Attribution Is Mostly Guesswork

Open any marketing dashboard and you'll find confident sentences: Google Ads drove 14 conversions. Organic search delivered 9. Social contributed 3. The precision is soothing — real numbers, decimal points available on request. Here is the uncomfortable truth underneath: nobody, including the dashboard, actually knows where those clients came from. Attribution — the assignment of credit for a sale to a channel — is not a measurement. It's an allocation rule, and the rule was chosen by whoever built the report.

The attribution reality: a client's real journey has many touches — the colleague's mention, the assistant's answer, the review skim, the article at 9pm, the ad that happened to be the last click. Attribution software sees a fraction of these and assigns all the credit by convention, usually to the final click before contact. The report isn't lying about what it saw; it's guessing about what mattered — and the guess is dressed as a decimal.

Where the guess breaks, concretely

The last-click lie. The journey above, run through standard attribution: the ad gets 100% of the credit; the referral that started it and the reviews that carried it get zero. Practices then double the ad budget and wonder why doubling didn't double anything — they paid twice for the receipt, not the cause.

The invisible channels problem — and it's growing. Attribution counts what leaves digital fingerprints. Word of mouth leaves none. The AI assistant's recommendation leaves almost none — a client told "call this practice" arrives typing your name, and the dashboard files them under "direct," the euphemism for no idea. The channels most valuable to a practice are precisely the least trackable, which means software attribution systematically under-credits your best assets and over-credits whatever's easiest to tag. (Did AI send that customer? digs into the newest blind spot.)

The incentive problem. Every platform grades its own homework: the ad platform's attribution finds the ads worked; the SEO tool finds search did it. None of them is fraudulent; all of them are counsel for their own defence.

How to hold attribution claims

Not by rejecting them — by demoting them from verdicts to hints, and cross-examining with tools that don't share their blind spots:

Ask the humans, always. The source question — "how did you hear about us?", verbatim, every enquiry — is unfashionable and beats the software where it matters: the client's own story of arrival captures the mention, the assistant, the review, in their weighting. Imperfect memory beats imaginary precision.

Compare the stories. Once a quarter, put the software's channel story next to the humans' answers. The gaps are the finding: dashboards say ads, humans say referrals — now you know where the credit actually belongs, and what to feed.

Settle real disputes with subtraction. When money rides on the answer — is this spend working? — the only honest referee is the holdout: pause it and watch. Would it have happened anyway is the question attribution pretends to answer and subtraction actually does.

Questions practices actually ask

Should I bother setting up conversion tracking at all? Yes — imperfect fingerprints beat none, especially inside a single channel (which ad, which page). The trap isn't tracking; it's promoting the tracker to judge.

What about the fancy multi-touch attribution models? They redistribute the guess more elaborately — smoothing credit across observed touches while remaining blind to the unobserved ones. For enterprise ad budgets, sometimes worth it; for a practice, the source question plus holdouts answers more, for less.

Clients say "Google" when asked. That's uselessly vague, right? Follow up gently once — "searching for something, or did someone mention us?" The second sentence usually surfaces the real chain: "my sister told me, I just googled the name." That's a referral with a Google receipt — file it accordingly.

So what do I put in the source column? The client's story, one line. Over months, the column becomes the practice's true channel mix — the humble spreadsheet that out-reports the dashboards.

---

Part of Measure What Matters — the instrument panel.