What Share of Your Conversions Is Actually Incremental?

There is no honest industry answer to this. The share of conversions that advertising actually caused varies by brand, by channel mix, by how strong the brand already is and by how much of the budget is chasing demand that existed anyway. Anyone quoting a single percentage across brands is generalising from a small number of published experiments onto businesses those experiments never touched. What does exist is a set of results showing how large the gap can get, and a procedure for finding your own.

What the published experiments actually found

Brand search at eBay. Paid brand-keyword ads scored as a top performer under observed-journey logic. When they were switched off, 99.5% of the forgone paid clicks came back through natural search. Almost the entire measured contribution of that channel was demand that would have arrived without paying for it.

Facebook advertising, against randomised ground truth. In the studies The Price of Being Found cites, observational methods overstated effects by roughly three times compared with randomised experiments on the same advertising. In one set of fourteen comparisons, six could not be statistically distinguished from zero at all.

Two things follow. The gap between claimed and caused can be very large, and it is largest exactly where targeting is best, because a system excellent at finding people about to buy produces impressive reported numbers while adding little.

Why your number is not any of those

Those results are about specific advertisers, specific channels and specific periods. A brand with weak awareness buying cold prospecting is in a different position from an established brand buying its own name. The honest statement is directional: the more your spend sits close to existing intent, brand search, retargeting, abandoned-cart flows, the wider the gap is likely to be, and the more it sits on genuine discovery, the narrower.

That is a hypothesis about your business. It is not a number, and treating it as one is the error this article exists to avoid.

The arithmetic that gets you close today

Two ratios, computed from data you already have, put bounds on the question in about an hour.

The claim ratio. Sum what every platform claims for a month and divide by the orders your store shipped. At 1.6, at least 37% of the claims cannot each be a distinct order. That is a floor on the double counting, not a measure of incrementality, and it is the fastest evidence that reported conversion volume overstates reality.

Coverage. Attributed conversions over store orders, which tells you what share of the business the reporting can see at all.

The one-hour claim ratio audit is the procedure for both.

The design that actually answers it

Withhold a channel from a randomly chosen group, fix the window in advance, and compare revenue per customer. That difference is the incremental share for that channel over that window.

Before running it, work out the smallest effect the design could detect at your volume. If your spend share times an honest incremental return sits below that floor, the test cannot answer the question and running it produces a null result you will misread as evidence of no effect. Whether your channels are measurable at all has the arithmetic, and incrementality testing for ecommerce is the playbook.

What to use between tests

Causality Engine uses causal inference on the GA4 history you already have to estimate what each channel caused.

What to do this week

• If you own the budget: compute the claim ratio. It is the cheapest evidence available that your conversion volume is overstated, and it takes an hour.
• If you have to defend the number: refuse to quote an industry incrementality percentage, including a flattering one. The refusal is more defensible than any figure you could cite.

The eBay result, the Facebook comparisons and the measurability arithmetic are from The Price of Being Found (Edition 2.10), Chapters 12 and 19, with the book's stated caveats: they are published findings about specific advertisers, not population estimates.