Incrementality

Incrementality is the measure of how much of an outcome a marketing activity actually caused — the conversions, sales, or signups that would not have happened without it. The crucial idea is the counterfactual: what would have happened anyway, with no campaign at all? Incrementality is the difference between the world with the marketing and the world without it. If a thousand people bought after seeing your ad, but eight hundred of them would have bought regardless, the campaign's incrementality is two hundred, not a thousand. This matters because a great deal of marketing takes credit for demand it did not create — retargeting someone who was already going to buy, or bidding on a brand term a loyal customer would have searched for anyway. Incrementality asks the only question that matters for spending decisions: did this activity create extra results, or merely accompany results that were coming regardless?

Measuring incrementality requires an experiment or a credible estimate of the counterfactual, because you cannot observe both worlds at once for the same people. The cleanest approach is a controlled test: hold out a comparable group from seeing the marketing, run it for everyone else, and compare. The difference in outcomes between the exposed and held-out groups is the incremental lift — the part the marketing genuinely caused. Geo experiments, ghost ads, and matched-market tests are practical ways to build that comparison at scale. The result is often humbling. Channels that look spectacular under last-click reporting frequently show modest true lift, while channels that build demand can be undervalued by reporting that only credits the final click. Incrementality cuts through that, replacing the question "who was nearby when the sale happened?" with "who would not have bought without us?"

Incrementality and attribution are easy to conflate and fundamentally different. Attribution divides credit for outcomes that already happened among the touchpoints that preceded them — last-click hands all the credit to the final interaction, multi-touch spreads it across the path. But attribution assigns credit by correlation, by who was present along the way, without asking whether any of them caused the result. It can hand a retargeting ad full credit for a purchase the customer was always going to make, simply because that ad was the last thing they saw. Attribution answers "which touchpoints get the credit for the conversions we observed?" — a bookkeeping question about dividing a pie whose size it never questions.

Incrementality asks the prior, harder question: how big is the pie that the marketing actually created? It measures causation, not correlation, by comparing what happened to what would have happened without the activity. The two can disagree sharply. A channel that attribution credits heavily can have low incrementality if it mostly intercepts demand that already existed, while a channel attribution ignores can be highly incremental if it generates demand that later converts elsewhere. The practical discipline is to use attribution for operational, day-to-day optimization within a channel, but to lean on incrementality for the bigger budget decisions — whether a channel is worth funding at all — because only incrementality tells you whether the money is creating results or just being present when results occur. Trusting attribution alone systematically over-rewards demand-harvesting and under-rewards demand-creation.

Measure incrementality by building a credible counterfactual rather than inferring causation from correlation. The gold standard is a controlled experiment: randomly hold out a comparable audience or set of geographies from the marketing, run it for the rest, and read the difference in outcomes as the incremental lift. Use these tests for the decisions that matter most — whether to keep funding a channel, how much a campaign truly contributes, whether a tactic creates demand or merely harvests it. Size the test so the lift is detectable, run it long enough to capture delayed effects, and pick a control group that genuinely resembles the exposed one. Treat the result, not the attributed number, as the answer to whether the spend paid off.

Beware the traps. The most common is trusting attribution as a proxy for incrementality and so over-funding channels that look efficient but mostly intercept existing demand. Another is a poorly designed test — a control group that differs from the exposed group, a window too short to capture lagged conversions, or a sample too small to detect real lift — which produces a confident but wrong answer. A third is running an experiment once and assuming the result holds forever, when incrementality shifts as markets, creative, and saturation change. And a fourth is ignoring that heavy spending can suffer diminishing incremental returns, so the lift from the next dollar is smaller than the last. Used rigorously, incrementality is the most honest measure of marketing's contribution there is, precisely because it refuses to take credit for what would have happened anyway.

Incrementality — the share of outcomes a marketing activity truly caused versus what would have happened anyway — measures causation rather than the correlation that attribution assigns.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where incrementality is a core concern: