ServiceMethodTeamFAQThe BlogLet’s talk
← All articles

Incrementality12 October 2026

Incrementality: how many of your affiliate sales would have happened anyway

It is the only question that decides whether a programme creates value or merely moves it. No dashboard answers it: it has to be tested.

By Léo Baumgarten
EU Market Lead, Affiliate

An incremental sale is one that would not have happened without the partner credited for it. A non-incremental sale is one you would have banked anyway, and on which you pay a commission. The two look identical in your dashboard. That is the whole problem.

How a non-incremental sale is manufactured

The mechanism that conflates them is well known. A buyer has chosen their product, lands on your checkout page, sees a "discount code" field, opens a tab, searches your brand, finds a deal site or installs a cashback extension, comes back and confirms. The last click belongs to the partner; the purchase decision predates it by several days. The partner created nothing: it inserted itself.

Attribution does not answer the question

No attribution tool solves this, for a fundamental reason: attribution splits credit between observed touchpoints, it never says what would have happened in their absence. Moving from last click to a multi-touch model makes the split fairer between partners without answering the question. The only way to answer it is to create a case where the partner is absent, and compare.

Three methods, in order of rigour

Three methods exist, in order of rigour. The on/off test deactivates a partner or a family of partners for several weeks and watches total revenue: not attributed sales, the total. The geo test activates the partner in some regions and not others, which neutralises seasonality. The holdout, finally, keeps a fraction of users out of the scheme; it is the cleanest method and the hardest to run in affiliate, where you do not get to choose who goes through a cashback site.

A few design rules avoid wasted tests. The period must be long enough to exceed your buying cycle. It must avoid peaks (Black Friday, sales, launches) which add variables impossible to isolate. And the metric watched must be total revenue and the share of new customers, never attributed sales volume, which falls mechanically the moment you switch a partner off and proves nothing.

The percentages we will not repeat

On results, we prefer to stay careful. Figures circulate (this percentage of non-incremental conversions, that proportion for cashback and coupons) but the accessible studies come mostly from platforms and measurement tools, which are interested parties, and press write-ups often cite reports that cannot be found. So we will not repeat those percentages. What is documented and solid is the method, and the fact that on/off tests run on loyalty partners have shown, in some cases, total revenue left roughly unchanged.

Adjusting rates rather than cutting partners

The practical conclusion is not to cut cashback and coupons. These partners secure baskets, recover abandoned carts and reach bases of identified buyers. It is to know what you are buying: a full commission on an incremental sale is an excellent investment, the same commission on a sale already won is a discount you granted without deciding to. Measuring the difference lets you adjust rates instead of removing partners: and that, in the end, is the whole job.

Sources

  1. CJ (Junction), A Study in Affiliate IncrementalityAn affiliate network’s study across 5.5 million transactions: useful on method, to be read with its author’s commercial interest in mind.
  2. Haus, Affiliate on/off test on loyalty partnersA D2C brand that switched off its US loyalty partners for seven weeks. Published by a measurement vendor, so an interested party.

Share this article