Incrementality: is paid search actually adding revenue?

Incrementality is the share of results that would not have happened without the marketing. It answers a different question from attribution: attribution asks which touchpoint receives credit, incrementality asks whether the activity caused anything. For paid search the gap between the two is often wide, and for SEO the question is harder to test. Neither channel is doing anything wrong by answering the credit question.

Why attribution overstates paid search

Paid search sits at the end of the journey. It captures people who are already looking for the product and records the click immediately before the purchase, which makes it appear responsible for revenue it merely observed. That is a property of the measurement, not a flaw in the channel.

• Brand searches: someone who already knows the company searches its name, clicks the ad because it appears first, and buys. The organic result would often have caught them.
• Repeat buyers: a returning customer does not need the ad, but the ad still receives the click and the credit.
• Organic overlap: on generic queries where the site already ranks well, the ad often appears above the site’s own organic result and pays for a click that was already there.

Where SEO and paid search overlap

SEO and paid search are not two independent channels competing for one click; they interact. A strong organic presence lowers the value of a paid click on the same query, and paid campaigns can raise branded search volume that later converts organically.

That interaction is one reason channel-level reporting rarely sums to total revenue. Each channel receives credit for demand the other created, and the sum of the parts looks larger than the business. It is a reporting artefact, not a sign that anyone has miscalculated.

Two channels can both report the same sale without either being wrong. They are answering a question about credit, not about cause.

Three tests that answer the real question

1. 01On/off holdoutPause a keyword group or campaign for two to four weeks and compare total revenue, not campaign revenue, against a forecast built from the same period last year plus recent trend. The difference is the incrementality.
2. 02Geo experimentTurn the campaign off in a set of comparable regions while leaving it running elsewhere, then compare the change between the two groups. This controls for seasonality and market-wide swings better than a single-market pause.
3. 03Matched-market comparisonChoose markets that have historically tracked each other closely, apply the change in half of them and measure the gap that opens. It is the same idea as the geo test, with more discipline about how pairs are chosen.

How to read the result

1. Compare total revenue, not attributed revenue. Attributed revenue falls to zero the moment a campaign pauses, which is not a finding.
2. Account for the change in organic clicks. Where organic picks up part of the traffic, the true loss is smaller than the paid report suggests.
3. Check the purchase cycle. A four-week test on a product with a two-month consideration window will look incremental when the effect is only delayed.
4. Treat anything inside the noise band as zero. If weekly revenue normally swings 20 percent, a 6 percent drop is not evidence.

What to do with the answer

If paid search turns out to be mostly credit-taking, cutting it is not the only option. Branded ads can defend against competitors bidding on the brand name, and that value appears as defence rather than as lift. The point is to pay for it knowingly, at a size the defence justifies.

For SEO the same question is harder, because a ranking cannot be paused. Periods when a page slipped, a site migration or a regional rollout can serve as natural experiments to see what changed in total demand. Whichever method is used, the question should stay the same across channels: incrementality is one question asked twice, not two reports compared.