Most of the time, the answer is something like “we’re using signups” or “we fire on form submission.” And when I ask why, the answer is usually “because that’s what we set up when we launched” or “that’s what was available”.

That’s the problem. The conversion event is not a setup task. It’s a strategic decision.

New strategies like Performance Max and Advantage+ have taken over the job of deciding who sees your ads. The algorithm decides targeting, placement, and bidding. Even creatives (imagery, videos and copy) are dynamically created by Google and Meta.

What you still control is what you tell it to optimize toward. That’s the event. Get it wrong, and you’re paying a very sophisticated system to find the wrong people.

In this article, I share 9 rules (tips?) I follow when structuring conversion events for paid campaign performance.

In this article, you will see:

• Why the conversion event is the most important decision in your campaign setup

• Why the event doesn’t have to be something the user does on your site

• How timing, value data, consent, PII, and click IDs all affect signal quality and, therefore, campaign performance

1. This is the most important decision you will make for your paid campaigns

Most marketers think the conversion event is a tracking task. Something for the data team to set up and for the media buyer to confirm is firing.

It’s not. It’s the single most important input into how your campaigns perform.

Platforms like Performance Max, Advantage+, and broad match with tROAS have removed most of the manual levers from campaign management. You no longer decide who sees your ads. The algorithm does. And the algorithm learns from one thing above everything else: the conversion event you give it.

Feed it a weak signal, and it will find users who match that weak signal. Feed it the wrong signal, and it will find the wrong users: efficiently and at scale.

This decision deserves the same strategic attention you give to budgets, channels, and creative. More, probably.

2. The event reflects what success actually means for your business

Not signups. Not form submissions. What does a genuinely good customer look like for your business?

For a nonprofit fundraising platform, a signup is meaningless. The valuable user is one that activates and starts raising money. For a B2B SaaS, it might be an account that invited a second user, or hit a usage milestone that historically predicts retention. For lead gen, it’s not a lead: it’s a qualified lead from the right company size.

Define that first. Then work backwards to find the right event.

3. The event doesn’t have to be something the user does on your site

This is where most teams leave performance on the table.

Ad platforms accept what’s called a synthetic event: a signal sent directly from your backend or CRM, after the initial online action. The event worth optimizing toward can be something that doesn’t exist in the client at all.

A user fills out a form, but the meaningful conversion is when sales marks them as qualified. A user signs up, but what you want to optimize toward is their first meaningful product action: something that happens in the backend, not in the pixel.

In all these cases, the signal worth sending is computed from data you already have and uploaded via API. Google calls this Offline Conversion Import. Meta receives it through the Conversions API.

4. Events that fire within 24 hours of the ad click have a higher chance of being attributed

Platforms attribute conversions within a lookback window and weight recent signals more heavily. Even though Google can attribute events up to 90 days after a click, a delayed event is a weaker signal. Attribution doesn’t break after 24 hours, but it degrades.

If your real conversion happens later, the solution isn’t to wait. Use a proxy event that fires within 24 hours and is a strong predictor of the downstream outcome you care about. That combination of fast-firing and predictive is better than the real event fired late.

5. Including value data (real or predicted) ensures ad platforms know what customers matter to your business

Without a value field in the payload, every conversion looks identical to the platform. A high-LTV subscriber and a low-intent trialist are indistinguishable. The algorithm optimizes for volume and brings cheap conversions, not valuable ones.

This is the practical difference between tCPA and tROAS. tCPA says: find me conversions at this cost. tROAS says: find me conversions at this return. You can’t run tROAS without value data.

An imperfect value signal is far better than none. You don’t need a model to start. Assigning value by plan tier, lead segment, or company size is enough. Start rough and refine from there.

6. Consent determines how much of the signal the platform can actually use

In markets with low consent rates, events arrive at the platform but can’t be matched back to campaigns for optimization. The event fires. The learning doesn’t happen.

This is not a reason to skip the event or assume tracking is broken. But it is a reason to prioritize server-side over client-side. Server-side events reach the platform regardless of browser consent state, which means more of your signal is actually usable.

7. Passing PII recovers signal that consent and cookies can’t

Hashed email and phone number let platforms match conversions to users deterministically, even when there’s no cookie. It’s the most reliable recovery mechanism available.

This is what Google calls Enhanced Conversions and Meta calls Advanced Matching. The implementation differs by platform, but the principle is the same: the more ways a platform has to identify a user, the more of your conversions it can attribute and learn from.

8. Capturing click IDs guarantees conversions are attributed back to the right campaign

When a user clicks an ad, the platform appends a click ID to the URL: gclid for Google, fbclid for Meta. If you capture and store that ID in your backend at the moment of the initial visit, you can tie any downstream event back to the exact campaign, ad set, and creative that drove it, even if that event fires later.

Without the click ID, the platform is guessing at attribution. With it, the connection is deterministic. User stitching (the process of linking a user’s backend record to their original click ID) can have a significant impact on attribution quality, especially for offline conversions and longer sales cycles.

9. Multiple sources can improve conversion attribution, but beware of duplication

Sending the same event from both client-side and server-side gives better coverage and higher match rates. But without deduplication, the platform counts the same conversion twice and the algorithm learns from inflated numbers.

Reported CAC looks great right up until someone fixes it, and then performance appears to crater. In reality, you were just measuring wrong the entire time.

Always pass a shared event ID across both sources and configure deduplication at the platform level: Meta’s event_id, Google’s transaction_id. QA it before launch, not after the numbers look suspiciously good.

Most teams optimise for the wrong thing and never realize it

The conversion event is not a tracking task. It’s the single input that determines what kind of customer the algorithm goes out and finds. Most teams set it at launch, treat it as infrastructure, and never revisit it. Even as their product, pricing, and customer profile evolve.

Getting it right means starting with a clear definition of what a good customer actually looks like for your business, then working backwards to an event that predicts that outcome, fires fast enough to be attributed, carries value data, and reaches the platform with as much signal intact as possible. That combination is what separates campaigns that improve over time from ones that plateau.

021 is a free newsletter on marketing measurement. If you want to read more about conversion tracking, attribution, and how to use data to improve paid performance, subscribe below.

Subscribe now

Multi-touch attribution (MTA) promises to solve the “who gets credit” problem by looking at the entire customer journey. In practice, it creates a different problem: every system builds those journeys differently, and the paths they construct systematically misrepresent which channels actually influenced the outcome.

The issue isn’t that MTA models are wrong. It’s that the touchpoint data feeding into them is incomplete and biased by design.

Search and direct traffic are over-represented because they’re easy to track. Upper-funnel channels like social and video are systematically excluded because impressions don’t generate clickable (or stitchable) trails through your analytics. Your Markov chain or Shapley model dutifully processes these incomplete paths and produces confident-looking attribution percentages that reflect tracking capability more than marketing effectiveness.

Triangulated attribution is a way to tackle this. Instead of relying on a single data source, you use multiple evidence streams (incrementality tests, self-reported attribution surveys, and platform performance data) to reach a more accurate view of channel impact. Rule-based methodology is how you operationalize it: a structured way to override MTA outputs or enrich journey data before attribution runs, using documented logic that turns experimental findings into repeatable adjustments.

This isn’t about replacing MTA with guesswork. It’s about making MTA reflect reality by correcting for the systematic biases in how touchpoints get recorded.

Where click-based attribution breaks down

Multi-touch attribution in tools like GA4, Snowplow, or custom MTA models only works if the touchpoints you’re attributing across actually represent the journey. But click-based paths have systematic gaps that make them fundamentally unreliable for measuring upper-funnel influence.

1. The impression problem

Someone sees your Meta ad on mobile during their morning commute. No click. They’re on the train. Three days later, they search your brand name on desktop at work and convert. Your analytics tool records a two-touchpoint journey: Brand Search → Direct. Your position-based model splits credit 40/40/20, giving most of it to search and direct. Meta, which created the awareness that led to that brand search, doesn’t appear in the path at all.

This isn’t a tracking failure. It’s an architectural mismatch. Impression-based channels like social, display, and video create awareness without requiring clicks. Intent-based channels like search and email capture existing demand with high click-through rates. When you feed only click-based touchpoints into MTA, you’re systematically overweighting demand capture and underweighting demand creation.

2. Cross-device (browser) journeys

Someone sees your YouTube ad on their phone, researches on their tablet two days later, and converts on desktop a week after that. Even with perfect cookie-based tracking, your analytics tool struggles to stitch this into one path. Most of these journeys collapse into “Direct” or “Organic” conversions in your reporting, which your MTA model then attributes to... nothing useful.

Cross-browser fragmentation creates the same problem even on a single device. A user clicks your Meta ad in the Facebook app (which opens in Meta’s in-app browser), browses your site, then returns hours later via Safari mobile to complete the purchase. Your analytics sees two separate sessions with different identifiers. The original Meta click never connects to the conversion. Your MTA model attributes the sale to Direct traffic, and Meta’s influence disappears entirely from your journey data.

This problem has gotten worse, not better. iOS privacy changes, cookie restrictions, in-app browsers, and the shift to mobile-first browsing mean more journeys fragment across both devices and browsers. Your MTA model doesn’t know these are fragments of the same journey, so it treats them as separate conversion events with artificially shortened paths. The channels that drove initial awareness, usually social and video, get systematically excluded from attribution.

3. View-through windows and platform logic

Meta counts conversions within a 7-click view and 1-day view window. Google Ads uses 30-day click and 1-day view by default. Your GA4 implementation might only count direct clicks with no view-through at all. These aren’t compatible measurement frameworks: they’re fundamentally different definitions of what “caused” a conversion.

When you export GA4 data into your MTA model, you’re processing journeys that already have Meta’s influence stripped out. The platform saw the impression, recorded the view-through conversion, and attributed it. But that impression never made it into your analytics event stream, so your MTA model can’t see it.

This Systematically Breaks Attribution

These aren’t random errors that average out, they’re systematic biases that consistently favor certain channels:

Your MTA model processes this biased data and produces attribution that makes search look brilliant and social look mediocre. It’s not measuring effectiveness. It’s measuring trackability.

This is why you need triangulated attribution: using multiple evidence sources to correct for systematic gaps in any single measurement system, then applying rules to operationalize those corrections.

How triangulated attribution works: two implementation approaches

Triangulated attribution means combining evidence from multiple sources (incrementality tests, surveys, platform data) to reach a more defensible view of channel impact. The rule-based methodology operates at two different layers depending on your data maturity and what you’re trying to achieve.

Channel-level override: adjusting MTA outputs

Override happens after your MTA model runs. You take the attributed conversions from GA4, Snowplow, or your MTA platform and reallocate credit between channels based on triangulated evidence that certain channels are systematically under- or over-credited.

How it works: Run your MTA model as normal on your click-based journey data. It attributes 800 conversions across your channels, with Direct getting 200, Brand Search getting 180, and Meta getting 45. You know from a geo holdout test that Meta drove an incremental 120 conversions during this period. Your override rule shifts 75 conversions from Direct and Brand Search to Meta, proportional to their original attribution, to reflect Meta’s true contribution based on experimental evidence.

How to implement: This lives in your BI or reporting layer: Looker, Tableau, or SQL views in your data warehouse. You’re creating an adjusted attribution table that applies percentage shifts to your MTA outputs. For example, during Sept 1-14, multiply Meta conversions by 1.67 (from 45 to 75) and reduce Direct/Brand proportionally to keep the total at 800. Your dashboard pulls from this adjusted table instead of raw MTA outputs. The original data stays intact; you’re just adding a transformation layer for decision-making.

The tradeoff: you’re limited to channel-level corrections. You can’t provide platform-level feedback for optimization, and you can’t attribute credit to specific creatives or audiences. But implementation is fast and doesn’t require modifying your attribution pipeline.

When override makes sense:

• Your user journey data is incomplete or poorly stitched

• You need to reconcile reporting quickly for budget planning

• You’re early in measurement maturity and need stakeholder alignment

• You have strong evidence from incrementality tests but limited technical resources

Most organizations should start with this type of override. It requires less infrastructure, produces results faster, and is easier to explain to stakeholders. Only move to insertion when you’ve validated the override approach, have the data quality and engineering resources, and need the additional granularity.

User-level insertion: enriching journey data

Insertion happens before your MTA model runs. You add synthetic touchpoints to individual user journeys where triangulated evidence shows that a channel influenced the conversion but wasn’t tracked in your click-based data.

How it works: You have 500 Direct conversions in your GA4 or Snowplow data. Survey data shows 30% of Direct converters recalled seeing your YouTube campaign in the past week. For 150 of those Direct conversion paths, you insert a synthetic “YouTube impression” touchpoint 3-5 days before the Direct visit. Now you re-run your MTA model—whether that’s position-based, Markov chain, or Shapley value—and it redistributes credit across these enriched journeys. YouTube now gets attributed credit, and Direct/Brand Search automatically decrease because the model sees a longer path.

How to implement: This requires modifying your attribution data pipeline before your MTA model runs. The key decision is selecting which user journeys get synthetic touchpoints. Your method should match your evidence quality:

• If your evidence is survey-based: Apply to survey respondents’ actual user IDs if you can match them, or randomly select 30% of Direct converters during the campaign period. Random selection assumes recall is evenly distributed.

• If your evidence is from incrementality test: Insert touchpoints only for users in the test geo or segment where you measured the lift. Keep insertion aligned with where you proved impact.

• If you have platform exposure data: Match user IDs between your analytics and platform impression logs. For users you know saw the ad, insert the touchpoint. Most accurate but requires data integration.

• If you have limited data: Apply probabilistically—add a 0.3-weighted synthetic touchpoint for all Direct converters, or use temporal logic (insert for conversions 1-7 days after campaign flight).

You’re modifying your user journey table (adding synthetic touchpoint records with channel, timestamp, and user_id), then feeding that enriched data into your existing MTA model. The model doesn’t know these are synthetic. It just processes richer paths.

When insertion makes sense:

• You have clean, stitched user journey data at the conversion level

• You need to feed adjusted conversions back to platforms for algorithmic optimization

• You want creative-level or audience-level attribution, not just channel-level

• You have the engineering resources to modify your attribution data pipeline

Building triangulated evidence: combining multiple sources

The power of triangulated attribution comes from not relying on any single measurement system. You combine incrementality tests, survey data, and platform analysis to validate adjustments from multiple angles. Each evidence source has limitations, but together they point toward a more accurate view.

Incrementality tests: the gold standard

Geo holdouts and controlled experiments give you causal evidence of incremental impact. When Meta’s geo test shows an 18% lift in conversions, you know Meta is driving conversions that your MTA model is missing or misattributing to other channels.

Translation to rules: If the test shows +120 incremental conversions and your MTA model only attributes 45 to Meta, you have a 75-conversion gap. Your rule might shift 70-75 conversions from Direct and Brand Search to Meta during the test period. The conservative approach is to start with 80-90% of the measured lift to account for measurement error and external factors.

Critical consideration: Apply temporal boundaries. If the test ran for two weeks in Q3, your rule should only affect those two weeks unless you have evidence the effect persists. Don’t extrapolate a short-term test result across an entire year without validation.

Self-reported attribution: directional evidence

Post-conversion surveys asking “How did you hear about us?” provide evidence of awareness and recall, but they suffer from recency bias, social desirability bias, and attribution errors by respondents themselves.

Translation to rules: If 22% of converters recall seeing your YouTube campaign, that doesn’t mean YouTube deserves 22% of credit—it means YouTube had measurable awareness impact. A conservative rule might allocate 8-12% of conversions to YouTube, discounting for recall bias. The specific discount factor depends on how your survey is designed and validated against other evidence sources.

This is where triangulation matters: survey says 22%, you discount to 10-12%, then validate against YouTube’s platform-reported view-through conversions to ensure you’re in the right range.

Platform discrepancies: reconciliation signals

When Meta claims 1,200 conversions and your GA4 shows 600, the gap itself is evidence, but you need to understand what’s causing it before turning it into a rule.

Translation to rules: Don’t use the 2x multiplier directly. Instead, investigate: Are the attribution windows different? Is Meta counting view-through conversions your analytics misses? Is there cross-device behavior? If you validate through incrementality testing that Meta drives 800 conversions (not 1,200), use that as your rule basis. The platform discrepancy tells you where to look, not what adjustment to make.

This is triangulation in action: platform says 1,200, analytics says 600, incrementality test says 800. You build your rule around 800 because it’s the most defensible number.

Triangulation example: YouTube campaign

You’re running a major YouTube campaign and need to understand its true impact. Instead of trusting any single source, you gather multiple evidence streams:

• GA4 attribution: 35 conversions attributed to YouTube (click-based only)

• YouTube platform reporting: 340 conversions (e.g. 7-day click, 3-day view window)

• Survey data: 28% of converters recall YouTube ad

• Brand search lift: +45% increase during campaign flight

• Geo holdout test: +18% incremental conversions in test markets

Your triangulated conclusion: Allocate 90-110 conversions to YouTube (conservative estimate from incrementality test), applied only during campaign flight dates. This reflects the 18% lift validated by the geo test, is directionally supported by the 28% recall rate and brand search lift, and falls well below YouTube’s claimed 340 (accounting for their aggressive attribution windows).

You’ve used five evidence sources to build confidence in one adjustment. No single source gave you the answer: the triangulation did. For sure, five sources is an overkill for most companies. So, start small: use MTA with a survey or an incremental test.

What actually changes: the competitive advantage

Triangulated attribution doesn’t reconcile numbers, it creates a measurable competitive advantage by making hard-to-track channels measurable and defensible.

Most companies systematically under-invest in social, video, and display because these channels look terrible in click-based MTA. Your competitors are making the same mistake. When you implement triangulated attribution using multiple evidence sources, you can see their true impact and invest accordingly while competitors keep cutting these channels based on incomplete data.

This is particularly powerful for channels with long consideration windows or heavy cross-device behavior. You’re able to defend budgets for YouTube campaigns that drive brand awareness, knowing the conversions will show up as “Direct” or “Brand Search” in raw attribution but being able to correct for that systematically using experimental evidence.

The competitive advantage isn’t just better measurement: it’s the willingness to invest in channels your competitors can’t measure properly.

“All models are wrong, but some are useful”

Triangulated attribution isn’t perfect measurement. It’s a pragmatic measurement that acknowledges every single data source has systematic biases, and corrects for them by combining multiple evidence streams into a more defensible view.

Your GA4 or Snowplow or custom MTA platform will continue to under-attribute impression-based channels and over-attribute last-click channels. Your ad platforms will continue to claim credit for everything within their attribution windows. Your surveys will continue to have recall bias. That’s fine. You’re not looking for a perfect source. You’re looking for convergent evidence from imperfect sources that collectively points toward a more accurate answer.

The goal is consistent attribution that’s good enough to make better decisions than you’re making today. Start simple: pick one channel where you have strong evidence of under-attribution: usually social or video. Run an incrementality test. Collect survey data. Compare to platform reporting. Build one override rule based on triangulated evidence. Validate it with stakeholders. Then expand to other channels as you build confidence and gather more evidence streams.

The alternative is continuing to optimize click-based MTA outputs that systematically misrepresent your marketing effectiveness or relying on hunch alone. Your competitors are making budget decisions based on incomplete data from single sources. You don’t have to.

We’ll cover how to turn post-purchase survey (“How did you hear about us?”) answers into measurable attribution inputs by calculating CAC, ROAS, and LTV from what customers say influenced their purchase. We also walk through how to handle non-responders properly so your numbers stay accurate, and how to integrate survey data into your broader attribution system without overcounting or misinterpreting the results.

