How to Settle Attribution Debates With Your Agency

Attribution debates with an agency end when the brand stops arguing about whose model is right and starts owning the scoreboard: orders from the store as the only fact about money, a coverage rate and a claim ratio computed by the brand, a design statement on every number the agency reports, and one registered holdout the agency helps design and cannot reinterpret. None of that requires distrusting the agency. It requires noticing what it is paid on.

The incentive, stated without blame

The Price of Being Found puts the structural problem in one sentence: the party that sells you the advertising also measures whether it worked, reports the result, and there is no independent audit. An agency sits one step removed and inherits the same scoreboard. Its performance review, its retainer renewal and its case studies all rest on platform-reported ROAS, and the platform grades itself. The book's author describes having been the person who, agency side, took the middle of three numbers for a board pack because the highest looked like marking your own homework and the lowest looked like an admission. It was not a scandal. It was a Tuesday.

That is why the debate cannot be settled by a better model from either side. It is settled by moving the scoreboard to a source neither side produces.

Four things to ask for

1. Orders as the scoreboard. State in writing that the store's order count, or finance's revenue, is the number the relationship is graded on. Platform reports become context. This costs the agency nothing if it is doing good work.

2. Coverage and the claim ratio, computed by you. Coverage is attributed conversions divided by orders. The claim ratio is every platform's claimed conversions, summed without deduplication, divided by orders. Both take an hour on data you hold. The one-hour claim ratio audit is the procedure. A claim ratio of 1.8 means the platforms collectively claimed 80% more sales than you shipped, and it ends the argument about whether the reports can all be right.

3. A design statement on every number. Experimental, quasi-experimental, or observational. Most agency reporting is observational and should be labelled as a description, not a causal claim. The book's observation is that most marketing arguments are two people comparing an experimental number to an observational one without noticing. The label ends those.

4. One registered holdout. Pick the channel whose spend share times honest return clears the smallest lift a test can detect at your scale. Have the agency help design it: geography unit, pre-period, minimum detectable effect, decision rule. Write it down before launch, somewhere neither party can edit. The agency's expertise improves the design; the registration removes the reinterpretation. Incrementality testing for ecommerce is the playbook.

The questions an agency should be able to answer

The book's vendor list applies to agencies too, and three of its questions do most of the work. Which of our channels are not measurable at our current spend? What would your reporting show if the advertising had no effect at all? What is the confidence interval, not the point estimate? An agency that answers the first with a list, the second with a placebo check, and the third with an interval is a partner. One that cannot is reporting, not measuring, and that is fine as long as everyone says so.

What this changes on the invoice

Nothing, immediately. What changes is the meeting. Reallocation decisions move onto the causal read or the holdout, and the agency's own work is judged on the number it cannot inflate. Good agencies tend to welcome this, because it also protects them from being blamed when a platform redefines a click, as Meta did in March 2026 according to the book's account, and the reported numbers move with no change in performance.

What to do this week

• If you have to defend the number: send the four asks to the agency in writing before the next monthly review, so the answers exist before the numbers do.
• If you own the budget: compute the claim ratio yourself once. It is the one number in the relationship that neither the platform nor the agency produces.

A causal read on your own GA4 export gives you per-channel incremental ROAS with intervals independently of anyone's reporting, which is the fastest way to arrive at the meeting with a number of your own. How to prove marketing incrementality in budget meetings is what to do with it once you have it.

As of 9 September 2026. The scoreboard problem, the three-systems identity and the vendor questions are from The Price of Being Found (Edition 2.10), Chapters 9 and 17, with the book's caveats.