Claim ratio

You meet it the first time somebody senior puts the ad platform's dashboard next to the invoicing or booking system and the two numbers are nowhere near each other. Usually the platform is higher. Sometimes, on an account whose offline conversion import has quietly stopped, it is dramatically lower. Either way the meeting ends with someone asking which number is real, and the honest answer is that the question is malformed.

It also turns up in the smaller version of the same argument: two ad platforms each claiming the same sale, a monthly report where reported revenue exceeds actual revenue, or a campaign that looks like it collapsed in a month when total real orders barely moved.

A platform-reported conversion is not a count of things that happened. It is a claim that the platform caused something, made under that platform's own attribution rules, inside that platform's own window, using that platform's own modelling for the conversions it could not observe directly. Your booking system, by contrast, is counting events. The two are different kinds of object, and there was never a design goal that made them agree.

So the gap is not an error to be found and fixed. It is the sum of every modelling assumption between a click and a settled order: how long the window is, whether view-through counts, how much of the drop from consent and app-level tracking restrictions is being filled in statistically, and whether two platforms are both taking credit for the same person.

Which is also why the gap moves. Every one of those inputs changes: the platform revises its modelling, consent rates shift, the channel mix changes, a promotion pulls demand forward. A ratio measured last quarter is a measurement of last quarter.

The cost is not the reporting discrepancy. It is every decision made on the wrong side of it. A target cost per acquisition set against inflated conversion counts is set too low, so the account bids for volume it is not really getting. Set against undercounted conversions, the same target is too high and the account throttles spend that was profitable.

The expensive version is the diagnosis error. A campaign's reported conversions fall by a third, and the reflex is to treat it as a performance problem: check the ads, check the landing page, cut the budget. If total real orders were flat over the same period and a sibling channel scaled up and absorbed the credit, then nothing broke and the intervention makes things worse. A conversion drop is an attribution question before it is a performance question.

The practical rule is short. Never report a platform conversion, cost per acquisition or return on ad spend without the source-of-truth figure beside it, and never apply a fixed correction factor to close the gap.

The fixed correction is the tempting mistake, because it feels like rigour. Somebody measures the ratio once, finds the platform is claiming 1.8 times reality, and divides by 1.8 from then on. That is worse than not correcting at all: it converts a moving, visible discrepancy into an invisible constant that is wrong in a new way every month, and it hides the one signal worth having.

Tracked as a per-channel time series instead, the ratio becomes an instrument. A stable ratio means the measurement setup is behaving and platform movements can be read at face value. A ratio that jumps means something changed in the measurement, not the market, and the first place to look is the tracking, not the campaign.

It is not an accuracy score. A claim ratio above one does not prove the platform is lying and a ratio near one does not prove it is right. Both channels could be over-claiming against a source of truth that is itself incomplete.

It is not incrementality. Claim ratio compares two counts. It says nothing about what would have happened without the spend, which is a different question needing a holdout, a geographic test or an unplanned pause to answer.

It is not a benchmark you can borrow. The ratio is a property of one account's tracking setup, consent profile, sales cycle and channel mix. Someone else's number tells you nothing about yours, and that includes ours.