A powerful simplification of the customer journey

Post-purchase surveys, also known as “How did you hear about us” (HDYHAU), often get dismissed because they do not feel rigorous enough. Analysts prefer hard data such as multi-touch attribution, MMMs, or incrementality testing. Surveys, by contrast, rely on human memory and subjective answers.

But that is precisely what makes them valuable.

When customers answer a post-purchase question, they simplify their journey into one clear story, the moment that felt like the reason they bought.

No pixel will ever see that.

It reveals the part of demand that tracking cannot capture and the part of the journey people actually remember.

What is a post-purchase survey?

Post-purchase surveys vary across industries, but this article focuses on the ecommerce version used for attribution. It captures what tracking cannot: the moment a customer believes they first discovered your brand.

Note: The same concept can work for other industries as well. For example, a SaaS business would typically trigger the survey right after a new user signs up.

• When it fires: After a customer completes their first purchase and the payment has been processed.

• Where it lives: On the order confirmation page or as a lightweight modal that loads post-transaction.

• What it looks like: A form with a single or multiple questions.

Why “HDYHAU” deserves a spot in your measurement mix

Adding a post-purchase survey does not replace MMM or modeled attribution. It fills gaps they cannot see.

1. It captures the untrackable

People discover your brand through podcasts, creators, word of mouth, or PR mentions. Channels that are typically difficult to measure. Surveys make those visible.

2. It adds context to quantitative data

Attribution models and analytics tools capture touchpoints, clicks, and last interactions. What they cannot capture is the moment that felt meaningful enough for a customer to remember.

The answers are subjective, and that’s the point. They reflect what buyers remember influencing them, which is often more useful than perfect precision. You are not recreating every touchpoint. You are identifying the moment that mattered.

3. It is simple to implement

Post-purchase surveys install in minutes.

4. It has low requirements

There is no need for years of historical data or a perfectly clean tracking setup. You can launch it today and start collecting useful signals immediately.

5. It acts as a good reality check

Survey does not replace MMM, or modeled attribution results; they serve as a sanity check against them. When numbers disagree, it helps you understand where perception diverges from data.

6. It is easy to understand

The results are intuitive. Anyone on your team can review the distribution of answers and see where customers say they came from.

7. It is easy to join with other data

Each response includes an identifier, such as order_id or email, making it easy to join survey data with customer and marketing spend data.

How to implement a first-order post-purchase survey for attribution

1. Choose the right tool

Look for one that integrates seamlessly into both your checkout flow and your data pipeline. KnoCommerce and Fairing are excellent options for Shopify, but any tool can work as long as it supports the core features you need:

• The ability to show the survey only to first-time customers

• Randomized answer choices to avoid biased selection

• A direct way to send responses to your warehouse, whether through an ETL, a webhook, or a simple automation tool

• Conditional follow-up questions, if you plan to collect more details

If you decide to add more questions later, partial-response capture becomes essential; otherwise, you risk losing the most important answers when someone abandons the survey partway through.

2. Craft and organize questions

Since the primary goal is attribution, we want the first question to focus directly on that. Start with a simple, familiar question that customers can answer without thinking:

“How did you hear about us?”

If you want to collect a second layer of detail, such as “Which creator?” or “Which show?”, make it an optional follow-up that appears only after the customer selects the relevant category.

Pro tip: To maintain strong response rates, small incentives can help if needed, though most brands already see solid participation without offering them. For first-order surveys, a response rate between 40 and 70 percent is common and generally reliable enough for attribution work.

3. Structure your choices

Answer choices should be clear, distinct, and easy to categorize later. The goal is to make it effortless for customers to choose the right option and for you to analyze the data accurately.

Best practices:

• Keep options mutually exclusive.
  Good: Social media, Search engine, Podcast, Friend or family
  Avoid: Social media, Instagram Ads, Facebook Ads (too much overlap)

• Use plain, recognizable terms.
  Good: Social media, YouTube, Podcast, Email, Search engine
  Avoid: Paid social, Video platform, Audio content, Newsletter, Organic discovery

• Align with your marketing spend categories as closely as possible, but prioritize clarity. When a trade-off arises, choose the option customers will understand instantly. Clean responses are more valuable than a perfectly matched mapping. More on that in the next section.

• Include “Other (open text)” at the end.
  Review “Other” responses regularly. If the same sources appear repeatedly, fold them into your main options. For higher-volume brands, an automated classifier can help maintain clean categories as responses scale.

• Keep it short.
  Limit to about 6–10 options to reduce friction and guesswork.

A clean, well-structured list ensures higher accuracy, fewer “Other” responses, and easier integration with your marketing spend data.

Pro tip: To test your list, ask people across your organization to act as customers from specific channels (Google Ads, Google Search, Facebook Ads, etc.) and answer “How did you hear about us?” If they hesitate or choose differently than expected, your choices need clarification or regrouping.

The inevitable mismatch: Customers do not distinguish between your technical marketing channels. Paid search and SEO often collapse into a single “Search engine” answer because most buyers cannot tell the difference.

This limits granularity, but you can account for it by grouping related spend categories on your side. For example, treat paid search and SEO as one unified search channel when running attribution.

The goal is not perfect precision, but a clean, consistent signal you can use for decision making.

From survey responses to actionable data

Up to this point, we’ve focused on collecting high-quality survey responses by choosing the right tool, crafting clear questions, and structuring clean answer choices. Now it’s time to turn those responses into something useful.

You will need some basic SQL knowledge for this part, since the next steps involve joining tables, mapping channels, and running simple aggregations in your warehouse.

This is where everything comes together, and your survey data becomes real insights you can use to make smarter decisions about where to invest.

Let’s build cool stuff.

1. Send responses to your data warehouse

Once your survey is live, set up an automated flow to send responses to your warehouse. You can do this via APIs, webhooks, or ETL tools such as Fivetran or Portable. For lighter setups, Zapier or Make can also handle the transfer.

2. Join with your customer data

Once the data lands, join the survey table to your customer table using email or customer_id. We join the data on a customer level because the survey only shows for the first purchase.

3. Aggregate the results

Aggregate your survey responses by channel to see how often each source is selected, and prepare the data for downstream calculations such as CAC and ROAS.

4. Map answers to marketing spend categories

Your goal is to align the heard_about_us field from your survey responses with the channel names from your marketing spend table. You can use a simple mapping table to connect survey answers to spend buckets.

5. Join with marketing spend data

Once your post-purchase survey data is connected to customers, the next step is to join it with your monthly marketing spend by channel.

Pro tip: If you spend a considerable amount on retargeting campaigns on specific channels, exclude that spend.

Accounting for non-respondents

Not every buyer will complete the survey, and that matters when you calculate metrics like CAC by channel. If you ignore the response rate, your CAC will be drastically inflated.

Connecting response rates to results

When you calculate CAC the usual way, you divide total spend by total new customers:

cac = Monthly marketing spend by channel / Monthly new customers by channel

*All the CACs from this table are wrong because they assume every customer completed the survey, which is rarely true.

How to adjust for response rate

To correct for this, scale your spend by the survey response rate for that specific cohort.

response rate = Monthly number of responses / Monthly new customers

adjusted_cac = (Monthly marketing spend by channel * Monthly survey response rate) / Monthly new customers by channel

This way, you are comparing data from the same cohort of customers and spend, rather than using a single global rate that may shift over time.

Achieving a normalized CAC

• It normalizes CAC across channels and time periods with different response rates.

• It highlights where data is thinner so you can interpret results with the right context.

• It keeps your post-purchase survey analysis aligned with real-world participation.

Imagine the following:

• You spent $110,000 on TV ads in November

• 324 new customers answered the post-purchase survey with “TV” in November.

• 5,589 new customers answered the post-purchase survey out of a total of 10,574 new customers in November. This means your response rate for November was 52.86% (5,589/10,574).

Here’s how your adjusted CACs would look:

Pro tip: To be extra rigorous, compare the group of customers who answered the post-purchase survey with those who did not.

Look for differences in AOV, repeat purchase rate, refund rate, or the source/medium reported by Google Analytics 4 for that transaction.

If the two groups behave similarly, you can trust your extrapolated results. If they differ meaningfully, consider segmenting your analysis or adjusting how you interpret the survey data.

Pro tip 2: To calculate the CAC for non-responders, you can use the monthly weighted average of your adjusted CACs.

Making decisions based on what people remember

Attribution will always be imperfect. You will never be able to isolate the exact channel or message that caused a purchase.

The goal is not to find a perfect truth, but to collect enough imperfect signals to make confident decisions.

MMM, modeled attribution, platform data, and intuition each tell part of the story. Post-purchase surveys tell the part that can’t be modeled: what people actually remember.

And when you connect those answers to your data warehouse, you create a scalable, quantifiable, and human signal that grounds your attribution system in reality.

MMM alone isn’t enough anymore

Marketing Mix Models (MMMs) are making a comeback.

As privacy restrictions tighten and cookies crumble, marketers are returning to MMMs as one of the few durable, privacy-safe ways to measure performance. But this resurgence is happening in a very different landscape: one where data volumes have exploded, leadership expectations from models have increased, model complexity has increased, and decision timelines have collapsed.

This is where Large Language Models (LLMs) come in. AI-assisted tools (or “AI Co-pilots”) are emerging as powerful accelerators inside marketing science workflows.

They can:

• Automate data extraction and summarization

• Translate complex model results into plain language

• Identify anomalies or shifts across time

• Draft first-pass performance memos and recommendations

With the right setup, these tools can shrink a multi-week storytelling process down to days or even hours. But this speed introduces a new problem: reliability.

Why now?

MMMs remain one of the go to tools for long-term performance measurement in marketing since the 50s, but they come with a major bottleneck: human interpretation.

A typical MMM produces dozens of outputs: incrementality, contribution numbers, Cost Per Actions, marginal ROIs, diminishing returns (response curves), confidence intervals, stability loops. All these at levels of granularity such as date/week/month, line of business, product, platform, region etc. Turning those numbers into something actionable takes both technical expertise and a storytelling skill. And, Data Scientists have been masters of that craft for a while! However, it takes a lot of time.

Even in well-staffed analytics teams, the process looks like this:

That can easily take 1–3 weeks per cycle.

In a fast-moving marketing environment, that delay means insights often arrive after the budget decisions have already been made.

Worse, MMMs can be opaque to non-technical teams. Marketers and finance leaders see tables of coefficients but lack context: Which channel truly moved the needle? How reliable is this estimate? How does this change if we shift spend mid-quarter?

Without faster and more interpretable translation, MMMs risk becoming beautiful models that never influence real decisions.

The role of LLM co-pilots in MMM workflows

The Promise of AI Assistance: The Rise of AI Co-pilots in Marketing Measurement

Marketing Mix Models (MMMs) have traditionally been data-intensive, slow to iterate, and hard to communicate. But with the rise of Large Language Models (LLMs), nearly every step of the MMM workflow is seeing automation or augmentation. LLMs fit in different parts of the MMM lifecycle:

1. Coding Assistance: LLMs can accelerate model development by helping Data Scientists write, debug, and refactor code for econometric and Bayesian MMM frameworks.

2. Missing Data Imputation: LLMs can suggest or generate plausible values using contextual and temporal patterns.

3. Data Summarization: Automatically describing correlation structures, feature interactions, and trends.

4. Agent-based Feature Engineering: LLM “agents” can propose, test, and evaluate new features or priors.

5. Drift and Stability Detection: Monitoring model consistency across time windows or geographies.

6. Post-Processing and Insights Generation: Transforming raw model outputs into coherent business stories.

LLMs as Co-Pilots for Decision Makers

A single MMM can produce hundreds of coefficients, elasticities, and saturation curves. For most executives, this output is overwhelming. Traditionally, a Data Scientist spends days (often weeks) translating these outputs into narratives, recommendations, and visuals that business leaders can act on.

LLM-based insight co-pilots can dramatically compress this timeline. They can summarize results, highlight ROI inflection points, compare trade-offs across channels, and generate decision-ready insights in plain language. For non-technical audiences, these co-pilots become interpreters of complexity - helping teams understand why certain investments outperform others and where budget reallocations could yield the most lift.

The LLM co-pilot can generate key insights such as:

1. Cost Per Action (CPA), where applicable

2. Revenue breakdowns by channel, platform, region, and segment over time

3. Cost metrics: CPC (Cost Per Click) and CPM (Cost Per Thousand Impressions)

4. Diminishing returns curves reflecting non-linear media effectiveness

Important: Consistency in field names, schema, and metadata is critical for context-aware reasoning. The co-pilot must not misinterpret fields or mix metrics.

“LLMs can’t reason their way out of bad data.” Ensuring high-quality, validated inputs is essential for reliable outputs.

How LLM for MMM Looks in Practice

Leading organizations are now deploying LLM-powered apps that let marketers and analysts query their Marketing Mix Models (MMMs) in plain language - no code or manual digging required.

Once new MMMs are trained, MLops pipelines automatically push model outputs (ROI curves, elasticities, forecasts) into a secure, centralized database. The LLM co-pilot then retrieves the latest version, grounded in brand-specific context and governed by internal data policies.

Stakeholders can ask:

“How did paid social efficiency trend last quarter?”

and receive instant, explainable answers - complete with charts, uncertainty ranges, and source tags.

The result: insights are fresh, contextual, and conversational, reducing decision time cycles.

Data requirements for trustworthy co-pilots

To enable LLM-based co-pilots to provide actionable insights, MMM data must be structured, granular, and consistent. Specifically, the model requires:

• Channel × Platform × Time granularity: Typically monthly or weekly breakdowns.

• Taxonomy information: If available, provide segment, product, or audience hierarchies.

• Performance metrics: Spend, impressions, CPM/CPC changes, and predicted ROI or contribution at the same granularity.

• Non-linear response curves: Include equations if available to capture diminishing returns or saturation effects. If not, the model can generate them by itself.

In practice, stakeholders need to provide data in a consistent schema, so the co-pilot can reason accurately. Many vendors, especially off-the-shelf models, do not provide this automatically. In most cases, in-house data teams can setup an automated connection with GenAI based applications for the model to tap into the latest model outputs. If this is not possible, the data can be loaded to the application as an excel file using which the model can generate insights and provide optimization recommendations.

Below is an example table illustrating the input format required for an LLM-based co-pilot:

Designing the co-pilot: prompt strategy & fine-tuning

Building a reliable LLM co-pilot for Marketing Mix Modeling requires not just structured data, but also explicitly engineered context so the model can reason accurately across multiple levels of granularity. This section describes how prompts, fine-tuning, semantic layers, and context engineering work together.

Prompt Engineering & Scaffolding

Prompts define the reasoning pathway of the model. Without careful design, LLMs can misinterpret or hallucinate insights. This can result in model producing incorrect recommendations and findings, which can ultimately result in bad business decisions.

• Chain-of-Thought Reasoning: Break analysis into sequential steps.

  Example: Instead of “Which channel is most efficient?”:

  1. Compute CPA, CPC, and CPM for each channel.

• Guardrails in Prompts: Explicit instructions to prevent errors:

  • “Always reference the source dataset and time period.”

• Few-shot Prompting: Provide examples of correct MMM outputs to teach the model expected reasoning patterns.

• Validation Queries: Embed self-checks in prompts:

  • “Does CPA equal Spend ÷ Conversions?”

Fine-Tuning on Brand-Specific Knowledge

Fine-tuning enables the model to understand brand-specific nuances, business rules, and historical performance patterns:

• Historical MMM reports and campaign performance summaries. Load recent MMM results.

• Business rules, e.g., minimum spend thresholds, segment-level filters, seasonality adjustments.

• Embedding structured data (tables, JSON, CSV) for accurate metric referencing.

Fine-tuning ensures the LLM learns domain-specific language and patterns, reducing errors and improving consistency.

Semantic Layer: Structured Knowledge for the LLM

A semantic layer provides the model with a formal representation of the data and relationships, which is critical for consistent reasoning:

• Entity definitions: What each field represents, e.g., Channel, Platform, Segment, Contribution.

• Hierarchies & Aggregations: Define roll-ups (e.g., segment → region → country) for correct summations.

• Valid Ranges & Constraints: Prevent impossible values (negative spend, CPC < 0).

• Mapping to KPIs: Link data fields to ROAS, CPA, revenue contribution, etc.

• Automatic Field Resolution: Prevents misinterpretation of similar metric names (e.g., “Spend” vs. “Budget”).

Semantic layers act as a knowledge graph for the co-pilot, ensuring data is interpreted consistently.

Context Engineering: Embedding Situational Awareness

Context engineering provides temporal, business, and domain context so the co-pilot can reason appropriately:

• Temporal Context: Include campaign dates, seasonal periods, and event flags to distinguish baseline vs. incrementality effects.

• Business Context: Define business objectives (ROAS, CPA targets, revenue goals) so the LLM aligns insights with decision-making priorities.

• Data Provenance: Track source systems, calculation methods, and assumptions for transparency and traceability.

• Scenario Context: Include multiple datasets (e.g., historical MMM, competitor benchmarks, media mix shifts) to provide situational grounding.

By embedding context explicitly, the co-pilot avoids common pitfalls like misreading coefficients, ignoring seasonal effects, or misattributing campaign impact.

The combination of structured prompts, semantic layers, context engineering, fine-tuning, and modular reasoning ensures the LLM co-pilot generates reliable, actionable insights while minimizing errors and hallucinations.

Reliability & Validation: Guardrails, Testing, and Continuous Improvement

LLM co-pilots accelerate MMM insights but carry risks of misinterpretation, spurious correlations, or overconfident outputs. This unified framework ensures outputs are trustworthy, auditable, and aligned with business objectives, enabling safe and scalable deployment.

Guardrails for Safe and Accountable Outputs

Purpose: Control, constrain, and guide the co-pilot to prevent errors and misalignment.

Below, you can visualise the different types of guardrails that can be implemented for distinct purposes.

Highlights:

• Semantic layers and context engineering ensure correct metric interpretation.

• HITL and DACI governance assign ownership and review responsibilities, ensuring auditability.

• RAG grounds outputs in factual data, reducing hallucinations.

Testing and Validation

Purpose: Verify correctness, consistency, and alignment with business goals.

• User Acceptance Testing (UAT): Compare LLM outputs against expert analyses. Validate KPIs such as CPA, ROAS, CPC, CPM, and contribution.

• Scenario Testing: Evaluate co-pilot reasoning on edge cases (e.g., high spend campaigns, seasonal variations).

• Error Logging: Maintain a repository of hallucinations, misinterpretations, and KPI deviations. Analyze patterns for continuous improvement.

Monitoring and Metrics

Key indicators to track co-pilot performance:

• Accuracy: % of outputs aligned with known MMM results.

• Hallucination rate: Frequency of unsupported conclusions.

• Coverage: Proportion of channels, segments, and time periods included.

• Confidence alignment: Correlation between reported uncertainty and actual error.

• Throughput: Time/resource efficiency for generating validated outputs.

Dashboards can visualize performance trends, confidence ranges, error patterns, and human review checkpoints to support scalable operations.

LLMs Add Speed and Intensity to MMMs, but Guardrails can’t be Skipped

LLM co-pilots can democratize MMM insights - but only if bounded by transparent, testable rules.

