The only way to know what your ads really did | MarTech

Every performance marketer knows this scenario. The marketing platform, whether it’s Google Ads, Microsoft Ads, Meta Ads, or TikTok Ads, shows 5xROAS.

You don’t believe it, so you pull the number from your backend, and it comes back 2x. Backend data must be true, right? You mark the campaign down as mediocre and move on.

The marketing platform’s 5x is inflated. The problem is, your backend’s 2x isn’t the truth either.

Marketing platforms count generously by design.

View-through conversions, modeled conversions where consent was never granted, and a conversion window that credits a click from three weeks ago all flow into the number it reports to you, and they all point the same way — up.

Marketing platforms grade their own homework and set your spend on the result.

The backend does the opposite, and it does it quietly. Most backend revenue reporting is last-click or something close to it. The order gets attributed to whatever the customer touched last, often a brand search or a direct visit, and the paid click that started the whole thing three weeks earlier gets nothing.

Your CRM isn’t lying to you. It’s answering a narrower question than you think you asked.

Same flaw, pointing the other way

Both errors come from the same place: the moment a conversion gets pinned to a single touch. The marketing platform resolves that ambiguity in its own favor and books the assist as a win. The backend resolves it in favor of the last click and books the assist as nothing. Same ambiguity, opposite defaults.

The gap between 5x and 2x is mostly due to the size of that disagreement, not to fraud.

Picture two thermometers on the same patient. One runs five degrees hot, one runs five degrees cold. Averaging them doesn’t give you the fever. It gives you a number that belongs to neither thermometer and describes nobody. That’s what reconciling a marketing platform figure against a backend figure actually produces.

If the backend systematically strips credit from the touchpoint that opened the journey, then whatever did the opening gets underpaid, and the error runs in one consistent direction: away from the top of the funnel. It doesn’t fall evenly, though. Where it falls hardest isn’t where you’d guess.

The backend can’t see the ads nobody clicks

The under-crediting isn’t spread evenly. It scales with distance from the click, and that determines which channels the backend quietly robs.

Impression-based formats are where it goes nearly blind. Social, display, video, and connected TV work by influence, not clicks. Someone scrolls past your Meta ad, doesn’t click, then searches your brand three days later and buys.

Your backend credits the last clickable touch, brand search, or direct, with the sale. The impression that created the demand gets nothing because there was no click for a last-click system to record.

To your CRM, aMeta campaign that drove a week of branded searchlooks like it did nothing at all. Not undervalued. Invisible.

It got worse after iOS 14. Even the social clicks that do happen lost their match back to the purchase, so the one signal the backend might have caught thinned out, too. The channels built on impressions are the ones a last-click backend is structurally least able to see, and they’re exactly the channels teams are now told to justify against backend revenue.

Search is the exception, and I’ll be straight about why that matters coming from me. Search lives on a click, and the click usually sits close to the purchase, so a last-click backend captures a fair share of what search actually did. It still shorts the upper-funnel work, the generic and research queries that seed a branded search later, but the gap is narrower and far less invisible than social’s. Search is the channel the backend treats least unfairly.

This is the uncomfortable part. I sell search. The format the backend robs blind isn’t mine, it’s social, the one I have the least commercial reason to defend.

So when I tell you the backend is systematically wrong about what your ads did, it costs me nothing to say search is fine, and everything to admit the channel being destroyed by last-click logic is the one I don’t run.

The damage isn’t abstract. Judge your impression-based campaigns by last-click backend revenue, and you’ll switch off the demand generation that quietly feeds every clickable touch downstream, branded search included.

Cut what doesn’t convert becomes cut what you can’t measure. The backend doesn’t merely undervalue the top of your funnel. It can’t see it, and acting on that blindness is how working awareness spend gets killed.

When two platforms claim the same sale

Everything to this point assumed one channel. Add a second, and the over-reporting stops being something you can argue about and becomes arithmetic that can’t be true.

Run Google and Meta together, pull each marketing platform’s reported conversion revenue for the same period, add them up, and set the total against what your backend says you actually made. For most accounts running at scale, the sum exceeds the actual number. Sometimes much higher.

That isn’t two platforms being a little generous. It’s the same sale booked twice.

A customer sees a Meta ad, searches your brand later, clicks the Google ad, and buys. Google logs the conversion. Meta logs it on a view-through basis, too, because it served an impression within its window.

Neither can see the other, so neither discounts a cent. You paid once, you sold once, and two dashboards recorded the revenue.

Meta’s view-through is the largest single source of the overage, and it pays to name who benefits from it existing: Meta, whose reported performance depends on claiming conversions it influenced at most and often didn’t touch at all. But Google runs its own version, counting engaged views and modeled conversions, which boost the number. This isn’t a social-versus-search point.

Add a third channel, and it compounds even more. Reconciliation can’t rescue this. You’re no longer lining up two or more numbers; you’re trying to split one sale among parties who have each already claimed the whole of it.

Reconciling gives you a third wrong number

The standard fix is to reconcile. Blend the sources, or move everyone onto data-driven attribution (DDA) and let the model split the credit.

