How to Prove Marketing Incrementality in Budget Meetings

Marketing incrementality is proven in a budget meeting by bringing numbers with the four properties a finance lead recognises: from a source that does not sell you media, with coverage stated, with the design stated, and with an interval and a floor. In practice that is five numbers from data you already hold, one qualified holdout, and a sentence that admits what you cannot know. The Price of Being Found calls the question every CFO is actually asking not "what was the ROAS" but "how much of this do you actually know".

Why the ROAS slide loses

A slide of platform ROAS figures cannot survive one question: which of these is true? Meta's, Google's and the store's numbers answer different questions and cannot be reconciled, only chosen between. The finance lead knows this even when they cannot name it, because the sum of the platforms' claims exceeds the revenue in the ledger. Black Friday ROAS: what platforms report vs orders shipped walks through the reconciliation problem; this article is the meeting itself.

The five numbers

1. Coverage. Attributed conversions divided by actual orders, by month. Reported cost per sale is the real one divided by this number, so a coverage of 0.6 means every cost figure in the deck is 67% above reality before any media price change. One published census found 48.6% of sessions with no resolvable source at one company; yours is one query.
2. Claim ratio. Every platform's claimed conversions summed without deduplication, divided by orders. Above 1 for almost any multi-channel brand. Its level says how generous the platforms are collectively; its movement usually means a definition changed, not performance.
3. The one qualified holdout. One channel whose spend share times honest return cleared the smallest lift the design could see, held out for eight weeks with the pre-period fixed. Reported as "revenue in the held-out regions moved by this much, interval this wide, from a design that could see anything above 8.3%".
4. The measurability list. Every other channel, split into measurable at current scale and not. The second list is the one finance has never been given, and the book calls it the more useful of the two, because it is where measurement theatre stops being paid for.
5. The anchor date. The date of the last holdout that qualified. Everything after it is model, not measurement, and confidence decays from it. If the honest entry is "never", write never.

Five Monday numbers for DTC owners is the owner's version of the same list, with the visibility rate and the mix effect added.

The script

The book suggests a short paragraph, and it works because every clause is checkable: here is what we can see and the half we cannot; here is what being found actually costs us all in; here is which of our channels we can measure at this scale and which we cannot; here is our last valid experimental anchor and how much confidence has decayed since; here is what I propose to do about each.

Say the second clause first if you say nothing else. Admitting the unresolvable half is what converts the rest of the deck from claims into evidence.

The question you will be asked next

"What happens if we cut channel X?" Two answers, depending on the channel. For the held-out one: the interval, and the decision rule you wrote before the test. For every other channel: "its spend share times its honest return is below what a test could detect at our scale, so we cannot know from a test, and we are managing it on stated judgement rather than pretending a null result was evidence". That second answer sounds like weakness and is the strongest sentence in the meeting, because nobody else in the company can say it truthfully about their numbers.

Why the first version makes you look worse

Coverage will be lower than anyone expected. The claim ratio will show the platforms collectively took credit for more sales than the company made. The anchor date will, for most organisations, be never. The book calls this the correct first result: it is what the situation was before you measured it, and a smaller number you can defend is worth more than a larger number you cannot. The organisations that get past it are the ones where somebody senior says that out loud, once.

What to do this week

• If you have to defend the number: compute coverage and the claim ratio for last quarter today. Two hours, no vendor. Put both on the first page of the deck, above any ROAS.
• If you own the budget: tell whoever presents to you that the store's order count is the number the meeting is graded on. Every platform report then becomes context rather than verdict.

A causal read on the GA4 export gives every channel an incremental figure with an interval between anchors and states which channels fall below its fit floor. The pricing page carries four answers about buying one; the five numbers above come first.

As of 9 September 2026. The four properties, the five numbers, the coverage census and the meeting script are from The Price of Being Found (Edition 2.10), Chapters 8, 19 and 20, with the book's caveats.