An Attribution Report That Says Which Channels to Cut

An attribution report a CFO will accept does four things a platform report cannot: it names the channels to cut with an interval and a floor behind each recommendation, it states the coverage of the data and the claim ratio of the platforms, it labels every number as experimental or observational, and it lists the channels it cannot measure at your spend instead of inventing a figure for them. A causal read on a GA4 export produces the first three in 5 to 10 minutes; the fourth is the honest part most reports leave out.

Why "cut channel X" needs more than a number

The platforms cannot tell you what to cut, because each grades its own work. The Price of Being Found's identity: the ad platform reports conversions it can associate with its own inventory inside its own window; analytics reports sessions it could resolve; the store reports orders. The sum of the platforms' claims exceeds the orders, and none of the three answers what would happen without a channel. A cut recommendation built on a platform's own ROAS is a recommendation the platform has already argued against.

The four properties of a defensible number are the fix. A cut that says "channel X: incremental ROAS 0.7, interval 0.3 to 1.1, observational read on a 40-day export, coverage 0.62" can be discussed. "Channel X: ROAS 1.8, data-driven" cannot, because nobody in the room can say what it means.

What the report contains

A causal read returns, per channel, an incremental ROAS estimate with its confidence interval and the platform-reported figure beside it, plus a ranked reallocation in plain language: scale, cut, leave alone. It is exportable, so the same page goes to the owner, the finance lead and the agency without three different decks. How to export and share attribution reports you can defend covers the sharing side.

Two numbers frame the recommendations. Coverage, attributed conversions over orders, says how much of the journey the estimates rest on. The claim ratio, summed platform claims over orders, says by how much the platforms collectively overstated; the one-hour claim ratio audit computes it from data you already hold.

How to read a cut

Sort by the lower bound of the interval, not the point estimate. A channel whose whole interval sits below break-even is a cut with a number behind it. A channel whose interval straddles break-even is a decision under uncertainty, and the honest recommendation is a staged reduction with a re-read: cut part of the spend, wait a fortnight or longer for slow channels, run the read again, and let the interval move before the rest goes. The fastest way to cut wasted ad spend by channel is the practitioner method. Break-even itself moves with margin and discount, so the comparison is against this quarter's break-even ROAS, not last year's target.

The line the CFO trusts most

The list of channels the report will not rank. Below the fit floor, roughly 5,000 euros a month in paid spend or 40 days of history, and for any channel whose spend share times honest return sits below what a holdout could detect, the honest entry is "not measurable at current scale, managed on stated judgement". The book calls that second list the more useful of the two, because it is where measurement theatre stops being paid for. A report that admits it is worth more in the room than one that ranks everything.

What the four CFO questions look like on this report

Which number is true: the store's orders, with coverage and the claim ratio beside the platform figures. How much do we actually know: the coverage rate and the anchor date of the last qualified holdout. What happens if we cut channel X: the interval, and the staged plan. Why believe you over the platform: the design statement on every line. How to prove marketing incrementality in budget meetings is the script, and the pricing page carries the four answers about buying a read.

What to do this week

• If you have to defend the number: run the read on last quarter's export and rebuild your top-line slide in the four-property format before the next review.
• If you own the budget: read the "not measurable" list first, then the cuts. The first list is where money is being managed on numbers that could not exist.

The interactive demo shows the report on a sample store with no signup.

Product facts as stated on causalityengine.ai on 9 September 2026. The three-systems identity, the four properties and the measurability list are from The Price of Being Found (Edition 2.10), Chapters 9, 15 and 19, with the book's caveats.