When properly designed, they shift marketing analytics from a specialized function into an organization-wide capability: everyone from a CMO to a campaign manager can query performance, simulate ROI, and see what drives growth. Yet democratization without discipline risks confusion. The future of analytics will favor teams that combine accessibility with accountability, co-pilots that explain why they recommend something, not just what to do.

The future of marketing analytics isn’t “AI instead of analysts,” it’s AI guided by analysts.

Human judgment remains the compass. Data scientists define the metrics, guardrails, and validation logic that keep AI grounded. Analysts interpret nuance that models can’t see - brand shifts, creative fatigue, or cultural moments. The best co-pilots don’t replace human intuition; they amplify it, compressing analysis time while expanding clarity.

“AI that earns trust beats AI that just sounds smart.”

As LLMs move deeper into decision workflows, credibility will matter more than cleverness. Trusted co-pilots will be those that cite sources, quantify uncertainty, and learn from feedback loops. They will act less like automated narrators and more like transparent collaborators, auditable, interpretable, and continuously improving.

The next era of marketing measurement won’t be defined by who uses AI first, but by who uses it responsibly.

Building trustworthy co-pilots means merging the best of both worlds:

• The rigor of data science

• The discipline of governance

• And the human judgment that turns data into direction

If you’ve enjoyed this article, connect to Aditya on Linkedin and subscribe to 021 below:

Subscribe now

It’s what I recommend to clients starting with server-side or dealing with Meta event duplication issues (which I’m afraid happens too often). You can start with a free account and test a single event before committing.

The decision between client and server side for conversion events

The client-side versus server-side decision isn’t about choosing better technology. It’s about understanding where your conversions happen and which tracking method will actually capture them.

If your most valuable conversions happen in the browser during the session, client-side works fine. But if they happen later in your backend (payment confirmed, lead qualified, subscription activated from expired trial), the browser can’t see them.

In this article, we’ll cover how to make the decision of where to fire your events from.

Attribution and optimization depend on seeing conversions

For paid campaigns, conversion tracking serves two purposes.

First, you need conversions attributed back to the campaigns that drove them. When someone clicks your ad and converts, the platform needs to know that happened. This is how it learns which audiences and targeting work. More attributed conversions means more data for the algorithm.

If conversions happen but don’t get attributed (cookie expired, click ID lost, browser blocked the script), the platform never sees them. You might have 100 actual conversions, but if only 60 get attributed, the platform only has 60 data points.

Second, you want to send the version of the conversion that matters to your business. A platform optimizing for “form submitted” will behave differently than one optimizing for “qualified lead accepted by sales.” The more the event reflects real value, the better the optimization.

The difference is where and when events fire

Client-side captures what happens in the session

With client-side tracking, the event fires directly from the user’s browser (or, from the app) when the action happens. Someone completes a purchase, the browser fires the event right then.

Usually, this automatically respects consent-mode restrictions, which are easy to implement with client-side GTM. You get an accurate picture of what happened in that session, captured in real time.

The problem: client-side is vulnerable to anything that interferes with the client. Ad blockers, privacy settings, connection issues, or the user closing the page before the script finishes. When any of these happen, the event never reaches your ad platform.

Client-side reflects exactly what happens in the session, at the moment it happens.

Server-side gives you control over timing and content

With server-side, the event is sent from your server directly to the ad platform. The data can reach your backend from the client, or from another source (your CRM, payment processor, etc.).

Because it fires from your server, ad blockers and browser privacy settings don’t affect it. More importantly, you can delay when the event fires. You can wait for payment confirmation instead of firing on checkout submission. You can calculate a lead score before marking someone as qualified. You can replace preliminary revenue with fraud-checked amounts.

The tradeoff: consent logic needs careful implementation. A browser automatically usually blocks itself when users don’t consent. Your server doesn’t. You need to explicitly handle consent signals.

Server-side lets you decide when to fire the event and what version to send.

Server-side attribution works better with PII

Client-side relies on cookies and click IDs

Client-side attribution uses click IDs from the browser (gclid, fbclid), cookies that identify the user, and optional hashed PII (though many teams avoid this for security reasons).

This works well when the conversion happens in the same session, the browser keeps the click ID, scripts aren’t blocked, and cookies are accepted.

It breaks down when the conversion happens later after cookies expire, the browser removes click IDs (Brave does this), there’s a cross-browser/device journey, or user settings block tracking.

Server-side can safely send PII for better matching

Server-side can also pass click IDs (and should), but the bigger advantage is PII.

Sending hashed email addresses and phone numbers from the server is safer than exposing them client-side. Platforms like Meta and Google use this to match conversions even when cookies and click IDs fail.

And because you can delay the event, you can enrich data with tools like Clearbit or Clay before sending. This generally improves attribution, especially for conversions outside the initial session.

Three ways to implement server-side tracking

Client-side implementation is straightforward. Most teams use Google Tag Manager or direct pixels. Simple to deploy, easy to debug, minimal engineering involvement.

Server-side has three main approaches with different tradeoffs.

Server-side GTM mimics client-side behavior

Server-side GTM works like the web container most marketing teams already use. The interface looks familiar, and marketing teams can often manage it after initial setup.

You get built-in debugging tools, easier deduplication between client and server events (both use the same GTM structure), and marketing can self-serve most changes.

The downsides: you need infrastructure to run the server or a hosted provider. And it’s less flexible than a completely custom pipeline.

This is the main use case for Stape, as they provide the infrastructure here. They handle the server setup, generate event_ids automatically for deduplication, and manage connections to platforms like Meta. You get server-side GTM without maintaining your own server.

CDPs centralize data but aren’t optimized for paid

Tools like Segment, RudderStack, and mParticle collect events once and send them to multiple destinations.

If you already have a CDP in your stack, the benefits are real. Event collection gets centralized, routing to multiple destinations is easy, and marketing can usually configure destinations without engineering.

But CDPs aren’t designed for paid media optimization. Debugging tools vary between platforms and can be difficult. Most don’t send events in real time, which hurts attribution and targeting. And deduplicating client/server events through a CDP is often complicated.

If you’re already paying for a CDP and using it extensively, it might make sense. However, deduplicating client-side and CDP-powered server events is usually quite difficult.

Custom pipelines give full control but need engineering

Some teams build custom backend pipelines that fire client side events, store them on a server, and then send to ad platforms.

The advantage is complete flexibility over timing, logic, and event structure (like, everything). The disadvantages add up: no built-in logging or debugging tools, every change requires engineering, easy to break attribution if not aligned with client-side logic, and ongoing maintenance becomes a permanent engineering responsibility. Most marketing teams do not love this.

This works for teams with strong engineering resources and very specific needs. For most marketing teams, it creates more problems than it solves.

What works for different business models

The right approach depends on where your conversions happen and when their value gets determined. Here’s how this plays out across common business models.

E-commerce usually works fine with client-side

For most e-commerce businesses, client-side tracking captures what matters. It works well when:

• Purchase happens quickly after landing

• Transaction occurs in the same browser they clicked from (works better for Google Ads than Meta’s in-app browser)

• Click IDs are reliably present

• Revenue value is accurate at transaction time

If your typical journey is: click ad, browse, buy in the same session, client-side captures it.

Server-side makes sense when you need to adjust revenue after returns or fraud checks, want to optimize for predicted customer lifetime value, or when significant purchases happen across devices.

If you want to set up CAPI for ecommerce, there are many many tools for the main platforms (like Shopify) that can make this straightforward.

Subscription products need server-side for meaningful conversions

For subscription businesses, the conversions that matter happen after the browser session ends.

Someone might sign up during the session, but the value gets determined when payment processes through Stripe, when they complete onboarding and activate, when they renew after the first cycle, when revenue clears fraud checks, or when they upgrade their plan.

The browser can’t see these. They happen in your backend, and sometimes days or weeks later. If you’re only tracking signup client-side, you’re optimizing for signups, not for paying, activated, retained customers.

Server-side lets you:

• Send the event when payment is confirmed, not just when someone submits their card

• Optimize for predicted lifetime value (ROAS) instead of signup count (CPA)

This fundamentally changes what the ad platform optimizes toward.

Lead generation works better when you wait for qualification

The browser only knows: this person submitted the form.

Your business cares about different questions. Is the lead qualified? Does it fit ICP? Did sales accept it? Did it convert to a customer?

Important point: enrichment tools like Clearbit don’t get sent directly to the ad platform. They help you decide which leads are valuable enough to send as conversions.

With server-side, you can:

• Wait to see if the lead passes qualification before sending anything

• Promote raw leads to qualified leads only after validation

• Avoid polluting optimization with low-quality submissions

• Send conversions based on CRM outcomes (sales accepted, meeting booked) instead of browser form submissions

• Include predicted lead score or LTV in the event data

The platform sees “high-value lead that fits ICP” instead of just “lead submitted.” You optimize for business value, not form volume.

Some conversions only exist in your backend

Some conversions can’t be captured client-side because they don’t happen in the browser.

Payment confirmed after third-party processing. Revenue adjusted after fraud checks. Sales-qualified status triggered in CRM. Trial automatically converted to paid after 14 days.

The browser closed hours or days ago. For these, server-side is the only option.

Deciding what you actually need

The decision comes down to two questions: where do your conversions happen, and when does their value get determined?

Client-side makes sense when conversion happens on-page during the session, click IDs are reliable, you don’t need enrichment or delay, simplicity matters, and you have limited engineering resources.

Server-side makes sense when important conversions happen outside the browser, you need to send a better version of the event (qualified, enriched, confirmed), backend systems define real conversions, and cross-device or delayed conversions are common.

You might need both when you want immediate on-page events for short-term optimization plus backend-confirmed events for long-term value. This is common in subscription and lead-gen models.

Start small and test

Don’t move everything at once. Start with your highest-value conversion that happens outside the browser. Test whether it improves optimization and campaign performance.

The biggest mistake teams make is treating this as all-or-nothing. You can run hybrid. You can start with one conversion type and expand only if it works. It’s also ok to run come campaigns optimizing towards client events, some server events and some both. As always, the answer for marketing is: it depends!

Take the first step with Stape

If you want to test server-side without building infrastructure, Stape makes it straightforward.

Start with a single event (qualified lead, subscription activation, whatever matters most) and see if it improves your campaigns. Their infrastructure handles the technical setup, deduplication, and platform connections.

Once you’re storing the client-side data on a server, you can start to enrich it with events that never made the client or predicted values. Then it starts to get really powerful.

Pricing starts at $17/month. Create a free account here to test before committing.

Maybe this is the year you finally test out-of-home.

Or build a real ABM program.

Or scale Meta more aggressively.

Or give Google’s AI Max a try.

Or increase your experimentation velocity across creative, channels and audiences.

All totally reasonable things to want. But here’s the thing I see over and over:

The marketing plan gets written long before anyone maps what needs to happen in the data layer to support it.

That’s usually when problems appear. Not because the initiative was wrong, but because the tracking or measurement foundation wasn’t ready for it.

A few examples I’ve seen:

• Testing OOH without a way to measure lift or model the impact

• Scaling Meta when conversion signals aren’t attributing enough to support optimization

• Running AI-led Google campaigns (either PMax or AI Max) without signal engineering/conversion values

• Trying to run more experiments when the reporting layer can’t judge them properly (or fast enough)

• Starting ABM with no scoring, routing or funnel measurement structure

So this year, I’m doing something new. Bare in mind, this is a sales initiative for me.

I’m opening a small set of free 30-minute planning calls to help marketers connect both sides of their plan:

The ambition + the data required to make it work.

What we can cover in the call

• Your 2026 initiatives (big bets, small bets, channel changes, experiments)

• The data, tracking and measurement requirements behind each idea

• What needs to happen first so initiatives perform as planned AND you’re able to measure their impact

• Where you might need specialist help vs what your team can handle internally

  Let's talk 2026 marketing plans

The kinds of projects I support

If you’re newer to my work, here’s where I typically help teams:

• FixMyTracking: reliable conversion signals for Meta, Google, TikTok and attribution

• Attribution and measurement strategy beyond “why don’t GA4 and Meta match”

• Creative and channel reporting systems that actually guide decisions

• Data foundations for scaling: event design, server-side tracking, CDPs, analytics architecture

• Experimentation frameworks: uplift tests, incrementality, structured creative testing

Who I work with

I don’t deliver everything alone. I have my set of trusted coworkers that include:

• Marketing scientists for lead scoring, predictive models, synthetic conversions

• Marketing analysts for debugging, QA, analytics troubleshooting + supporting much of the work I do

• Data engineers for server-side tracking, pipelines, BigQuery, CDPs and integrations

If you’re shaping your 2026 marketing plan and want help making sure the data side isn’t a bottleneck, grab a slot here:

Let's talk 2026 marketing plans

Bring the wishlist. The bigger the better.

First, let’s go back to 2015

In 2015, I was leading growth at HER, a dating app where we scaled from 2,000 to 1 million users. Every morning started the same way: open the campaigns, check the CPIs, adjust bids on the different ad groups segment per audience, pause the underperforming ads, boost budget on the campaigns that were crushing.

We had persona decks that could wallpaper an office. We knew exactly which Facebook pages our users liked, what time of day converted best, which creatives would have the highest CTR. We split-tested everything, manually tweaked bids at least twice daily, and felt genuinely in control of our campaigns.

That was 2015. Platforms have changed completely since then, but most marketing teams haven’t.

We’ve moved from controlling campaigns to feeding algorithms. Yet we’re still setting them up like it’s 2015.

How paid media has evolved

The transformation happened gradually, then suddenly. This was already the trend “pre-AI”, but now we’re in fast mode.

What we controlled in 2015:

• Built audiences from scratch using granular targeting

• Set manual bids down to the penny

• Chose exactly which placements our ads would appear on

• Made every targeting decision ourselves

• Created detailed persona decks

What platforms control in 2025:

• Performance Max decides where your Google ads appear across Search, Display, YouTube, Gmail, and Discovery

• Meta Advantage+ combines your creative assets and serves them to whoever the algorithm thinks will convert. They go as far as creating assets now.

• Broad match with tROAS lets the platform find customers you never would have targeted manually

• Auto-bidding optimizes spend in real-time across millions of signals

The algorithm is now the media buyer. Our job is to teach it what good looks like.

This isn’t a minor shift in tactics. It’s a fundamental restructuring of how paid media works. Platforms don’t want us micromanaging campaigns anymore because their algorithms genuinely perform better when given consolidated data and clear objectives.

But here’s the problem: most teams adapted their campaign types without adapting their approach. They’re running Performance Max campaigns but still thinking like it’s 2015.

There are habits we haven’t yet outgrown

Walk into most marketing teams today and you’ll see the same patterns from a decade ago, just wrapped in newer interfaces.

Over-segmentation remains the biggest culprit

Teams split campaigns into eight ad sets because they want to “test different audiences,” when the platform already has access to billions of signals that make those manual segments obsolete. They’re limiting the algorithm’s ability to learn by spreading conversions too thin across too many containers.

Another significant consequence of this is budget allocation. Instead of letting the algorithm optimize for ROAS and shift budget across different campaigns based on performance, marketers are manually defining budget for each audience. This leads to inefficiency.

Siloed testing produces misleading results

Teams test a new creative in a separate campaign, see it perform worse than established ads (because it’s in the learning phase with no historical data), and conclude the creative is bad. They’re drawing conclusions from a fundamentally flawed test setup. The creative might be brilliant, but they’ve handicapped it by isolating it from the campaign’s accumulated learning.

Nowadays, creative is the targeting. There are better methodologies for understanding creative performance.

Attribution obsession consumes countless hours

Teams try to make their Facebook Ads Manager numbers match GA4, then panic when they don’t align. They adjust attribution windows, create custom reports, schedule meetings to discuss the discrepancy.

Meanwhile, they’re missing the fundamental point: your ad platform doesn’t care about GA4. It optimizes based on the conversion signals you send it, not what appears in your analytics dashboard. In fact, I’d go as far as saying: if your Meta and GA4 conversions are matching, then you’re not doing Meta right.

Uniform targets ignore customer value

Teams set a $50 CAC target across all campaigns, treating every customer as equally valuable. But a $500-per-year subscriber and a $5 trial user are not the same. Without value data, the algorithm can’t tell them apart and will happily spend your budget acquiring the cheapest conversions, not the most valuable ones.

Generally, I feel like this is a problem everyone is aware of. But they try to fix it with reporting: by measuring which campaigns/strategies lead to the highest ROAS. The best way to approach this is, however, by feeding this data into the ad platform.

We say we trust the algorithm, but we don’t trust it to learn.

What actually matters now: your conversion signals

Here’s what most marketers get wrong about conversion tracking: they think if their tag fires, they’re good. They check the Facebook Pixel Helper, see the green checkmark, and move on.

But a working tag doesn’t mean you have useful signals. The algorithm doesn’t see your dashboard. It sees your payload.

The four inputs the algorithm actually needs

Platforms need four things to optimize effectively, and most campaigns are missing at least two of them.

1. Sufficient Frequency

Meta’s learning phase requires about 50 conversions per week per ad set. Google’s algorithms need similar volumes. Fire 10 conversions per week and your campaign never exits learning phase, it just perpetually resets, unable to establish patterns or optimize meaningfully.

This is why consolidating campaigns almost always improves performance. You’re not giving the platform less control; you’re giving it more data to learn from.

2. Signal Consistency

If you’re sending the same conversion from both client-side pixels and server-side events without proper deduplication, the platform sees twice as many conversions as actually occurred. Your reported CAC looks fantastic, but you’re teaching the algorithm to optimize toward inflated numbers.

When you eventually fix the duplication, performance appears to crater, but really, you were just measuring wrong the entire time.

3. Attribution Compatibility

Modern attribution requires multiple data sources:

• Client-side tracking catches some conversions

• Server-side catches others

• Proper PII (like email addresses) helps platforms match conversions back to ad clicks even when cookies fail

Teams running only client-side tracking could miss 80%+ of their conversions in a post-iOS14 world. The platform thinks those campaigns perform worse than they do and shifts budget away from them.

4. Value Data

This is the most commonly missing piece. A $500 annual subscriber and a $5 monthly trialist should not look identical to Meta’s algorithm, but without value and currency fields in your conversion payload, they do.

The platform will optimize for volume rather than value, delivering you a steady stream of low-quality, cheap conversions while the valuable customers remain out of reach.

Barbara’s Five Rules for Optimization events

Through eight years of fixing tracking for companies like Microsoft, WeTransfer, and dozens of startups, I’ve developed five rules that consistently separate high-performing campaigns from struggling ones.

The business impact of better signals

These technical details translate directly to business outcomes I’ve seen repeatedly across client engagements:

Performance marketers become strategists again. When your conversion signals work properly, the nature of the job changes. You stop spending hours tweaking bids, debating audience exclusions, or reconciling attribution reports. That time gets redirected toward work that actually differentiates your brand: developing positioning that resonates, creating thumb-stopping creative, testing new value propositions. The algorithm handles mechanical optimization across millions of micro-moments. You focus on strategic decisions it can’t make: what story to tell, which pain points to address, how to position against competitors.

Faster learning phases mean campaigns become profitable sooner. Instead of spending three weeks stuck in learning while burning budget, campaigns with strong signals exit in days. They start optimizing toward real conversions faster, which means less wasted spend during the ramp-up period.

Reduced troubleshooting time might sound minor, but I’ve seen teams spend 15 hours per week debugging why campaigns aren’t performing, only to discover their conversion events were firing inconsistently or missing value data. Fix the signals, and those hours get redirected toward creative strategy and landing page optimization, work that actually moves the needle.