DDA presents itself as the answer to exactly this problem, but it still comes from the same marketing platform whose top-line number you already learned not to trust. Reallocating credit among paid touches still can’t show you the conversions that would’ve happened without ads. It reconciles. It doesn’t measure.

Whatever number reconciliation hands you, it’s still built from attribution, and attribution answers “which touch gets the credit,” never “would this have happened anyway.” Those are different questions. The reconciled figure is a more expensive guess at the first one.

This is where automated bidding turns the problem worse instead of better.

PointSmart Bidding, or an agent layered on top of it, at the reconciled ROAS, and it optimizes toward it with total confidence, pouring budget into whatever the flawed number rates highly. The mistake a junior analyst makes reading the wrong column, an agent makes faster, at scale, and with more conviction.

Wrong inputs don’t become correct just because the optimizer is sophisticated.

In the EU, the marketing platform side of this gap is structurally wider, not narrower. Consent mode means a real share of conversions are modeled rather than observed, and modeled conversions inflate exactly the top-line figure you’re trying to reconcile against. The German or European practitioner lining up a consent-mode platform number against a last-click backend is averaging two instruments that sit further apart than the U.S. case, not closer.

Dig deeper:Why attribution and impact are no longer the same thing in PPC

Build it or buy it, it’s still attribution

The honest reaction to all this is to go looking for a better attribution setup, and there are two doors: Build your own or buy one of the packaged suites. Both are worth understanding before you spend an engineering quarter or a five-figure subscription on either.

Building your own sounds like the sophisticated move, and the pitch is seductive: collect every touch server-side, resolve identity across sessions and devices, run your own multi-touch model instead of Google’s, own the whole thing.

On a slide, it looks clean. In practice, it’s a standing data-engineering commitment, not a project with an end date. You’re deduplicating conversions across marketing platforms that each count conversions differently, matching users who clear cookies and hop between devices, and rebuilding the pipeline every time a marketing platform changes its API or window.

Most teams underestimate this by an order of magnitude. The ones who pull it off have a data engineer whose actual job it is, not a paid search lead doing it on Fridays.

The backend you’re stitching to isn’t a fixed quantity. Some commerce platforms are far smarter than others. Shopify can show you the customer journey to a sale, the sequence of touches that led there, which is real raw material to work with.

Plenty of CRMs and backends do nothing of the kind. They stamp the last touch and call it revenue. So “check your backend” isn’t uniform advice. What you can even see depends on the commerce platform you’re on, and that alone decides whether building anything on top of it is realistic.

The buy option is the packaged multi-touch attribution suite: Triple Whale, Northbeam, Rockerbox and the rest of the cottage industry that grew up after iOS 14 to rebuild the signal the pixel lost. The better ones now bolt marketing mix modeling and incrementality testing onto the attribution, which is the part actually worth paying for, and the road the next section takes.

On the attribution itself, here’s my take, and it’ll annoy a vendor or two. These suites earn their complexity only at seven to eight figures of revenue, where a dedicated analyst runs them and a single budget call moves enough money to justify the overhead.

Below that, you’re buying a laboratory to settle a question a whiteboard could handle. The precision outruns both your data volume and the stakes of the decision, and a small account would learn more from one clean holdout than from an annual subscription to a dashboard it half configures and never quite trusts.

None of it, built or bought, cheap or dear, changes the category.Multi-touch attribution is still attribution. It answers which touch, with more granularity and better-looking charts, and it never answers whether. The most sophisticated version of the wrong question is still the wrong question, which is why the number that finally settles the argument has to come from somewhere else entirely.

Turn it off to get the truth

There’s one number in this whole discussion that answers “would this have happened anyway,” and you can’t pull it from any dashboard. You get it by turning the channel off somewhere and watching what your backend revenue does without it.

That’s incrementality: The revenue that exists because the ads ran and wouldn’t exist otherwise. It’s the only figure that deserves to steer spend. You don’t need a measurement vendor to start.

• Choose a geo holdout over a conversion-lift study if you run enough volume to split by region. It’s cleaner, and you own the data.
• Hold the channel out of comparable markets for a full purchase cycle, not a week. Four weeks minimum.
• Measure the delta in backend revenue between held-out and live regions, not the marketing platform’s reported conversions. It can’t see its own absence.
• Run the test before you trust any reconciled or data-driven number. The holdout tells you whether the model was even close.

With more than one marketing platform in the mix, hold each one out in its own window rather than trying to reconcile its reported results. You can’t divide a sale among platforms that each claimed the whole thing.

One holdout won’t settle every budget question. For broader decisions, marketing mix modeling can provide a top-down view of each channel’s contribution from aggregate spend and outcomes. Use it alongside holdouts, and let attribution handle the day-to-day. No single one is the truth. They check each other.

So stop asking which number to trust. That question has no good answer because it assumes one of the two instruments on your desk is calibrated, and neither is.

Ask the other question instead. What does your backend revenue do when the campaign goes dark?

Everything that matters is in the answer.

Every marketing platform in your stack runs hot, each in its own way. The backend runs cold.

Until you turn a channel off and read the patient directly, you’re still averaging a row of miscalibrated thermometers and calling it a diagnosis. It costs me something to say the number flattering my channel is the unreliable one, too, but a wrong number that happens to favor you is still a wrong number.