The Maze: Last-click attribution and post-purchase surveys disagree because they answer different questions. One records the conversion session. The other asks which channel entered the buyer's memory first. Matching them at order level maps who created demand and who collected it.
Google closes demand that other channels started. In Curtis Howland's matched matrix, 47% of buyers who remembered TV or streaming as their first touch arrived through Google at the final click. The same was true for 45% of podcast-led buyers and 44% of YouTube-led buyers. Even among people who named TikTok, Google was the largest last-click destination at 34%. This is not evidence that Google caused those purchases. It is evidence that search often sits closer to checkout than the channel that made the brand memorable.
Channel “agreement” varies sharply. Google-search awareness matched a Google UTM for 71% of orders. Instagram or Facebook matched Meta for 38%. TikTok matched TikTok for only 22%. Friend referrals behaved differently: 48% landed as direct or none, while 38% ended on Google. That makes platform-reported ROAS a map of captured journeys, not necessarily originated demand. Google's own GA4 documentation says paid-and-organic last click assigns 100% of a key event's value to the final non-direct channel. The model is doing exactly what it was designed to do. The management mistake is asking it a first-touch question.
The useful unit is the matched order. A survey response alone has memory bias. A UTM alone has journey loss. Joining both observations on the same transaction creates a two-axis diagnostic. Fairing's Last Click Report uses this same pivot: declared source versus Shopify last-click UTM. Operators can then compare channel rows, estimate response-adjusted acquisition volume and identify where platform credit diverges from customer memory. But extrapolation should wait until response rates and row counts are stable. Howland suggests more than 30% response and at least 100 answers per row; those are practitioner guardrails, not universal statistical guarantees.
Disagreement should trigger tests, not automatic budget moves. The matrix omits the brand, geography, field period, sample size, spend, frequency and incremental lift. Buyers can misremember. Response order can bias clicks. Untracked exposure can disappear. A search click may be both capture and genuine discovery. The right next step is to use the mismatch as a hypothesis: social or offline media creates awareness, search harvests it. Then test that hypothesis with geo holdouts, lift studies, model comparisons or controlled spend changes. Fairing's survey guide similarly frames survey-plus-UTM data as a simplified first/last-touch view—not causal proof.
The payoff is a better budget conversation. Search can keep credit for closing intent without claiming every sale originated there. Social can argue for awareness using matched customer evidence. Finance gets a testable bridge between platform ROAS and blended acquisition cost.
The same post-purchase form can also pull more work from a single customer moment. “Why did you buy today?” informs creative. “What almost stopped you?” exposes objections. Site issues create a CRO backlog. Missing-product answers become a demand file. Attribution is the opening question; the operating system is the larger prize.