Smarter budget allocation happens automatically when platforms have good data. Instead of spreading budget evenly across campaigns because you don’t trust the platform’s optimization, you can consolidate into fewer campaigns and let the algorithm shift spend toward whatever’s working. I’ve seen this change alone improve efficiency by 30-40% for clients who were previously over-segmented.

Most teams think they need better targeting. They actually need better signals.

Common conversion pitfalls (and fixes)

Even with the best intentions, conversion signals often go wrong. Here are the issues I encounter most frequently and their practical solutions.

What if my event doesn’t fire enough?

The problem: Your primary conversion event happens too rarely (say, 20 times per week) and you’ll never exit learning phase.

The fix: Use a higher-volume proxy event that occurs earlier in the funnel. For subscription businesses, optimize for “Start Trial” instead of “Subscribe.” For e-commerce with high-consideration products, optimize for “Add to Cart” instead of “Purchase.”

Yes, you’re technically optimizing for a different event, but you’re giving the algorithm enough data to learn patterns. Just make sure you include predicted value data so it knows some add-to-carts are more valuable than others.

What if it doesn’t match GA4?

The problem: Your ad platform’s conversion data doesn’t align with GA4 and you’re spending hours trying to reconcile them.

The fix: Stop trying to make them match. It’s expected. They’re measuring different things through different methodologies.

Optimize for algorithm learning, not report matching. Use GA4 for understanding user journeys and website performance. Use platform conversion data for campaign optimization.

What if it doesn’t carry value?

The problem: You’re sending conversions without any value data attached.

The fix: Always aim to send revenue or value data with your conversions:

• E-commerce: Send the transaction value

• Subscription products: Send the predicted LTV based on which plan they chose

• Lead gen: Assign values to different lead types based on your historical close rates and deal sizes

• Mobile apps: Model predicted LTV based on early user behavior, even if it’s not perfectly accurate

An imperfect value signal is far better than no value signal at all.

The mindset shift: optimization ≠ measurement

Many teams conflate optimization with measurement, and it causes endless confusion. They’re trying to use the same data for two different purposes, then getting frustrated when it doesn’t work perfectly for both.

For Optimization: Use Platform Conversions

Facebook Ads Manager, Google Ads, TikTok Ads Manager should receive:

• Clean, high-frequency conversion events

• Value data attached

• Fast implementation (fire within 24 hours)

• Comprehensive coverage (don’t worry if numbers don’t match your source of truth)

Their job is to teach algorithms who to show your ads to.

For Measurement: Use Incrementality Tests

Holdout tests, geo experiments, conversion lift studies, marketing mix modeling should:

• Tell you whether your ads actually caused incremental business results

• Answer the question: is this advertising working?

• Guide strategic decisions about overall budget allocation

These are different questions requiring different tools. Your optimization data needs to be fast and comprehensive so algorithms can learn quickly. Your measurement data needs to be accurate and causal so you can make strategic decisions.

Stop trying to make everything match. Start making everything learn.

Beware of the role of data in campaign optimization

If your campaigns still look like they did in 2015, it’s not your creative, it’s your signals.

The platforms have evolved dramatically. The algorithms are genuinely sophisticated now, capable of finding customers and optimizing bids better than we ever could manually. But they’re only as good as the data we feed them.

Most teams are still fighting the algorithm instead of feeding it. They’re over-segmenting campaigns, tracking too many events, missing value data, and wondering why performance has plateaued.

The solution isn’t more manual control. It’s better conversion signals.

I launched FixMyTracking this year specifically to help advertisers fix their ad platform conversion tracking. After eight years of freelancing in marketing data for companies like Microsoft, Veed.io, WeTransfer, and Superbet, I kept seeing the same issues repeatedly: smart teams with sophisticated strategies, held back by broken conversion signals they didn’t even know were broken.

Subscribe now

Dynamic URL parameters, also know as ValueTrack or auto-tagging parameters, solve this by automatically pulling campaign identifiers from ad platforms at click time. Instead of hardcoding campaign names or manually typing IDs, you use placeholders that the platform populates when someone clicks your ad.

Here’s what that looks like:

Manual approach:
https://example.com/?utm_source=google&utm_campaign=summer_sale_2024

Dynamic approach:
https://example.com/?utm_source=google&utm_campaign={campaignid}

At click time, {campaignid} becomes 12345678 or whatever the actual campaign ID is.

Three problems are solved by dynamic parameters

1. Manual Work and Human Error

Building UTM parameters manually (this includes using a Google Sheet or another front end tool) takes precious time away from marketers and invites mistakes. Typos in campaign names. Inconsistent capitalization. Missing parameters. Different formatting conventions across team members.

These errors compound when you’re managing dozens or hundreds of campaigns. Cleaning the data after the fact takes time, and you can never be completely sure you’ve caught everything.

Dynamic parameters eliminate this by pulling values directly from the ad platform. The platform knows the campaign ID, ad group ID, and ad ID. It populates these automatically at click time. No typing, no errors.

2. Campaign Learning Phase Resets

When you change a destination URL in an ad (including the UTM parameters), many platforms reset the campaign’s learning phase. The algorithm treats it as a new ad and starts optimization from scratch.

This is particularly painful when you realize you made a tracking mistake and need to fix it. You can either leave the broken tracking in place or accept the performance hit from resetting learning.

The exception here is if you’re using tracking templates, like Account-level tracking. You can update tracking parameters without changing the ad’s final URL. The platform doesn’t treat it as a new ad, so learning continues uninterrupted.

3. Unreliable Joins Between Systems

The core technical problem is joining ad platform data with your first-party conversion data.

Different channels pass back data in different ways. Sometimes you can connect ad data (spend, clicks, impressions) straight to first-party results.

Without stable identifiers, you can’t answer questions like:

• Which specific ads drove registrations that converted to paying customers?

• What’s the LTV of users acquired from Campaign ID 12345678 versus Campaign ID 87654321?

• How does performance differ between ad groups within the same campaign?

Campaign names aren’t reliable join keys because they change, aren’t unique across accounts, and contain special characters. IDs are stable, unique, and work consistently across exports and APIs.

Dynamic parameters for IDs standardize this by creating a one-to-one link across all sources. Every campaign, ad, or placement can be tied directly to business outcomes measured via your first-party data.

Why use IDs, instead of names

Most teams initially want to pass campaign names in UTMs instead of IDs. This feels more intuitive. You can read the URL and understand what campaign it came from.

But campaign names create problems.

Names change. You might start with “Q1_Acquisition” and later rename it to “Q1_Acquisition_Mobile.” If your reporting depends on matching names exactly, you’ve just broken your historical data.

Names aren’t unique. Two campaigns might have similar or identical names, especially across different markets or account structures.

Names contain special characters. Campaign names often include spaces, parentheses, or other characters that break URLs or require encoding. This creates ugly URLs and potential parsing errors.

Names expose internal logic. If your campaign name is “High_Intent_Premium_Users_Retargeting,” that information is now visible in every URL that gets shared or clicked.

IDs solve all of these problems:

• Stable: Campaign ID 12345678 never changes, even if you rename the campaign ten times

• Unique: Every campaign, ad group, and ad has a unique identifier across the platform

• Clean: IDs are just numbers. No special characters, no encoding issues

• Private: An ID reveals nothing about your strategy or targeting to people who see the URL

• Joinable: IDs map directly to ad platform exports, API data, and any other system that references the same campaigns

The tradeoff is readability. Looking at utm_campaign=12345678 tells you nothing about what the campaign does. But that’s what your ad platform naming conventions are for, you join the ID to the campaign name in your reporting layer.

How dynamic parameters and URL work together

Dynamic parameters and naming conventions serve different purposes. Understanding this distinction prevents confusion about what each system provides.

Dynamic parameters:

• Capture stable, machine-readable IDs

• Used for joins between ad platforms and BI systems

• Do not contain strategic context (geo, funnel stage, creative format)

• Automatically populated at click time

Naming conventions:

• Define structured campaign/ad names to encode strategy

• Example: RO_Acq_Prospecting_Search_Q4

• Enable grouping, segmentation, and parsing in reports

• Manually created when setting up campaigns

The two systems work together. Dynamic parameters give you the join keys. Naming conventions give you the strategic dimensions.

The best practice is to use both IDs and names

The most effective setup combines both systems:

1. Use dynamic macros for joins (IDs)

2. Use structured naming conventions for strategic dimensions

3. Combine both in reporting:

   • Join UTMs with platform exports via IDs

   • Parse names for strategy-level reporting

This gives you the stability of IDs and the context of structured names.

Common mistakes I’ve seen clients make when it comes to UTMs

Don’t Use UTMs for Internal Links

UTMs are for external traffic coming to your properties. They should not be used for navigation between pages on your own site, like homepage banners, in-product CTAs, or support links.

Using UTMs internally resets attribution. Tools that rely on last non-direct attribution will incorrectly assign credit to the internal source (e.g., utm_source=homepage_cta) rather than the true origin (e.g., Google Ads).

If you want to understand the role of internal touchpoints in the user journey, use event tracking or custom dimensions instead. GA4, product analytics platforms, and tag management setups offer ways to track internal behavior without polluting acquisition data.

Beware of Duplicate Parameters

If both tracking templates and final URLs define the same parameter, behavior differs by platform. Test carefully before rolling out.

Empty or Malformed Values

Some macros may not populate in certain conditions:

• Using {keyword} on display or video campaigns

• ATT-limited iOS ads on Meta

• Campaigns without the structural level you’re trying to capture, like Ad Groups on PerformanceMax

Empty or malformed values can break reports. QA across different campaign types before rolling out. Consider adding fallback logic in your reporting layer to handle missing values gracefully.

Most ad platforms today offer dynamic parameters

Whether before it was only available for Google and Meta campaigns, it is now a standard practice across many ad platforms.

Note the inconsistency. Google uses single curly braces. Meta uses double curly braces. TikTok uses double underscores. This is why standardizing your parameter names matters more than the platform’s macro syntax.

Not all platforms support dynamic parameters natively.

X (Twitter)

X has no official dynamic parameter support. There’s also no bulk ad upload functionality, making it impractical to manually add IDs to URLs for every ad.

Programmatic (Open market, PMPs, Direct Deals)

Programmatic channels typically don’t offer standardized dynamic parameters. Instead, granularity comes from how you structure your campaign taxonomy and ad operations setup.

For DSPs like The Trade Desk or DV360, you’ll need to work with your ad server (e.g., Flashtalking, Campaign Manager) to understand what tagging is possible. Often this means building a consistent naming structure within the DSP and ad server that can be parsed in reporting, rather than using dynamic parameters.

How to start rolling out dynamic parameters

1. Start with One Platform

Don’t try to implement dynamic parameters across all channels at once. Start with your largest or most important channel.

Google Ads is usually the best starting point because it typically represents the largest share of paid media spend, the tracking template setup is straightforward, and account-level templates mean you only configure once.

2. Define Your Parameter Structure

Decide which parameters you’ll capture and how you’ll name them.

Minimum viable setup:

utm_source=google&utm_medium=cpc&utm_campaign={campaignid}

More complete setup:

utm_source=google&utm_medium=cpc&utm_campaign={campaignid}&utm_content={adgroupid}&utm_term={keyword}

With custom parameter:

utm_source=google&utm_medium=cpc&utm_campaign=summer_sale&campaign_id={campaignid}&adgroup_id={adgroupid}&ad_id={creative}

Keep parameter names consistent across platforms. Don’t use campaign_id on Google and cid on Meta. Standardization makes reporting easier.

3. Understand Where Parameters Go

How you implement depends on what the platform supports and your existing setup.

Platforms with tracking templates (Google Ads, LinkedIn, Microsoft Ads):

If you’re already using tracking templates, you can add or update dynamic parameters at any time without affecting ad delivery. Changes don’t reset learning phases.

Platforms requiring ad URL updates (Meta):

Parameters must be added to each ad’s final URL. Changing ad URLs can reset learning phases. Plan to implement this alongside other campaign changes (new creatives, budget adjustments, or new launches) to minimize performance impact.

4. Test with One Campaign

Before rolling out to your entire account, validate that everything works correctly.

Create a test campaign or use an existing small campaign. Add the tracking template. Click your own ads. Verify that:

• Parameters populate correctly in the landing page URL

• Your analytics platform receives the data

• The IDs match what you see in the ad platform

• No parameters are empty or show the literal macro text

Common issues to watch for:

Parameters show as literal text: utm_campaign={campaignid} instead of utm_campaign=12345678. This means the macro syntax is wrong for that platform.

Parameters are empty: The URL shows utm_campaign= with no value. This happens when the macro doesn’t apply (e.g., using {keyword} on display campaigns), or when special characters in the macro break reporting.

URL encoding breaks: Special characters in your destination URL conflict with parameter syntax. Use {lpurl} instead of hardcoding the landing page.

5. Verify Reporting Joins

Export campaign data from your ad platform with campaign ID, campaign name, impressions, clicks, and spend. Join this data with your analytics or conversion data using the campaign ID from your UTM parameter.

If the join fails or shows mismatches, common causes include data type mismatch (string vs. integer), leading/trailing whitespace in IDs, different ID formats between systems, or timing delays between ad platform and analytics.

Fix these issues in your test campaign before expanding.

6. Roll Out Gradually

Once your test campaign works correctly, expand to more campaigns.

Add the tracking template at the account level (if supported). This automatically applies to all current and future campaigns. Remove any campaign or ad-level tracking that conflicts with your new setup.

Monitor for issues during the rollout. Check that all campaigns show populated parameters, no campaigns show literal macro text, and reporting joins continue working.

7. Expand to Other Platforms

Once Google Ads works reliably, apply the same approach to other platforms.

For each new platform:

• Translate your parameter structure to that platform’s macro syntax

• Test with one campaign

• Verify joins work correctly

• Document platform-specific quirks

• Roll out to all campaigns

The goal is consistent parameter names across platforms, even though the macro syntax differs.

The technical part is easy, but the coordination can be a pain

Dynamic parameters automate tracking by pulling campaign identifiers directly from ad platforms. This eliminates manual UTM building, prevents learning phase resets, and enables reliable joins between ad data and business outcomes. The implementation is straightforward: start with one platform, test with one campaign, verify joins work, then roll out gradually.

The harder part is coordinating across teams. Media buyers need to understand why campaign naming still matters even with automated IDs. Analysts need to know how to join the new data structure. BI developers need to confirm the reporting stack can handle it. Everyone needs to agree on when and how to roll it out without breaking existing dashboards.

This is workflow infrastructure. It’s not exciting, but it’s the kind of change that compounds. Six months from now, you won’t remember the setup work. You’ll just have cleaner data and fewer tracking fires to put out.

If you’ve enjoyed reading this article, don’t forget to subscribe below! And if you’d looking for someone to guide your team, from data engineers to marketers, through this change, do reach out.

Subscribe now

The Not So Hidden Cost of Broken Tracking

Ever had that sinking feeling when your CMO asks why Facebook is reporting 5 conversions while your CRM shows 75? Or why your dashboards suddenly show conversion rates plummeting overnight? Welcome to the world of broken marketing tracking, where decisions stall and ad spend efficiency suffers.

In this guide, I'll walk you through my process for diagnosing and fixing marketing tracking issues. By the end, you'll understand:

• How to systematically identify where tracking failures occur

• Which debugging tools can help for different ad platforms

• How to build systems that make future troubleshooting easier

• Ways to establish trust in your data across teams

This isn't theoretical. These approaches have helped clients managing $1M+ media budgets across fintech, e-commerce, B2B SaaS, and other industries where data reliability directly impacts business outcomes.

The Algorithm Advantage You're Missing Out On

When campaigns perform well, tracking systems get ignored. But watch how quickly "let's check the tracking" becomes the rallying cry when numbers disappoint. This pattern repeats across almost every marketing team I've worked with.

In today's ad ecosystem, broken tracking isn't just annoying: it's inefficient. Modern ad platforms have moved beyond manual targeting to algorithm-driven optimization that relies on conversion signals to function. When tracking breaks, you're essentially forcing platforms back into manual mode, abandoning the sophisticated AI and machine learning technologies that make them effective.

Think about it: Meta's algorithm can process thousands of audience signals and bid adjustments per second, but only if it knows what a conversion looks like. Google's Smart Bidding can optimize for your specific business outcomes, but it needs clean conversion data to learn from. Without reliable tracking, you're stuck with manual audience targeting and bid management: an inefficient approach that wastes both time and money.

Or worse, spending money in the wrong place and not knowing until your bottom line starts reflecting it

For teams managing serious media budgets, this isn't a technical footnote. It's the difference between growth and waste. I've seen companies improve incremental CPAs not by changing their ads or landing pages, but simply by fixing broken tracking systems that were feeding poor signals to ad platforms. The algorithms were there, ready to optimize, but they needed better data to work with.

Five Steps to Troubleshoot What's Broken Now

When tracking breaks, you need a systematic approach to identify and fix the problem. Here's my process:

Step 1: Document Your Event Architecture

This sounds ridiculously obvious, but you'd be shocked how often teams can't answer basic questions like:

• What events are we measuring for each ad platform?

• When exactly do these events fire?

• Where do they fire from? (e.g. cGTM, Meta CAPI, GA4)

• What variables do they pass?

Create clear documentation that answers these questions and map out the data flow of the events. A simple flowchart will do. Without this basic alignment, troubleshooting becomes nearly impossible.

Step 2: Understand What Each Event Actually Measures in the Ad Platform

"We had 157 conversions" means absolutely nothing without context.

Attributed conversions is different from all conversions.

Some platforms only show conversions they take credit for like Google Ads. Meta Events Manager, on the other hand, shows all events fired, regardless if they were attributed to Meta. This alone can lead to confusion.

It’s core to look into what’s configured in your ad platform. What is your lookback window? Are you looking into post-click only, or also post-view? If you’re using catalog products, are all products verified?

Together, Steps 1 and 2 are critical to close the gap between marketing and engineering/data teams. I’ve seen engineering teams adamant that “tracking was working” because they completed step 1, while ignoring the impact of Ad Platform configuration in how events are measured.

Step 3: Compare Against Your Source of Truth

You need a reality check: a system that captures what actually happened, not what platforms claim happened.

For purchases, this is your order database. For leads, it's your CRM. Whatever it is, identify this source and use it as your benchmark.

Just don't expect perfect matches. That's not how this works. Discrepancies will always exist. I mostly use this to confirm that events are not double firing or duplicating.

Step 4: Verify Events Are Actually Firing When They Should

Before you can fix attribution, you need to confirm your events are firing when they should. Here are different ways to test them:

GA4 Debug View: Navigate to Admin → Debug View in GA4. Perform the conversion action on your site while Debug View is active. You should see the event appear in real-time.

Meta Test Events: Go to Events Manager → Test Events. Watch for the event to appear in the testing interface.

Chrome Network Console: Open Developer Tools → Network tab. Filter for your tracking domain (gtm, facebook, etc.). Perform the conversion and look for the HTTP requests being sent.

GTM Preview Mode: Click Preview in GTM, then navigate to your site. The preview pane shows exactly which tags fire and when.

I wrote here on how to make these steps easier with AI.

Step 5: Check Attribution

An event can fire successfully but still fail to attribute properly. It’s important to make a distinction here, specially when developers are saying “everything is firing as it should”.

Common reasons for “broken attribution” are:

UTM Parameter Issues: Your events might fire, but if UTM parameters are missing or named incorrectly, platforms can't attribute conversions to campaigns. Check that utm_source, utm_medium, and utm_campaign are properly formatted and consistent.

