Attribution Is Lying to You! Here’s why…

For every agency owner who has ever presented a client report and quietly wondered if the numbers were actually true.

There is a number in your dashboard right now that your client believes represents reality. It probably does not.

That number is your cross-platform ROAS. And the way it is calculated — the way every major ad platform calculates it — is fundamentally broken in a way the industry has collectively decided not to talk about loudly enough.

Here is what is actually happening.

A customer discovers your client’s brand through a TikTok video on a Tuesday. They do not buy. On Thursday they see a Meta ad in their feed while scrolling before bed. They still do not buy. On Saturday morning they search Google for the brand name, click an ad, and purchase.

Google reports a conversion. Google takes 100% of the credit. TikTok reports nothing. Meta reports nothing. Your Meta campaign ROAS looks weak. Your TikTok campaign looks like it produced zero return. Your Google brand search campaign looks like a hero.

None of this is accurate. The customer’s decision involved all three channels. But your platform data tells a story where only the last click mattered — and you are making budget decisions based on that story every week.

**This is last-click attribution, and it has been the dominant measurement methodology in digital advertising for over a decade.**

The industry has produced increasingly sophisticated alternatives — linear attribution, time-decay models, data-driven attribution, Markov chain models. Each one attempts to distribute credit across touchpoints more fairly than last-click. Each one requires a fundamental assumption about how customer journeys work that cannot be verified. Change the model and the credit distribution changes. The campaign rankings change. The budget recommendations change. You have not discovered truth — you have selected a different set of assumptions.

The deeper problem is structural. Each platform’s attribution model is designed by that platform, optimized to justify that platform’s ad spend, and reported through that platform’s dashboard. Meta’s model credits Meta. Google’s model credits Google. They are not neutral arbiters of your marketing performance. They are competitors for your budget who also control your measurement system.

After Apple’s iOS 14.5 changes in 2021, Meta could no longer observe a significant portion of purchases made by iPhone users. Their solution was to begin estimating — statistically modeling — the conversions they could no longer directly observe. Some percentage of the purchase events in your Meta reporting right now are not observed purchases. They are Meta’s algorithm predicting that a purchase probably happened. You are evaluating campaign performance against a number that is partly fabricated by the company whose revenue depends on that number looking good.

Add GDPR and CCPA consent declines — which run at 40-60% in European markets and significantly affect US data quality too — and the conversion dataset underlying every attribution model is incomplete by design. You are modeling customer journeys using data from the subset of customers who consented to being tracked. That subset is not representative of your full customer population.

**The attribution industry has been selling complexity as a solution to a problem that complexity cannot solve.**

Multi-touch attribution, Markov chains, Shapley values — these are sophisticated tools applied to a fundamentally broken data foundation. They produce precise-looking outputs from imprecise inputs. The precision is not real. The confidence it creates in budget decisions is not warranted.

So what do you do instead?

You stop trying to assign credit across a customer journey you did not design and cannot fully observe. You start asking a different question: is each campaign doing the job it was built to do, at a cost that justifies its budget?

A brand awareness campaign’s job is to reach the right audience at an efficient CPM with enough frequency to create recognition — not to generate trackable conversions. A prospecting campaign’s job is to drive qualified clicks and build the audience pool that your retargeting campaigns will close — not to produce last-click ROAS. A direct response campaign’s job is to convert in-market buyers — and that is where ROAS, measured cleanly with view-through attribution stripped out, is actually meaningful.

When you evaluate each campaign against the right metric for its function rather than forcing every campaign through the ROAS lens, you stop making decisions based on a number that was never designed to measure what you are measuring.

The question is not which channel gets credit. The question is whether every dollar you are spending is in the right place right now — doing the right job, at the right cost, for the right audience.

That is what Kaivo tries to answer.

—
*Kaivo is an AI-native advertising intelligence platform built for digital agencies.*