← Blog
ResearchJune 22, 20261 min read

The Incrementality Era: Why Last-Click Is Finally Dying

Only one in five marketers still trust last-click. In 2026, incrementality testing became the most-trusted method in marketing — here's what that changes.

By The Ad Spend
A man rests his head and arm on a cubicle partition, pensive.

Ask a paid-media veteran what their reported ROAS actually means and watch them squirm. The number is inflated, and everyone knows it — branded search and retargeting take credit for conversions that would have happened anyway. In 2026, the industry finally did something about it: incrementality testing went from a specialist's tool to the most-trusted measurement method in marketing.

The trust has already shifted

The collapse of faith in last-click is documented. Only 21.5% of marketers say last-click reasonably reflects a platform's long-term impact, per an eMarketer/Snap survey, and three-quarters are actively moving away from it. What replaced it at the top of the trust ranking is causal measurement: in Haus's 2026 Marketing Decision Confidence Index, 60% of US senior decision-makers said they trust independent incrementality testing most — twenty points ahead of MMM and nearly double in-platform reporting. Adoption has followed: 52% of US brand and agency marketers now run incrementality tests.

The numbers that embarrass last-click

When you actually test, the gap is stark. Across 225 DTC geo-incrementality experiments, Stella found branded search returned a median incremental ROAS of just 0.70x — the lowest of any channel — versus a 2.31x portfolio median. No wonder 78% of decision-makers believe at least one of their current channels fails incrementality testing if measured properly.

How to run a test that means something

A geo-based holdout — holding back a matched region from exposure while the rest of the market sees the ad — is the standard for isolating causal lift. The test needs to run long enough to collect statistical power, and results need to be read against a planned sample size, not stopped early because one variant looks good. Done right, the result is a number you can trust: what the campaign actually caused, not what the platform attributed to itself.