Consent Management: Privacy settings can prevent events from attributing even when they fire. This is especially problematic in Europe where consent rates can be below 50%. I’ve seen this be a common cause for MMP hiccups.

Four Activities To Improve the Speed and Confidence of Troubleshooting

Once you've fixed immediate issues, invest in systems that make future troubleshooting easier:

Activity 1: Store Raw Data of Your Source of Truth

Sometimes an event breaks in specific user journeys that basic debugging won't reveal.

In these cases, raw data becomes your savior. Set up GA4 for BigQuery or store your MMP data in your warehouse. Server-side tracking can offer a major advantage here. If you’re using only client-side and that’s your source of truth, then make sure to log these events.

From a business perspective, investing in proper data storage isn't just a technical nice-to-have. It's insurance against future diagnostic challenges. When a client implements comprehensive raw data collection, troubleshooting time typically decreases significantly.

Try as best as you can to centralize your data tracking in one place. If you only have to manage a single GTM container, one Segment account or if you send everything from BigQuery directly, then the risk of discrepancies (and double counting) also decreases.

Activity 2: Pass Identifiers for Isolation

Debugging without proper identifiers is like searching for a specific grain of sand on the beach.

Pass unique variables like transaction_id, opportunity_id, and user_id across your systems. These become crucial for pinpointing exactly where and when failures occur.

The organizational impact here is significant: when marketing, product, and data teams share consistent identifiers, cross-functional diagnostics become more efficient.

Activity 3: Build Proactive Monitoring

Don't wait for stakeholders to notice tracking problems. Build systems that alert you first:

Timeline Event Count Reports: Create basic timeline reports showing event counts over time. When counts suddenly drop or spike, something's broken. These visual indicators catch problems that raw numbers might miss.

Discrepancy Alerts: Set up automated monitoring between systems. If GA4 and Meta normally differ by 15% but suddenly show a 40% gap, investigate immediately. These alerts become your early warning system.

Regression Tests: Work with engineering to turn a simple “When this happens → What should happen” sheet into automated test cases. These run with every release to confirm marketing events still fire correctly—preventing new features from silently breaking your tracking.

The key insight here is that regular monitoring is far less expensive than emergency troubleshooting. A simple weekly review of key tracking metrics can prevent major crises.

Activity 4: Simplify to Reduce Failure Points

Tracking will break again. It's inevitable. Minimize the damage with these approaches:

Track only what's absolutely necessary. Every new event increases maintenance complexity. Fewer tracking points = fewer potential failures.

This relates directly to one of my core beliefs: most marketing problems are workflow issues, not tool problems. Complex tracking setups create process challenges that technical solutions alone can't solve.

Tracking Quality Determines Campaign Performance

Data isn't useful if people don't trust it, and broken tracking creates both trust and technology problems.

The true cost of broken tracking isn't just technical. It's that you're forced back into manual campaign management instead of leveraging the sophisticated algorithms that could be optimizing your spend.

Without reliable conversion signals, you lose access to the best technology ad platforms offer. Smart Bidding becomes unreliable. Lookalike audiences lose their effectiveness. Campaign Budget Optimization can't function properly. You're essentially paying premium prices for advanced platforms while using them like basic display networks from 2010.

Get this right, and everything else becomes possible. Get it wrong, and you're not just guessing expensively. You're abandoning the algorithmic advantages that separate winning campaigns from wasteful ones.

Do you need help fixing your tracking?

I’ve recently launched FixMyTracking, a conversion tracking auditing service. It’s as simple as it sounds: I investigate the events from your paid campaigns and prepare clear fixing instructions and recommendations for you to share with your engineering team.

Here's what's happening: last month-ish, I've launched FixMyTracking, a new service that helps advertisers fix broken ad platform conversion tracking. (Yes, it’s a descriptive name.) And I'm hiring someone to work with me part-time, freelance, remote.

I launched FixMyTracking to fix the thing that breaks every campaign

Every project I took started with the same broken foundation

For over 10 years, I've helped teams improve campaign performance through better data. First as a growth marketer, then as a freelancer.

And I kept seeing the same pattern: smart marketers. Solid creative. Decent strategy. But tracking was broken or, at the very least, tracking the wrong thing. So campaigns couldn't optimize.

Conversion events were missing, misfiring, duplicating, or poorly mapped. Sometimes the data was working, but teams didn't trust it, so decisions stalled.

Either way, the outcome was identical: wasted spend, confused teams, and performance that never matched potential. As a defensive mechanism, marketers often turned to manual control: splitting ad sets, tweaking bids manually, debating attribution windows. Not harvesting the best of 2025’s algorithms.

And here's the kicker: Every project I took on started with fixing this foundation. No MMM, no incrementality modeling, no sophisticated ROAS analysis until there was confidence the basics were bulletproof.

So I built FixMyTracking to solve exactly that problem.

It's a one-time audit that actually fixes your conversion tracking

FixMyTracking is a comprehensive, one-time audit of your conversion tracking setup across platforms like Meta, Google Ads, TikTok, and more. You get a thorough investigation of all your conversion signals, a detailed blueprint showing what's broken and exactly how to fix it, plus a wrap-up call to walk through results and plan next steps.

No retainers. No junior analysts. No outsourcing to overseas teams. Just clear, actionable fixes so your ad platforms can finally optimize properly.

If your campaigns feel stuck and you suspect tracking might be the culprit, you can book a quick call to see if it's a fit.

I'm hiring someone who gets excited about debugging conversion events

Launching FixMyTracking is a big step for me. It's the first time I've transformed part of my consulting work into its own brand. While I'm excited, I also know I can't scale it alone.

So I'm hiring.

If you get genuine satisfaction from hunting down rogue conversion events, debugging CAPI setups, or untangling messy UTM schemes, this role might be perfect for you.

The role is part-time, freelance, remote, focused on technical tracking audits for paid media. It's ideal for marketing analysts or technical marketers who want to go deeper into signals, not just surface-level reporting.

You'll work directly with me (no corporate layers or middle management). You'll see firsthand how high-growth teams use and misuse their marketing data. And you'll get an inside look at how a premium solo consultancy actually operates.

Read the full job description and application process here. Please please please (how Sabrina Carpenter of me) forward this to anyone who might be interested.

What I'm hoping for (and how you can help if you want)

So that's my news: FixMyTracking is officially live, and I'm hiring (and admittedly a little nervous about it). If you've read this far, thank you for being part of this journey.

If you want to share this job posting with a sharp marketing analyst you know, I'd genuinely appreciate it. If you want to book a FixMyTracking consultation, the initial call is free. Or if you just want to send encouragement, I'll take that too.

Thanks for reading an atypical issue. Regular marketing measurement content returns next time.

The Measurement Mayhem We're Living In

Let's face it: the mobile marketing ecosystem has been turned upside down since Apple dropped the App Tracking Transparency (ATT) bomb in 2021. What was once a relatively straightforward process of tracking user journeys from ad impression to in-app purchase has become a chaotic scavenger hunt for reliable data.

For growth teams managing substantial media budgets, this has created a frustrating reality: those attribution tools you've relied on for years—the ones your boss is still expecting pristine reports from—simply aren't providing the same level of certainty they once did. Yet your CEO still wants to know exactly which campaigns are driving revenue, preferably by yesterday.

This guide explores how sophisticated marketing teams can measure and optimize mobile app campaigns without defaulting to an MMP, providing platform-specific approaches and their genuine business implications. While centralizing attribution through an MMP remains valuable for many organizations (and keeps things simpler), understanding these alternative approaches offers several competitive advantages:

1. Privacy resilience: Building first-party measurement infrastructure creates protection against further privacy changes (and trust me, they're coming)

2. Cost efficiency: Reducing dependence on third-party attribution tools can significantly lower costs

3. Enhanced data ownership: Direct implementation gives you more control over your measurement methodology

4. Platform-specific optimization: Tailored approaches for each channel can yield better performance than one-size-fits-all attribution

First things first: What is an MMP?

Mobile Measurement Partners (MMPs) are specialized analytics platforms that help app marketers track and attribute user acquisition across advertising channels. Popular MMPs include Adjust, AppsFlyer, Branch, Kochava, and Singular. These platforms traditionally served as the single source of truth for attribution data, helping marketers understand which ads led to app installs and subsequent in-app actions.

A key advantage of MMPs is their ability to provide centralized attribution that handles deduplication across campaigns. When users interact with multiple channels during their journey (for example, seeing both a Google and Meta ad before installing), MMPs apply attribution models to credit the appropriate source. This deduplication capability is difficult to replicate with platform-specific attribution solutions.

The Post-ATT Reality: How MMPs Have Changed

Before diving into platform-specific approaches, let's acknowledge how fundamentally different MMPs are in our new privacy-centric reality.

Pre-ATT World (Before 2021)

Mobile Measurement Partners like Adjust, AppsFlyer, and Branch operated with remarkable precision in the pre-ATT world:

1. Deterministic attribution: MMPs could directly link ad interactions to app installs and post-install events using device identifiers (IDFA on iOS, GAID on Android)

2. Cross-platform visibility: User journeys could be tracked across channels, allowing for sophisticated multi-touch attribution

3. Real-time optimization: Ad platforms received granular conversion data, enabling algorithmic optimization toward high-value users

4. User-level reporting: Marketers could analyze individual user behaviors, LTV calculations, and cohort performance

Post-ATT World (2021-Present)

The introduction of App Tracking Transparency fundamentally altered this landscape:

1. Probabilistic attribution dominance: With IDFAs largely unavailable, MMPs shifted to probabilistic methods using IP addresses, timestamps, and device characteristics to make educated guesses about attribution

2. Limited data granularity: Privacy thresholds and conversion values significantly restrict the depth of information available

3. Delayed reporting: Real-time optimization has been replaced by aggregated, delayed reporting windows

4. Increased complexity: Implementation now requires platform-specific knowledge and ongoing maintenance

For marketing teams, this means MMPs have transformed from comprehensive attribution solutions to integration layers that aggregate data from various platform-specific measurement technologies—often at significant cost but with diminishing unique value.

Platform-Agnostic Measurement Approaches

Some ways of measuring mobile app campaigns work regardless of the platform.

Option 1: Web-to-install campaigns

For most of the cases in this article, we’re referring to traditional “install” campaigns that take the user directly to the App Store listing. However, taking the user to a web page first is an alternative that can decrease acquisition costs while also improving measurement. In this case, you can use:

• Ad Platform tags (like Meta and Google conversion pixels)

• APIs (like CAPI)

Option 2: SKAdNetwork (SKAN)

SKAdNetwork provides some post-install data but with significant limitations:

• Allows tracking of conversion values (0-63) that can encode basic user behavior

• Provides campaign ID but no keyword-level data

• Subject to privacy thresholds that further limit reporting

• Requires specialized implementation knowledge

It can be leveraged alongside deterministic-attribution (an MMP or a custom-built solution) to create a deduplicated one-view reporting.

Option 3: Incrementality Testing

For larger organizations with sufficient campaign volume, incrementality testing offers a method to measure impact of different initiatives.

• Identify geographic regions with similar characteristics

• Run ASA ampaigns in test regions while keeping control regions dark

• Measure the lift in overall conversions between test and control regions

• Calculate the incremental impact of campaign spending

This approach requires:

• Significant campaign volume

• Statistical expertise for proper test design and analysis

• Patience, as results aren't available in real-time

Option 4: "How Did You Hear About Us" (HDYHAU) Surveys

While less technical, implementing in-app surveys can provide directional insights:

1. Prompt new users with a simple survey asking how they discovered your app

2. Track responses mentioning for example "App Store Search" or similar terms

3. Compare survey data with ASA spend to estimate efficiency

This method is most valuable as a directional signal rather than precise attribution. You can also get around a product implementation by emailing new users a survey. I’ve personally done this before, and it’s a great starting point.

Option 5: MMM (Media Mix Modeling)

Media Mix Modeling offers a holistic approach to understanding marketing effectiveness across channels:

• Uses statistical analysis to determine marketing impact on key business outcomes

• Measures effectiveness across all channels, including those with limited direct attribution

• Should account for external factors like seasonality, market conditions, and competitor activity

Implementation can be done through specialized MMM vendors or in-house data scientists, often leveraging open-source frameworks like Meta's Robyn.

MMM isn't for everyone. It typically requires 1-2 years of historical data across multiple channels and significant analytical resources, making it better suited for established apps with diverse marketing strategies.

Platform-Specific Measurement Approaches

Let's examine how to effectively measure campaigns on each major platform without an MMP, including implementation requirements and business implications.

Apple Search Ads (ASA): The Measurement Enigma

Apple Search Ads presents perhaps the most challenging measurement scenario without an MMP, as its core functionality focuses exclusively on install attribution with no native capacity for tracking downstream events like trials or purchases.

Technical Implementation

Apple Search Ads provides basic install attribution out-of-the-box through a proprietary matching system that doesn't require user opt-in. However, there are several critical limitations:

1. Install-only optimization: ASA campaigns can only optimize for installs, not downstream conversions like trials or purchases—regardless of whether you use an MMP

2. No feedback loop: Unlike other platforms, there's no native mechanism to send conversion data back to ASA for optimization purposes

3. Limited keyword insights: Without downstream conversion data, it's impossible to determine which keywords drive meaningful business outcomes versus merely generating installs

For organizations needing more robust ASA measurement, the only MMP alternative for measuring deterministic post-install behaviour (downstream conversion data) is to build something custom.

AdServices API (Custom Integration)

The AdServices framework offers the most comprehensive solution, as this is the same API that MMPs themselves builds on top of. It requires significant development resources, but on the other hand I’ve also seen in FixMyTracking development teams run around in circles trying to understand why MMPs are not working as intended.

This implementation allows you to:

• Deterministically attribute installs to specific ASA campaigns

• Connect user activity to the originating campaign

• Build custom reporting based on this attribution

However, it requires:

• iOS development resources

• Backend infrastructure for token validation

• Regular maintenance as Apple updates the framework

Meta (Facebook & Instagram): The SDK+CAPI Power Couple

Unlike ASA, Meta offers multiple viable options for measurement without an MMP, with a hybrid approach providing the most comprehensive solution.

Meta SDK + Conversions API Hybrid Approach

This approach combines client-side tracking via Meta's SDK with server-side events through the Conversions API, providing comprehensive measurement across the user journey:

1. Meta SDK Implementation:

   • Tracks app installs and early funnel events directly in the app

   • Captures user engagement metrics that require on-device context

2. Conversions API:

   • Tracks high-value events like trials and purchases server-side

   • Sends conversion data directly from your servers to Meta

   • Leverages PIII (like user emails) to enhance attribution

3. RevenueCat Integration:

   • RevenueCat already tracks subscription events

   • Configure RevenueCat to send events to Meta CAPI

   • Establish a robust data pipeline from RevenueCat to BigQuery for analysis

The hybrid approach provides several advantages:

• More comprehensive coverage than using an MMP alone

• Redundancy in tracking critical conversion events

• Higher match rates by combining multiple identification methods

And, if you rely on SKAN, using Meta’s SDK means Meta controls how conversion values are set—making optimization easier but limiting your visibility. With an MMP SDK, you define the conversion value logic yourself, giving you more transparency and flexibility across ad channels.

Advantages of CAPI over client-side tracking (MMP or SDK)

Regardless of whether you’re deciding whether to SDK or MMP, conversion tracking with APIs provide unique advantages for campaign optimization.

• Greater control over what data you share: With the Conversions API, you choose exactly which events and user data to send to Meta, aligning with both business goals and privacy requirements.

• Flexible event timing: Unlike client-side SDKs or MMP postbacks, CAPI lets you send events after the moment they occur—perfect for lead scoring, onboarding milestones, or any conversion logic that depends on post-signup behavior.

Business Impact

The stakes for proper Meta measurement are extraordinarily high:

The Algorithm Optimization Gap: Meta's algorithms are only as good as the data you feed them. Without proper conversion tracking, you're essentially asking Meta to optimize for the wrong outcomes. In practice, this means paying premium CPMs for users who install but never convert.

Real-World Consequences: One wellness app I worked with discovered they were losing approximately 40% of attributable conversions by relying solely on Meta's SDK without the Conversions API complement. Their actual CAC was substantially lower than reported, and they had been unnecessarily constraining growth based on incorrect data.

Campaign Performance Impact: When another client implemented the SDK+CAPI hybrid approach, they saw a 22% decrease in cost-per-trial within two weeks—not because their actual performance improved, but because Meta's algorithms finally had accurate data to optimize against. Their growth team had been fighting an uphill battle with incomplete data for months.

The bottom line: if Meta represents a significant portion of your acquisition budget, implementing proper conversion tracking isn't optional—it's the difference between scaling profitably and potentially killing campaigns that are actually performing well. You can’t rely on MMP alone.

Other social platforms

Over the last years, other social ad platforms have been following Meta’s route and launching their own CAPIs too. That is the case for TikTok’s Events API and Pinterest’s API.

Google: Built-in Solutions Already Available

Google provides the most straightforward MMP alternative through its Firebase platform:

• Firebase Analytics can track installs and in-app events

• Integration with Google Ads provides attribution for app campaigns

• Data flows into BigQuery for cross-platform analysis

If you're already using Firebase, you likely have the foundation for Google Ads measurement without an MMP. The key challenge isn't implementation but ensuring proper monitoring and troubleshooting frameworks to maintain data accuracy and catch potential errors.

For Google Ads web campaigns (which you might explore in the future), Google Analytics would be the primary tracking mechanism. Both Google Analytics and Firebase feed data into BigQuery, allowing you to analyze performance across platforms in a single environment.

Business Impact

The Firebase approach to Google Ads measurement offers several business advantages:

Direct Algorithmic Feedback: Unlike some platforms, Google Ads can directly utilize Firebase conversion data for optimization, creating a tight feedback loop that improves campaign performance. Keep in mind Google owns both sides (Firebase and Ads), and in cases you’re advertising Android—well, they also own that.

Cost Efficiency Without Compromise: Firebase provides measurement capabilities that rival most MMPs for Google campaigns at no additional attribution cost. One gaming client realized they were spending over $100,000 annually on an MMP that was primarily being used for Google Ads campaigns—all of which could have been tracked effectively through Firebase.

Enhanced First-Party Data Collection: Because Firebase data flows directly into BigQuery, marketing teams gain richer insights beyond simple attribution, including behavioral patterns that can inform product development and retention strategies.

tROAS campaigns: You can only run Google Ads tROAS campaigns if you’re using Firebase. Otherwise, you’re stuck with tCPA. This is Google’s “carrot” way of onboarding companies on Firebase.

However, if you’re using something else than Firebase for product analytics (e.g. Mixpanel, Amplitude), then supporting Firebase just for the Google Ads use case and don’t plan on running tROAS, this could not be the best use of resources.

Programmatic Networks (e.g., AppLovin): The Most Challenging Case

Programmatic networks like AppLovin present the most difficult measurement scenario without an MMP:

• These networks rely heavily on postback systems between MMPs and their platforms

• AppLovin has an SDK, but it's for publishers (apps that want to monetize by showing ads), not for tracking install attribution

Creating your own solution would require building:

1. Click Tracking + Attribution Logic

   • Capture user identifiers from ad clicks (device ID, click ID)

   • Match these clicks to the user using your own logic on app open or install

2. Event Tracking + Real-Time Postbacks

   • Log in-app events (like purchases) in your backend

   • Send these events back to AppLovin via server-to-server (S2S) postbacks in near real-time

This represents significant technical complexity similar to building your own mini-MMP—likely not justified unless you're spending millions on these networks.

Making the Decision: MMP vs. Alternative Solutions

When deciding whether to continue using an MMP in 2025, consider:

1. Ad platform mix: If you're heavily invested in programmatic networks, an MMP may still be necessary. If you're primarily using Google and Meta, alternatives are more viable.

2. Technical resources: Building and maintaining non-MMP attribution systems require development resources and ongoing maintenance. Engineering teams often have an inherent bias toward building custom solutions, so take these recommendations with a grain of salt when evaluating the true cost.

3. Measurement priorities: If campaign-level ROAS is sufficient (vs. creative or keyword-level), simpler attribution models may work.

4. Budget considerations: It's very difficult to troubleshoot install attribution, so the math will usually skew toward purchasing an MMP. Most MMPs charge per paid install, so costs scale linearly with spend, making it more cost-efficient than dedicating valuable engineering resources to building and QA-ing custom solutions.

5. Data infrastructure: Strong data warehouse capabilities make alternative solutions more feasible by allowing you to centralize and analyze data from multiple sources.

For many companies with moderate budgets focused on major platforms, a hybrid approach may work best—using platform-specific solutions for Google and Meta while using an MMP primarily for programmatic networks or considering whether to invest in these channels at all given the measurement challenges.

There’s no perfect way forward

The reality is that perfect attribution no longer exists in the post-ATT world, whether you use an MMP or not. While ATT has primarily affected iOS, the impact is significant given that iOS accounts for the majority of app revenue globally. With Google's Privacy Sandbox on the horizon for Android, the attribution landscape will continue to evolve.

The key is matching your measurement approach to your specific marketing mix, technical capabilities, and business objectives. For many companies, the centralized attribution and deduplication capabilities of MMPs still provide value that outweighs their cost, particularly when factoring in the engineering resources required to build and maintain alternative solutions.

Whatever approach you take, focus on establishing frameworks for monitoring and troubleshooting to ensure data accuracy, as the technical implementation is just the beginning of maintaining reliable attribution in a privacy-first world.

P.S. Obviously, I can’t escape myself, so I’m afraid this “Thought Leadership” is still very tactical with multiple practical examples of how I’m using AI for measurement. I hope you enjoy the read!

Data tooling can give “superpowers”

Marketing measurement has always been a messy tangle of technical requirements and creative thinking. With the modern data stack, especially post-dbt, we saw a split between data engineers and analytics engineers. Data engineers built the plumbing and make sure it doesn't leak, while analytics engineers (domain experts) actually knew who needed water and how much.

AI is now creating a similar wave by putting data capabilities directly in the hands of creative marketing professionals. No, marketers won't suddenly transform into analytics engineers (thank God, can you imagine the Slack arguments on discrepancies?), but they'll finally be able to own parts of the workflow that previously had them begging technical teams for help.

The 6 ways I’ve been using AI for marketing measurement

Like Timo and I often repeat in the Attribution Masterclass, people equate marketing measurement to MTA (Multi-Touch Attribution). But there are tons of other activities within measurement. And I’ve been using AI to help me with some of these.

1. Pixel troubleshooting

Nothing kills marketing momentum faster than a broken tracking pixel. I’ve seen marketing teams watch their campaign burn money while they waited for engineering to get to their ticket in the next sprint.

AI has been making troubleshooting significantly easier for me. I export GTM containers and Chrome network console logs, throw them at GPT and ask for documentation. What’s firing, when does it fire, what’s the conversion event ID, what variables pass with it, etc.

Marketers often still need engineering to publish the fix, but we can own the root cause analysis and maybe even share the necessary code changes making it all significantly faster.

2. Data analysis

Translating natural language into code is the bread and butter of AI. If used wisely, it can help even the best coders move faster.

SQL queries: Before AI, if you wanted to tweak an attribution model, you'd have to either learn SQL or wait for a data analyst to help. Now you can describe what you want in plain English, and AI generates the SQL code you need. That said, the quality of the output heavily depends on how well you can describe your requirements, and you'll still need someone who can validate that the query actually does what you think it does. Understanding the data you’re gathering is also a must-have.

Excel complications: We've all spent hours troubleshooting VLOOKUP formulas that inexplicably stopped working. I use AI all the time to create Excel formulas, like extracting values from strings (e.g. from naming conventions) or cleaning up data.

Regular expressions: Regex remains one of the most intimidating aspects of data manipulation. I don’t use it often, but sometimes I encounter a situation where Regex is the only way out. Previously, due to lack of practice, anything-Regex would take me hours, but luckily for me AI can handle many common regex patterns. I have to say though, it’s still a lot of “trial and error”. But again, faster.

Do you still need some technical foundation? Absolutely. The learning curve has become more manageable, but this isn't a magic wand that eliminates the need for data understanding. I still think knowing the basics of SQL goes a long way. It's more like having a really good data assistant: helpful, but you still need to know enough to ask the right questions and spot when something doesn't look right.

3. Data Cleaning

Raw (marketing) data is messy. Campaign names with typos, inconsistent UTM parameters, duplicate entries, and fields that should be numbers but are stored as text—it's a constant headache that eats up hours of manual work.

AI has become my go-to solution for cleaning up these datasets. I can describe the cleanup I need in plain English: "Remove duplicates based on email and campaign ID," "Extract the campaign source from these messy UTM strings," or "Standardize these campaign names that have different capitalization and spacing." What also works really well for me is sample data: I upload some messy data and how it should look when clean, and then I send GPT the full dataset.

4. Reading documentation

API documentation has long been a barrier between marketing teams and technical implementation. These dense manuals often seem designed for developers rather than the marketers who need to use them.

AI can help translate technical documentation into more accessible language. When I encountered an error with Branch (a mobile measurement partner) for a client, I was able to get a clearer explanation by sharing the docs with AI. This approach worked well in that specific case, though results can vary depending on the complexity of the API and how good the documentation is. It’s surprising how many tools (looking at you Google) have poor documentation.

This can save significant time for straightforward implementations, but it's worth noting that AI sometimes misses nuances in API behavior that only come up in production. For critical integrations, you'll still want technical review before going live.

5. Building landing pages

Campaign-specific landing pages—for example, for a podcast sponsorship or influencer partnership—can be a great way of isolating results of marketing initiatives.

AI-powered tools can significantly speed up this process for many use cases. I've been able to create custom landing pages for podcast appearances much faster than traditional methods. Recently, I even created the entirety of FixMyTracking using Lovable.

There are still a lot of caveats, like maintaining design systems but I believe this will only improve with time.

6. Creative analysis

This has a big caveat: I haven't used AI for this yet, but I know of multiple vendors who are attempting to solve this. And I believe this is possible and will be very valuable if/when done right.

Until now, most creative analysis—like understanding what features, messaging, or image types perform best in ads—has been dependent on manual naming conventions. Humans had to input information about what's actually in the ad: whether it shows the product, mentions a discount, what’s the hook, etc.

But this approach has all kinds of problems. People can't be bothered following naming conventions consistently. There are always errors. And manual tagging can harm speed of testing.

AI vision models could analyze creative assets directly, identifying elements like messaging hooks, visual features, and emotional tone without human input. The ability to automatically "interpret" creatives, extract characteristics, and cross-reference performance across your entire creative library could eliminate one of the biggest bottlenecks in creative optimization.

The future for marketing measurement

Anyone that reads this newsletter or my Linkedin knows that I always aim to be balanced and not overhype things. However, I think those that claim AI is not genuinely transformative for marketing is either wrong or just wanting to be a contrarian. It is. It is giving technical superpowers to creatives. It's changing what skills matter and how teams are structured.

The hiring implications are real. Teams need fewer people who can write SQL or build dashboards, but more people who can design good experiments and interpret results strategically. The technical barrier to doing measurement has dropped significantly, but the analytical thinking barrier remains as high as ever.

AI is changing who can do measurement work and how quickly it gets done. It's not changing the fact that good measurement requires people who understand how marketing actually works in your specific business.

But there’s one thing AI cannot change, and it’s that marketing measurement is a competitive advantage.

While AI democratizes the technical skills needed for measurement, it also creates a new risk: the commoditization of insights. When everyone can generate the same reports through AI prompts and reach the same conclusions about which channels or audiences to target, we end up in bidding wars for identical inventory using identical playbooks.

The brands that will be the ones asking better questions, finding unique audience segments, identifying undervalued channels and translating standard metrics into differentiated strategy. AI can help you analyze data faster, but it can't tell you what makes your business unique or help you avoid competing for the same users as everyone else.

Harnessing the best out of ad platform optimization

Marketing teams spend millions optimizing toward sign-ups, even when sign-ups are not all created equal. Some churn quickly. Some convert late. Others bring in outsized revenue, renew again and again, and invite their entire company. You miss that nuance if you’re optimizing paid campaigns for sign-ups only by running tCPA instead of tROAS.

Businesses with long sales cycles or freemium models face difficulties moving from tCPA to tROAS because partial or complete revenue only comes after conversion lookback windows (always a maximum of 30 days). Without predicting the value of conversions (visitors, signups, or leads), marketers can’t harness the best of ad platforms’ smart bidding algorithms and are stuck targeting poor-quality and high-quality leads.

In this article, we’ll cover:

• What is the impact of conversion value prediction, and which businesses need it the most

• What are the different types of conversions you can predict value for, and which one(s) should you use

• What’s required to implement this technically

• How to run your first test—including setup, model design, and analysis

Whether you’re working with long sales cycles, freemium models, or just struggling to scale CAC efficiently, this guide will help you assess if conversion value prediction is worth implementing—and how to do it well.

In what contexts is bidding using predicted conversion value impactful?

In most cases, predicting conversion value is a complex activity involving both data and marketing teams. It’s not a necessary activity, so it shouldn’t be treated as a default playbook. But it can create a meaningful business impact when:

1. Conversion to a paying customer happens a long time after an ad click
   In long sales cycles, for example, for medical equipment, financial products, or enterprise tools, conversions may take weeks to months, making it too late for smart bidding to react.

2. LTV varies significantly across user segments
   SaaS is a clear example. Depending on seat count, pricing tier, and churn, one user might be worth 50x another. Optimizing based on cost per signup ignores this nuance. This is also relevant for mobile games, where you commonly see 20% of users are responsible for 80% of the revenue (“whales”).
   

In these cases, revenue prediction helps bring valuable data back into the platforms—so you can bid higher for leads that matter, and lower for the ones that don’t.

Why predicted values can really impact paid performance

Paid platforms like Meta and Google no longer deliver the best results with manual audience targeting. Instead, performance depends on the quality of creatives and the signals you feed back—especially when running conversion or value-optimized campaigns.

B2B companies, for example, often bid based on a cost per lead target which treats all conversions equally (setting a target CPA bid in Google ads). But this leaves massive inefficiencies.

Let’s say you’re targeting a $5 CAC for all leads. But some users are only worth $3, while others are worth $100. If you're capping bids at $5 across the board, you're potentially blocking yourself from acquiring those $100 leads—because you’re not willing to bid more aggressively when it makes sense.

Imagine that winning a $100 lead would require a $10 CAC. Under your current ceiling, that user never enters the funnel—even though the return would be 10x. That’s money left on the table.

With conversion value predictions, you can:

• Run tROAS campaigns that bid based on predicted conversion value

• Adjust your bidding dynamically based on firmographic, behavioral or other data

• Power better optimization with stronger data signals, even before conversion to a paid customer

Example: Architecture of a conversion value prediction model for B2B

In this example we have a typical funnel for a B2B company, with different user actions leading to revenue. We can build separate models to predict values for each funnel step using machine learning or other scoring methodologies.

1. probability of the visitor converting into lead

2. probability of lead converting to paid customer

3. the predicted revenue once the customer converts

Each model may have different data inputs. The further down the funnel, the more information we can collect which can be used in prediction.

Once we have developed these models they are then combined in order to find the predicted value of the lead.

This value can be passed to smart bidding as conversion value for the lead allowing us to use ROAS instead of bidding on a cost per lead target.

What’s technically required for building a revenue prediction model?

For B2B, a common and useful conversion to predict the value of is “lead.” If you want to assign a predictive value to a whitepaper download or call booked, you’ll need a few things in place first.

1. Data Warehouse and Platform IDs

You’ll need to store user data—and tie it back to campaign identifiers—in a warehouse or structured environment.

Important considerations:

• You must link marketing identifiers (like gclid, fbclid) to in-house user IDs

• Meta doesn’t require fbclid, but Google does require gclid for conversion value import

This connection allows you to pass lead values back to platforms like Google and Meta and effectively run tROAS bidding strategies.

2. Additional Data Points

Data quality determines model quality. Common variables in conversion value prediction models include:

• Data collected at conversion: Country of signup, device type

• Behavioral data: Visits to pricing or product pages, number of sessions, pages viewed, or time on site

• Firmographic data: Company size, industry, revenue, tech stack

• Product data: High-intent behaviors (e.g. invited colleagues, downloaded assets)

3. Model

You’ll need a model to assign predicted values to leads. This can range from simple rules-based logic (e.g. pricing page visits + firmographic data) to machine learning models trained on past conversions.

If your company has no existing conversion prediction, start simple:

• Use only one data source, like enrichment tools (e.g. Clearbit, GA4 etc)

• Use a rule based model instead of machine learning (e.g bid double on leads based in the US)

• Start with a binary model (likely/unlikely to convert) and iterate

Over time, you can layer in more signals and increase complexity.

⚠️ Common pitfall: Overengineering the first model. Better to be roughly right and usable than precisely wrong and overfit.

Start with what you have. Then refine.

4. Upload process

You’ll need a way to push your conversion value predictions to the Ad Platform. This is usually done through Offline Conversion Upload API. For this you will create a table with the following information which is then pushed via API.

• gclid: Click ID allowing google to associate this conversion to a click

• Conversion time: Time at which the conversion occurs

• Conversion Value: Predicted revenue from your model

Make sure to run QA to verify that the number of conversions you are uploading matches the number of conversions tracked using pixel.

Example: Rolling out your first paid test with a predicted lead value

This isn’t a one-week project, but much of the work can happen in parallel. Here’s a practical rollout plan:

1. Assess if you need it.
   Not all products benefit. If your LTV doesn’t vary much, or your conversions happen within the attribution window, conversion value prediction may not be worth the effort.

2. Ensure your infrastructure can support it.
   Make sure your tracking system and data capture are properly set up to store the necessary data.

3. Link gclid to internal user data (for Google Ads).
   You need to be able to trace users back to the campaign or keyword level to assign value accurately.

4. Build ROAS reporting.
   For later assessing the impact on campaign performance. Start measuring ROAS and CAC early.

5. Create a revenue prediction model.
   Start small. Don’t overcomplicate your first version. Most teams don’t update models frequently, but you can test improved versions quarterly or when new data becomes available.

6. Run a test.
   Set up your platform to receive conversion values. Then compare performance using ROAS and CAC. Analyze how campaigns behave when optimizing for predicted value.

7. Analyze and iterate.
   Refine your model. Check for model drift. Monitor platform optimization quality. This is a long-term investment, not a one-off test.

Implement conversion value prediction with expert help

You need both the model and the marketing strategy. You need someone to wrangle the data—and someone who understands how ad platforms actually behave.

That’s exactly why I (Barbara) team up with David Loris, a marketing data scientist. Together, we’ve helped companies set up their first lead scoring systems and prove their impact with real media spend.

If you’re ready to:

• Define which conversion to predict the value for

• Understand if this is a worthwhile investment

• Run a first scoring test

👉 Schedule a free call with us to talk through it.

Without the correct context, you do, indeed, risk making some questionable business decisions. Just because a report says “Unique Users” or “Direct Concversions”, it doesn’t actually mean you’re getting that.

In this article, we cover misconceptions about what some common metrics represent and how to best leverage marketing data.

All marketing data are estimates

First things first: marketing data is not that maths, it’s analytics. It’s interpretations. It’s unlike finance, where you have an X number of sales of Y value. The numbers you see on marketing reports are not “real”. Afterall, how can any tool or service be sure that a campaign was responsible for convincing someone to purchase a product? Spoiler alert: they can’t.

Obviously, that doesn’t mean you shouldn’t use marketing data. Leveraged wisely, it is a powerful competitive advantage that can help you find and acquire valuable users cheaply.

But to make decisions based on marketing data, you need to be aware of how it’s measured and how it can be wrong. As one of my favorite quotes says:

Metrics you should take with a grain of salt, and why

Google Ads conversions

Would a user have converted anyway (via organic search) if they hadn’t clicked on your paid search impression?

The very nature of Paid Search implies cannibalization. The most obvious example of cannibalization is brand search because most companies rank a top page for their brand search.

The more successful you are at SEO, the bigger a concern paid search cannibalization becomes.

Just because Google Ads says you have 100 conversions, it doesn’t mean that these are incremental conversions. That also doesn’t mean you should put all your spend on brand search.

Actually, make that all ad platform conversions

Although Paid Search is the most common example, the cannibalization effect is not just restricted to Google Ads. Retargeting audiences (running on social, YouTube, programmatic, etc.) all have cannibalization problems. Perhaps these strategies have “snagged” the conversion, but were they really responsible for them in the first place?

The issue with ad platform conversions is not just cannibalization. Sometimes, the problem is the exact opposite.

Certain strategies “underattribute”, due to cross-device journeys, long lookback windows or lack of clicks.

Estimated search volumes from ahrefs, SEMrush, etc

From my experience, these are so often very very wrong. I’ve had multiple clients where search volume for their intent keywords were deemed very low with sometimes literally zero predicted searches per month.

However, when we start running paid search for those keywords, we see there’s indeed existing demand for them. Sometimes even thousands of monthly searches.

The only way you can really know the search volume of a keyword is to bid for it on Paid Search.

Unique users, anywhere but your own

Tools–-from ad platforms to website analytics like Google Analytics-–use different signals to identify if a user is indeed “unique”. Common leveraged parameters are: device identifiers, cookies, browser headings, IPs, location.

But, as you can suspect, none of these things guarantee a user is unique. I have multiple devices. My IP has changed. My cookies have expired. I access websites from different browsers.

The real number of unique users is always less than what’s reported.

Unique users, even your own first-party data

Ever relying on a product identifier, like your very own user_id doesn’t guarantee an actual measurement of unique users. That’s because certain products require users to create multiple accounts. (You can read more about first party data entitles here.)

For example, every time I get a new invitation to a client’s Slack or join a new community, I have to create a new account.

There are also products where you don’t need to create multiple accounts to use, but people still do. Teens have “ristas” and “instas”: several Instagram accounts for unique use cases. Multiple X accounts are also common.

Facebook has previously gotten schtick for counting these “multiple accounts” as multiple users in their shared MAU reports. And we can’t forget Elon’s statement that Twitter’s MAU was overinflated because of bots.

Average time on site

This is a problem of using “averages” in general. Depending on the size of your data set, it can be very easy for a single user to affect your “Average time on site” metric. If they leave the tab open for a long time, historically, that has counted as time on site.

If you want to measure time on site as an engagement metric (fair enough, I say) I would suggest either:

• Using BigQuery to determine sessions and session length

• Creating a “timer event” that fires once the user has spent an X amount of time on site, for example, 15 seconds. I’ve used this a lot as the primary optimization event for brand awareness campaigns

Subscribe now

How to best leverage marketing data

Understand edge cases from each metric

As we’ve seen, there are several use cases where the metric doesn’t really mean what it means. Different metrics have different edge cases.

The best way to leverage marketing data is to understand how each metric is calculated. How are conversions from a certain platform attributed? How are sessions defined? How are users counted?

There’s no short path here, just an ongoing learning process on the “whys” of different data.

This understanding is beneficial not just for data interpretation, but also for identifying when vendors in the data space are overselling what they can deliver. RFPs become a lot smoother when you know what’s being promised is technically impossible.

Track across multiple sources

It’s also helpful to track comparable metrics across different sources. Discrepancies can tell you important stories you were previously unaware of and highlight how essential it is to understand how numbers are calculated.

Test, test, test

Marketing metrics are not very useful if they live in silos. Analytics is better supported with hypotheses and tests.

A/B tests are great, and control groups are even better. For paid media, holdout tests can deliver the best overview of what’s working than any marketing tool can deliver.

Do you trust the data you use to report on campaigns?

If you’re struggling to make sense of what metrics to use or what data sources to rely on, please do reach out! I’m a consultant who helps companies drive marketing results with the help of attribution, analysis, education, and more.

Book a free 15-minute consultation call to see if I can help:

Book free consultation call

If you’re a marketer, data leader, or growth-focused executive wondering how data can improve marketing results—and how I can help you with that—, then this article is for you.

I provide the most value for companies running across many marketing channels

I specialize in working with companies running multiple marketing activities across channels: paid search, YouTube ads, affiliate programs, and organic SEO, to name a few. The common thread? These companies are navigating complexity—different channels, formats, and KPIs—and need clarity on what’s working, what’s not, and why.

For example, comparing YouTube ads to organic SEO isn’t just a case of apples to oranges. It’s more like comparing apples to a fruit salad. You can’t rely on one data source or methodology to measure both effectively. That’s where I come in—to create a measurement framework that makes sense of it all.

With over 15 years in marketing and analytics, I’ve partnered with tech companies like WeTransfer, VEED.io, Ashby, Dentsu, and Microsoft on everything from audits to retainer-based projects.

Improving campaign performance: from better decisions to data activation

In a nutshell, I help companies improve marketing performance through better data. Here’s what that looks like in practice:

Path 1: Smarter Decision-Making

I help companies use their data to allocate budgets more effectively by understanding what’s driving performance. This involves:

• Implementing holistic measurement to identify the best performing channels, like MMM, rule-based MTA, and zero-party data.

• Identifying best-performing campaigns and creatives within channels, with activities like server-side tracking and ROAS reports.

• Surfacing actionable insights instead of just numbers, with activities like testing roadmaps and naming conventions.

• Making data accessible by all relevant stakeholders, with centralized dashboards and improved in-platform reporting.

Path 2: Data Activation for Campaign Optimization

Data isn’t just for reporting—it’s a tool to optimize campaigns. With data activation, we can:

• Improve tracking and measurement of ad platforms, giving them better data signals to identify the most valuable users to reach with activities like lead scoring, lead prediction, and server-side tracking.

• Drive campaign efficiency with better audience segments for targeting, retargeting, and exclusion, with activities like implementing RETL and CDPs.

There is such a thing as “the right time” to work together

Here are some common issues my clients face. If you’re facing these, then it could be the right time to explore a project together:

• Reporting feels unreliable, and you’re not confident in the numbers. Decisions are based on flawed data—or no data at all.

• Campaign performance has declined, and you don’t know why.

• Data sources are misaligned, and discrepancies make analysis impossible or untrustworthy.

• The marketing team has the tools but isn’t leveraging them effectively to improve performance. Or the opposite–they have many plans for how to best leverage data, but aren't able to bring that vision to life.

• You’re unsure if you need an MMM, MMP, CDP, or something else entirely. Vendors are promising you the world, and you’re unsure on what to trust.

• It’s time to scale spend. You’re about to take paid media from 1 to 100 and you must build on a solid data foundation.

The different ways we could work together

An audit is usually the best way to get started

Most of my projects start with a comprehensive marketing analytics audit. Why? Because a solid understanding of your current setup–from both a technical and people perspective–is the foundation for improvement.

During an audit, we cover:

• Stakeholder interviews to understand data sources, reports, and pain points.

• Campaign analysis to assess performance measurement practices.

• QA tracking on both server and client sides.

The result is a detailed roadmap that outlines opportunities, prioritizes actions, and provides clear next steps.

Following these steps, you’ll receive a roadmap of what to tackle, how, and when. Whether fixing tracking, educating your team, or integrating new tools, you’ll have a strategic plan tailored to your needs.

Retainers are ideal if you need either ongoing or custom support

For ongoing support, I offer retainer packages. These allow me to address marketing and data challenges as they arise. Typical activities include:

• Diagnosing and resolving data discrepancies.

• Investigating sudden changes in campaign performance.

• Reviewing RFPs and vendor proposals.

• Implementing recommendations from the audit, like new data integrations or tracking events.

Bringing in the right expert for your problem

If you’ve already uncovered a marketing data strategy you want to double down on, I can help bring that to life. I partner with some of the best marketing data engineers and marketing scientists in marketing to execute on:

• Implementing server-side tracking for ad platforms

• Building a lead scoring model to send as conversion value and improve ad performance

• Rolling out MTA reporting and bidding for search and other click-based campaigns

And, and and–I also produce content like this for data products

In addition to consulting, I create content for companies looking to connect with data-savvy audiences. Clients include dbt Labs, Mixpanel, and Amplitude. I’ve produced case studies, long-form articles, video courses, social media assets and more.

If you’re building a technical product and need compelling content for marketers, analysts, or data leaders, I can help.

Well, ready to get started?

After nearly 10 years of consulting, I follow these steps for new projects:

1. Intro Call: We’ll discuss your challenges and assess if I’m the right fit.

2. Follow-Up Call: We’ll outline the project goals and deliverables.

3. Access Required: I’ll send a list of the stakeholders and data sources I need access to.

4. Kickoff: Contracts are signed, first payments are made, and we get started.

If you’re interested in learning more, then book a free consultation call:

Let's chat data!

If you want to sign up for future classes, head to our website and join the waitlist. Our next free cohort (on March 27th) is already sold out, but you can purchase the 4-hour on-demand course here.

Why attribution matters now (and ever)

Modern marketing teams face an unprecedented challenge: while channels and touchpoints multiply, our ability to track them traditionally diminishes. Privacy changes are breaking conventional attribution models just as marketing teams need better insights than ever to justify and optimize their spend.

In the course, we taught marketing and leaders how to:

• Build attribution frameworks that work in a privacy-first world

• Measure the true impact of both brand and performance campaigns

• Make confident budget decisions with incomplete data

• Build stakeholder buy-in for sophisticated measurement approaches

• Implement practical attribution solutions that drive results

Attribution fundamentals and the risks of broken attribution

We started by estabilishing why traditional attribution approaches are breaking down. This foundational knowledge is particularly crucial for teams investing heavily in both brand and performance marketing, as misattribution can lead to drastically undervaluing brand initiatives while over-crediting bottom-funnel campaigns.

As one participant noted: "We're spending six figures monthly on marketing, but can't confidently say which channels drive results."

We explored several critical challenges:

• Privacy changes (ATT, GDPR) limiting tracking capabilities

• Non-clickable touchpoints like podcasts being hard to measure

• Cross-device journeys breaking conventional models (specially for social)

• Different stakeholders interpreting data differently

• No single tool or model fully capturing marketing's impact

Key Learning: The Multi-Method Framework

Rather than seeking one perfect attribution model, we learned why successful teams need multiple complementary approaches:

1. Click-Based Attribution

• Best for: Direct response campaigns, search marketing

• Limitations: Misses upper-funnel impact

2. Marketing Mix Modeling (MMM)

• Best for: Channel-level budget decisions

• Requirements: Significant spend ($100K+/month)

3. Incrementality Testing

• Best for: Measuring incremental sessions or conversions from strategies

• Methods: Geographic holdouts, A/B tests

4. Zero-Party Data

• Best for: Capturing the user-perceived impact of campaigns that don’t generate clicks

• Methods: Post-purchase surveys, sales calls, lead forms

Multi-Touch Attribution (MTA) and why it’s best for click strategies

We examined the practical implementation of MTA, using real examples to understand its strengths and limitations. While MTA isn't the complete solution many vendors promise, it remains valuable for optimizing digital campaigns when implemented correctly—particularly for businesses with primarily online customer journeys and significant digital ad spend.

Case Study: The Modern Customer Journey

We analyzed this common scenario: A user views a YouTube ad, clicks a Meta prospecting ad, performs an organic search, and converts through a branded search ad. Through this example, participants learned:

• Why last-click attribution misses critical touchpoints

• How mobile app interactions break tracking

• Why view-through conversions matter

• The impact of cross-device journeys

• How cookie limitations affect data collection

Implementing and rolling out MTA

Participants learned practical implementation steps:

• Setting up proper UTM parameters

• Implementing tracking across platforms

• Handling cross-device attribution

• Working with privacy restrictions

Advanced measurement techniques, like MMM and incrementality

Sophisticated measurement approaches, like Marketing Mix Modeling and Incrementality Testing, are gaining traction. These advanced techniques become essential when dealing with significant marketing budgets ($100K+/month) across multiple channels, particularly when traditional tracking methods can't capture the full customer journey due to privacy restrictions or offline touchpoints.

Marketing Mix Modeling

We covered:

• Data requirements (12+ months historical data)

• Budget considerations

• Implementation steps

• Common pitfalls

• When to use MMM vs. other methods

Incrementality Testing

Participants learned:

• Different testing methodologies

• Setting up holdout tests

• Measuring true incremental impact

• Statistical significance requirements

• Real-world implementation challenges

• Session x Conversion Incrementality

Building your attribution roadmap, getting started with an MVP

We provided a practical framework for implementing attribution in phases. This was particularly valuable for teams needing to build stakeholder buy-in and create a clear path to better attribution, showing how to start with quick wins while building toward more sophisticated measurement. After all, for a successful attribution strategy; marketing, data and engineering teams must work together.

Our recommendation was to start with implementing best practices and rolling out MVPs alternative forms of measurement to get buy-in, such as:

• Implement proper UTM parameters

• Set up basic analytics tracking

• Start collecting zero-party data

Common Pitfalls We Addressed

Throughout the course, we helped participants avoid common mistakes:

1. Over-reliance on Last Click

• Impact: Undervaluing awareness channels

• Solution: We demonstrated multiple attribution methods

2. Tool Obsession

• Impact: Focusing on tools over strategy

• Solution: Participants learned to start with business questions

3. Seeking Perfect Attribution

• Impact: Analysis paralysis

• Solution: Understanding model limitations

4. Ignoring Incrementality

• Impact: Misattributing natural demand

• Solution: Practical incrementality testing approaches

Don’t forget to join our upcoming cohort or purchase a recording

Our next free live Attribution Masterclass is this month, and it’s already sold out. However, you can always enter the waitlist for future cohorts on the site or purchase the 4-hour on demand course here.

Register at Attribution Masterclass

This article covers what data sets and views you need to build a ROAS report at your organization.

CPAs vs. ROAS: The Full Picture

CPAs (Cost Per Acquisition) help you measure how much you’re paying for users to perform a specific action, like subscribing to your product. They pinpoint which strategies within your paid campaigns are bringing in the most new subscribers at the lowest cost. However, CPAs don’t tell the whole story.

User behavior post-subscription can vary significantly depending on the acquisition channel. Even within the same channel, the behavior can differ widly. For example, in paid search, we can see the difference between user acquired from brand terms versus intent terms.

CPAs alone miss this context because certain audiences have higher or lower:

• User retention

• Seats per account

• Cancellation rates

• Upselling

Without an overview of how much revenue each new account brings (ROAS), you risk misallocating your budget. Strategies with high CAC might also have higher ARPU, making them more cost-effective than other low CAC initiatives.

Here’s a common scenario:

Disclaimer: Measuring ROAS doesn’t mean measuring incrementality isn’t necessary.

Why Conversion Events Are Not Enough for Subscription Products

While CPAs rely on conversion events for measurement, subscription products complicate things with multiple revenue-generating actions. For example, renewals, changing plans, inviting users to your plan, and credit usage can all impact ROAS.

It would be nice if we could set all these user actions as conversion goals in our ad platforms, pass a monetary value for each of them (either predicted or real), and use this to calculate ROAS.

But, alas, attribution gets in the way.

Each of these events could be attributed to a different campaign instead of the original campaign that brought in the new user. For instance, a renewal conversion event might be attributed to “Organic” or a CRM campaign if you’re using UTMs for those, and not your original Google Ads campaign.

The Solution: Measuring Ad Spend, Conversions, and Revenue Separately

The solution is to divide the problem into three data views and later join them to create a ROAS report. We’ll cover the four steps to get this in place.

Step 1: Store Your Ad Platform Spend

A prerequisite to calculating ROAS is storing spend data. Duh. This can be done using an ETL tool (like Fivetran or Funnel) and sending the data to a Data Warehouse. BigQuery has out-of-the-box integrations for Google and Meta ads.

You should be able to create a view where you store spend from all your ad platforms, per day and per campaign:

Disclaimer: In order to simplify this article, we’ll explain how to report on ROAS at the campaign level. However, it’s recommended to go deeper: ad-level (for Social) and ad group and keyword level (for Search).

Step 2: Calculate and Store Revenue Data Per User or Account

Subscription products’ revenue comprises various events, such as the act of subscribing, monthly renewals, adding new seats, and changing plans. These events are typically tracked on your payment infrastructure (such as Stripe) and/or your product analytics tool.

To calculate your ROAS, you need an overview of all revenue an account or user has generated. For simplicity, assume you have a single-user product with no seat-based pricing.

Create a view that sums all revenue from different events throughout each user’s lifetime:

Disclaimer: If you want to have set lookback windows or calculate payback periods, you also need to include Date in this view.

This data should be stored in the same data warehouse as your ad spend.

Step 3: Track the Initial Conversion Event

The third table to go into your data warehouse is your attributed conversions.

In your product analytics tool of choice (Google Analytics, Amplitude, Mixpanel, Snowplow, etc.), track your key conversion event. Usually, this is the conversion event highest up the funnel.

The tracking of this conversion event must comply with the following requirements:

• The conversion event must be attributed to a single campaign.

• The conversion event must pass a unique user value also present in your revenue table, such as user_id.

• The campaign must be a unique identifier in the form of a non-changeable ID, also present in your ad platform reporting. In this case, we’re using a campaign_id.

Create a view like this:

For a lead event, the unique identifier is usually an email address. This means you need to include the user’s email address in the revenue table.

Subscribe now

Step 4: Join and Report on the Data Sets

Once you have all three tables stored in your data warehouse, it’s time to join the data sets. They will be joined on different keys:

1. Ad Spend: Joined by campaign ID.

2. Revenue: Joined by user ID.

3. Conversion Event: Joined by campaign ID and user ID.

Once you’ve created this holistic view of ad spend, conversions, and revenue, you’ll be able to report on ROAS for your subscription product’s campaigns:

Depending on your naming convention, you can further segment your campaign names into separate columns for “Market,” “Strategy,” “Audience,” etc., to find the most cost-efficient segments to allocate your budget to.

Drive improved results with ROAS

Calculating ROAS involves more than just measuring CPAs; it requires separate views of ad spend, conversions, and revenue. But, by moving beyond CPAs and incorporating ROAS, you ensure that every marketing dollar is spent wisely, optimizing not just acquisition but also long-term profitability.

When I built a ROAS report for VEED.io, we were able to increase overall subscriptions (with a flat CAC), improve impression share, and acquire higher-value users. This was possible thanks to smarter budget allocation.

If you’re interested in calculating ROAS for your paid media initiatives, then head to my website and reach out.

Similarly, marketing measurement can become a business advantage. When you measure campaigns that are hard to attribute, you identify strategies that your competitors overlook. This is where you find the most cost-effective CPAs.

But how do you get there?

This guide outlines eight marketing measurement strategies to understand campaign performance better and answer difficult budget allocation questions.

First of all, not every $1M is the same

In this guide, I assume that if you’ve hit $1 million annual spend, you’re spreading your eggs between many baskets. At this point, you might have YouTube, Google Ads, and Meta running concurrently.

If your $1M is allocated to a single channel—a common example being Paid Search—you don’t need all that much. MTA (multi-touch attribution) could be all you need.

In fact, I find important to highlight that having all budget allocated in a single strategy is great if that strategy has a positive ROAS and a comfortable payback period. The wider the channel mix, the more complex it is to manage campaigns and measurement. So, do “squeeze all the juice” of your working strategies before expanding.

Now, let’s get to those best practices.

The 8 Measurement Best Practices To Implement

1. UTM and Ad Platform Naming Conventions

Even in 2025, click-based attribution is a crucial part of all marketing measurement. I don’t need to go into the importance of UTM naming conventions. Dynamic UTM values—from Meta’s to Linkedin—have made implementing and maintaining consistent UTM structures very, very easy. So please use and abuse them. Lukas Oldenburg has a great article on designing UTM naming conventions if you want guidance.

However, many teams focus solely on UTM naming conventions. In fact, effective naming conventions must work seamlessly across advertising platforms, analytics tools, and CRM systems. Your naming framework should enable data two things:

• Data consolidation: joining data between platforms, e.g. Meta and Mixpanel.

• Granularity: segmenting insights meaningful analysis.

I’ve written in-depth on how naming conventions enable this.

Poor naming conventions can prevent joining first-party data with Google Analytics data, severely limiting your ability to understand customer journeys and make data-driven decisions. This impact compounds as marketing operations scale and more channels are added to the mix.

Market information, audience segmentation, and creative elements all need consistent representation across platforms. You’re unlikely to make decisions based on CTRs reported on Day of the Week. But you could learn valuable audience information on CTR per Creative Messaging. Naming conventions will enable you to segment reporting at the level you require.

2. The Critical Importance of Early Data Transfer Setup

The GA4 UI interface has significant limitations that many teams don't realize until it's too late. Due to cardinality constraints, you cannot report things very granularly through the UI. Even concepts like user-level attribution (super important for subscription products!), where you report all revenue dependent on the campaign that brought in the user, are impossible without querying.

GA4's BigQuery integration only stores data from when you set up the integration. If you set up the integration in month six, you've permanently lost granular data from months one through five. The data simply cannot be recovered retroactively.

Google has made the BigQuery integration process smooth because it's one of their main monetization strategies for Google Analytics today after deprecating Universal Analytics. Despite the easy setup, many companies still postpone this critical integration, not realizing the permanent loss of granular historical data they're incurring.

This is also relevant if you’re not using GA4. Queryable data from your CDP (e.g. Segment or Rudderstack) or event analytics (e.g. Mixpanel and Amplitude) can also be crucial if one of these tools is your product’s data source of truth.

3. Build and Maintain a Testing Roadmap

A testing roadmap should help you achieve three things simultaneously: 1) maintaining a high testing cadence, 2) ensuring actionable next steps, and 3) making insights easily understandable across the organization.

Creating a roadmap can be as simple as a well-structured spreadsheet in Google Sheets or a complex layout in Miro, Notion, or Jira. The key is choosing a tool that's easily accessible by the whole team—pick something that’s already used in their day-to-day work.

Your testing roadmap should function as a structured database of various tests, organized by team or section—whether that's product marketing, growth marketing, lifecycle, paid media, or onboarding. It keeps everyone aligned, maintains testing cadence, and helps prevent losing valuable insights.

Documentation becomes crucial in maintaining a comprehensive record of your tests. Think of it as leaving a trail of breadcrumbs that can guide future marketing strategies. Even tests from years ago can provide value, so you don't want to lose them through employee turnover.

Every test requires several key elements: test name, hypothesis, methodology, metrics, and timeline. Running tests without defining your hypothesis and methodology beforehand is strongly discouraged—you risk being unable to assess results due to incorrect setup or missing necessary metrics. This article is a handy place to start thinking about testing roadmaps.

For example, quarterly incrementality tests on YouTube and Meta prospecting campaigns help teams optimize budget allocation while accounting for seasonal variations. These regular cadence tests build a valuable repository of knowledge about channel performance over time. We’ll go deeper into this on the next section.

4. Measure True CACs Through Incrementality Testing

Incrementality refers to the additional outcomes—be it sessions, conversions, or other metrics—that result from a specific marketing strategy. It represents the difference between the outcomes with a particular marketing activity versus the outcomes without it.

Most companies start their incrementality journey with Paid Brand Search. The hypothesis is straightforward: would someone who saw (or clicked on) a brand search ad have converted anyway? The methodology involves turning off Paid Brand Search campaigns for a set period and comparing results with Organic search data (with Google Search Console).

Beyond Brand Search, prospecting initiatives like YouTube and Meta campaigns should undergo regular incrementality testing. One e-commerce client discovered their Meta prospecting campaigns were driving 40% more revenue than what click-based attribution suggested. When they reallocated that budget, the impact on overall sales was significant.

5. Channel-Specific Measurement Approaches Instead of a “One Source of Truth”

Meta's conversion tracking provides crucial signals for algorithm optimization, while Google Analytics offers a deduplicated view of cross-channel customer journeys. That’s why their conversion numbers don’t match. Neither source represents absolute truth. Both contribute to understanding marketing performance.

Different channels serve different purposes in the customer journey and require tailored measurement approaches. This doesn't mean working in silos, but rather understanding how various data sources complement each other to build a complete picture of marketing performance. The answer here is to educate stakeholders so they understand the value of a “triangulated” approach to attribution.

6. Server-Side Tracking for Top Funnel or Cross-Device Initiatives

Server-side tracking is considerably more complex to set up and maintain than client-side tracking. My rule of thumb is to avoid data complexity until it’s absolutely necessary. However, there are use cases where implementing server-side tracking can change campaign performance from night to day.

Meta CAPI, in particular, is the prime example here. The cross-device journey that happens on Meta ads is simply not measured effectively with click-based tags. By exporting PII (like emails and names) from converters from your backend to Meta, you can increase the number of Meta-attributed conversions. Meta can use this precious data signal to identify who to target with your ads and improve your campaign performance.

Whether this is worth to implement for your brand will depend on how much you’re spending on Meta or other top of funnel initiative that can be improved with server-side measurement.

7. Implement Discrepancy Monitoring to Identify Errors Proactively

The last thing you want is to waste your analysts’ time chasing down rabbit holes to find out why different data sources are reporting different numbers. Unfortunately, there's no universal threshold for acceptable discrepancies.

I've seen Meta and Google Analytics conversions with a 90% discrepancy, and everything was working fine! In other cases, a 10% discrepancy indicated serious tracking issues. This variability makes it crucial to monitor trends rather than focus on absolute values.

Regular monitoring helps establish what "normal" looks like for your specific implementation. Once you understand your baseline, you can more easily identify when something requires investigation.

One client saved significant ad spend when their monitoring system detected a discrepancy between their brand safety tool (IAS) and ad server (Campaign Manager), revealing geo-targeting setup errors in their programmatic campaigns. This discovery prevented continued waste of advertising budget and highlighted the importance of systematic monitoring.

In most cases, this will require a Data Warehouse, but it can also be done with Modern Spreadsheets. Evaluate the right time to invest in this based on your spend level and data quality.

8. Leverage Marketing Mix Modeling (MMM) for Confident Budget Allocation

At $1M+ annual spend across many channels, Marketing Mix Modeling (MMM) can help you understand true marketing impact. MMM's primary value lies in analyzing strategies that perform poorly under click-based attribution, like TV or YouTube campaigns. Traditional click-based measurement often fails to capture their true impact— not because these channels don't work, but because the measurement approach doesn't match how they influence customer behavior.

But not every strategy that is a fit for MMM. You need certain levels of volume and variation. If you’re wondering when to use MMM or click-based attribution, this article can shed some light.

Building Your Measurement Foundation

Proper measurement infrastructure isn't built overnight, but each element described above forms a crucial part of a comprehensive measurement strategy. Start with the fundamentals:

• Implement consistent naming conventions across platforms

• Set up your GA4 or CDP data warehouse integrations

• Create and maintain a structured testing roadmap

• Begin incrementality testing. Paid Brand Search is a good start

• Monitor discrepancies systematically if you’re suffering from poor data quality

• Review whether server-side tracking can drive the performance of your campaigns

• Consider MMM for budget allocation as you scale

Remember that poor measurement doesn't just affect reporting–-it impacts every decision your marketing team makes. It means leaving money on the table. And the cost of inadequate measurement compounds as your marketing spend grows.

But what do you really need?

What should you start with? What do you prioritize? With marketing, unfortunately, the answer is always: it depends. It depends on your spend level, your channel mix, your sales journey, your skillset, your data stack and more.

I’m a marketing measurement consultant who can help you identify the biggest data opportunities for your marketing campaigns. I usually start a project like this with a Marketing Analytics audit to identify what your team and your campaigns need.

If you’re interested in that, do reach out:

Book a free consultation call

This guide shows marketers how to accurately measure affiliate performance across content partnerships, coupon sites, and loyalty programs by combining Marketing Mix Modeling (MMM), engagement metrics, and multi-touch attribution.

Written with the help of affiliate experts Mike Currey and Lacie Thompson

The origin story of this post was an earlier project of mine for a leading B2C brand. Much like everyone else, they migrated from Universal Analytics to Google Analytics 4 and suddenly reported a very different number of conversions for their Affiliate campaigns. Only 40% of conversions reported on UA were attributed to Affiliate on GA4.

I shared this story on Linkedin. There, I chatted about how Affiliate campaigns should be measured with Mike and Lacie. With many decades of Affiliate experience between them, they helped me uncover common gaps marketers make when measuring the strategy:

Mike Currey: 20+ years in digital marketing, 10+ in Affiliate alone. He ran T-Mobile, Blue Nile, and Quince’s Affiliate programs.

Lacie Thompson: Founder of LT Partners, an Affiliate marketing agency acquired by New Engen.

The peculiarities of Affiliate campaigns and how it differs from paid media

What we’re covering as “Affiliate” in the article

In this article, we’re referring to “Affiliate” campaigns in which publishers or individuals (“partners”) are compensated based on measurable results. These campaigns usually have a dedicated tracking link per partner, enabling conversions to be attributed to each individual.

Affiliate is a wide-reaching and varied initiative. Some examples are:

• Blogs or influencers (known as “Content”)

• Coupon and deal sites

• Toolbar add-ons, like Honey

• Employee benefits partnerships

• Card-linked offers (CLO) from banks and credit cards

• Customer loyalty benefits

Affiliate “Ad platforms” are usually campaign managers’ source of truth

Paid media campaigns are usually set up via ad platforms. You create campaigns and bid for clicks on Google Ads. Affiliate campaigns, similarly, are created and maintained on Affiliate platforms.

These platforms are shared by advertisers and publishers alike. Both parties use them to visualize campaign results. Since this data usually defines “payouts” (how much compensation each partner should receive), they’re commonly used as Affiliate campaigns’ “source of truth.”

There are many affiliate platforms, some of the most established being Impact, Rakuten and ShareASale.

Differences between Affiliate reporting and traditional paid media reporting

For paid media reporting, measuring conversions and other metrics that indicate quality traffic (e.g., time on site) in the ad platform helps campaigns self-optimize. These data signals help Google, Meta, et al. understand which users you want more of.

The compensation for each partner is defined by which KPIs were chosen and how they’re attributed.

In Affiliate campaigns, advertisers measure metrics in the affiliate platform to calculate how much each partner should receive. The compensation for each partner is therefore defined by how metrics are chosen and how they’re attributed. Ensuring partners are compensated fairly enables them to continue to provide value.

Best practices for measuring affiliate performance

Only reporting on 1) last-click attribution and 2) purchases for Affiliate campaigns is a mistake. While conversions are certainly important, this narrow view can overlook significant contributions made earlier in the customer journey and, most importantly, can overestimate the impact of certain sources. This can lead to undercompensating valuable partners and misinvesting marketing budgets.

Define UTMs to report on results per partner

Affiliate campaigns need reporting like every other campaign type to understand traffic quality and direct-attributed conversions. Therefore, Affiliate links should include UTMs to enable reporting on Google Analytics, Snowplow, Mixpanel and similar tools.

Usually, Affiliate naming conventions include Partner IDs and Content Type. This granularity is necessary to define unique KPIs for each Content Type. In most cases, advertisers cannot measure Affiliate campaign results per creative.

Tackling last click attribution’s limitations with MTA (Multi-Touch Attribution)

Some Affiliates, such as browser add-ons like Honey, only show up at the tail end of a transaction. But are these conversions incremental? Last-click attribution would credit Honey for the conversion, but this ignores the initial touchpoints, like Content affiliates (e.g., blogs or influencers), that introduced the user to the brand.

A multi-touch attribution (MTA) model that accounts for various interactions across the journey can help here. You can assign different weights to each Affiliate’s contribution, ensuring that credit is distributed more equitably. For example, you might reduce the weight of tools like Honey that appear primarily during the purchase phase and increase the credit for affiliates involved in prospecting (e.g., content creators).

Leveraging engagement metrics for partners that are driving awareness

Not all Affiliate strategies should be measured or compensated solely on conversions. In some cases, traffic quality or new user acquisition might be more relevant indicators of performance, especially for top-of-funnel Affiliates like blogs or influencers.

• Compensation for prospecting: Content Affiliates can be evaluated on the percentage of new visitors they generate. A better model for these partners might involve compensating them per new user website visit, rather than just for last-click conversions. It’s important to remember that if you don’t compensate partners fairly, you risk losing them from your plan.

• Diversifying KPIs: For deal or coupon sites, metrics like conversion rate are critical, but for content partners, it’s important to track engagement metrics like time on site, number of pages visited, key pages visited (e.g. pricing) or bounce rate to assess the quality of traffic they bring.

MMM is valuable, but must be structured differently

Marketing Mix Modeling (MMM) is a useful tool for understanding the performance of top-of-funnel activities. It helps brands measure strategies that don’t generate clicks (e.g. OOH, video) and with long-term impact (when conversions occur outside of the lookback window). But, unlike paid media, where daily impressions and spending are input in the model, Affiliates only have “spend” when conversions occur. This shifts how MMM should be structured for affiliate campaigns:

• Adapting MMM Inputs: Traditional inputs like daily impressions aren’t available, and spend is contingent on performance. Instead, Affiliate MMM data input will rely primarily on click date.

Lacie worked on a recent project where a client’s MMM analysis reported a $28 ROAS for Affiliate. However, on GA4’s last-click model, the strategy only received a $6 ROAS. The client doubled down on Affiliate and saw an increase in sales corresponding to a $20 ROAS. This is a good example of how MMM can help identify opportunities in marketing campaigns that under-report on click-based attribution, like Affiliate.

Post-consumer surveys don’t work as well

Zero-party data like “How Did You Hear About Us” (HDYHAU) is a form of attribution that can be very helpful in understanding strategies that don’t measure well with click-based tracking. On paper, that’d include Affiliates, but that’s not the case. Affiliates are so embedded in the customer's life (employee benefits, loyalty programs, discounts), users don’t even realize they’re being served that content.

Beware: the nature of compensation leads to a fraud risk in Affiliates

Due to the compensation structure in Affiliate marketing, there’s a higher risk of fraudulent activities such as click farming or bot traffic inflating results. Here are steps to mitigate these risks:

• Click Quality Monitoring: If Affiliates are being rewarded for clicks, it’s essential to audit click quality. Affiliate tools (like Impact or CJ Affiliate) provide visibility into suspicious activity, allowing you to detect unusual spikes in clicks that don’t correlate with quality traffic or conversions.

• Voucher and Coupon Site Leaks: Coupon leaks, where discount codes are shared on unauthorized platforms, can lead to unintentional attribution and revenue loss. Ensuring that each voucher is assigned a specific partner helps mitigate this faster. Mike suggests including the partner's name in the voucher to disincentivize others to share it.

• Compliance Tools: Tools like BrandVerity and SearchMonitor also provide a more comprehensive array of “protection” features to avoid spending on “stolen” clicks.

• Expiration dates: Expiring vouchers after a set number of days can also mitigate risks of code sharing. For example, vouchers that are no longer valid for 7 days after their creation.

Most measurement best practices fit Affiliates, but customization is needed

Just like any other marketing strategy, it’s crucial to ensure that the way you measure Affiliate campaigns aligns with the specific goals of each campaign. Misaligning KPIs or relying too heavily on last-click attribution could lead to misinvesting budget or even cutting funds for strategies that are essential for driving new users to your brand. For example, content affiliates—who are often responsible for bringing in new prospects—might not get the credit they deserve if you’re only looking at conversion metrics.

If you want to audit or improve your marketing measurement, please reach out to me! If you’re looking for expert Affiliate help, Lacie’s agency is a great place to start your search.

And don’t forget to subscribe to 021 Newsletter. We cover everything from advanced attribution models to actionable insights that help you optimize your marketing efforts and budget.

Valuable marketing analysis leads to insights that can directly translate into campaign changes. For a recent project with VEED, the company wanted to scale thir 6-figure monthly Paid Search 20% month-over-month without sacrificing efficiency. 

Our analysis led to significant tests on Paid Search: the new model was tested in one of the most important campaigns, which resulted in an increase in subscriptions (20% increase with a flat CAC), impression share (10% increase), and ARPU (12% increase). In this article, I cover how we achieved that with:

• Dashboard and models that calculate ROAS across different granularity levels

• Changes to campaign taxonomy, switching from a traditional market-level to use-case level

The challenge of scaling spend, with a 20% MoM target

VEED is a market-leading video editing tool with over 4 million users worldwide. Search has been a backbone in their growth—VEED.io has 156,000+ backlinks. The organisation was spending a six-figure sum on Google Ads across Paid Search and PerformanceMax and was looking into increasing that spend 20% month over month.

But to scale efficiently, the organisation needed to expand its budget while maintaining CAC—a familiar yet tricky challenge. To do that, VEED needed to identify which strategies to double down on and which to remove from the plan. That’s where I came in.

VEED’s bidding strategy and campaign taxonomy

Target CPA for “subscription” and market-level taxonomy

I always start projects with an audit. This one was no different.

VEED was running target CPA strategies. The conversion event fired when a user successfully subscribed for the first time. They were using Enhanced Conversions and had a high consent rate. Almost all subscriptions were occurring within the lookback period.

Their campaign taxonomy was conventional, with “market” defined at the highest level (Campaign) and “use cases" at Ad Group level. Most companies use this structure to control the advertising budget for each country.

Gap in existing structure: homogenous treatment of users and use cases 

I identified a possible gap that could harm their campaign efficiency: the CAC bidding structure treated all converted users as if they had the same ROI. 

But since VEED is a subscription product, it was likely that not all users were worth the same to the business. Some would have higher churn or retention or even multiple seats (for B2B users), leading to a diversity in ARPU. Therefore, our analysis needed to go beyond CAC and identify the most valuable users.

A possible variable that could affect ARPU was market: for example, users in the United States being more valuable than users from Brazil. Another likely impacting variable was the product use case. For example, certain editing features were mostly sought out by businesses and some features are only used sparingly. 

Measuring ROAS on a keyword-level

The analysis VEED needed

Based on these gaps, I identified that we had two challenges to tackle with our analysis:

1. Move from CAC reporting to ROAS reporting. Instead of reporting on Google Ads conversions alone, VEED’s first-party payment data needed to be integrated to identify which attributed subscribers had repeated payments or multiple seats.

2. Calculate ROAS across different levels. Data needed to be connected and segmented in order for us to identify the most valuable use cases, keywords, and markets. For this, we needed to leverage Ad Platform naming conventions.

Building the ROAS report

The solution put forward was to build a report that calculates ROAS on different levels: Campaign, Ad Group and Keyword. This report consisted of three separate data sets from VEED:

• Ad Platform Spend, where we used Google Ads Reporting on BigQuery.

• Attributed events, VEED’s custom attribution model table that attributes new users to a utm_source, utm_campaign and utm_term.

• Revenue, VEED’s table that stores recurring payments on a user-level.

We built two different reports on Metabase, so it could be easily accessible by leadership:

• Overall ROAS: Calculates the total return on ad spend for each campaign, market, ad group, use case and keyword.

• ROAS in X days (payback period variation): Calculates ROAS in a custom lookback period. For example, how much revenue was recouped in the first 30 days of the campaign?

To gain insights on a use case level, we leveraged the existing Ad Platform naming convention. This enabled us to report on use cases as a whole (e.g., Add Subtitles) regardless of which campaign (or market) they belonged to.

Disclaimer: this report was only possible for Paid Search, since Performance Max doesn’t enable you to measure results on a keyword-level.

Increasing spend by 20% while flatlining CAC

Identifying the most valuable strategies

After building the report, I analysed the results and prepared recommendations for leadership. The biggest uncovered insight was regarding use cases. Certain features were very sought after, but did not lead to high value retaining users. On the other hand, certain features were likely to lead to attributed subscribers that had significantly higher ARPU. 

With this information, VEED could identify which strategies to stop and which to receive further budget.

Shifting towards a use case budget allocation

Based on this insight, the agency prepared a test where the Campaign focused on a single use case (feature). Within the Campaign, each Ad Group targeted a market. The initial result was positive, so the new taxonomy was scaled across multiple use cases.

Since VEED now knew which use cases were the most valuable, they could max out spend in these keywords before allocating budget to lower quality search terms.

Increasing scale while maintaining efficiency

The granular insights on ROAS, accompanied by changes in the campaign structure, enabled the company to:

• Grow the overall number of attributed subscriptions by 20% while maintaining a flat CAC in the tested campaigns

  • Increase impression share for their Generic terms by 10%

  • Acquire users with a 12% higher ARPU

Moving forward with ARPU prediction as offline conversions

VEED is continuing data projects that will enable them to optimise campaigns for tROAS over tCPA. Because recurring payments occur outside the maximum 30-day lookback period of Google Ads conversions, the company is looking into leveraging lead scoring to predict ARPU. In this future set up, campaigns would optimise towards an offline event that fires from the API instead of the tag-based subscription event.

Upcoming guide on how to measure Google Ads ROAS for subscription products

In this article, I covered the business impact the ROAS dashboard had on VEED.io’s Paid Search performance. My next article will be going deeper into how to build your own ROAS dashboard, including:

• Naming convention requirements

• Data sets structure

• Examples of ROAS reports

If you’re interested in learning how the sauce is made, don’t forget to subscribe (for free!) and be notified as soon as the guide is live.

Subscribe now

If you would like a granular ROAS report built for your organisation, reach out and let’s chat ✨projects.