Incrementality Testing Just Crossed the Tipping Point: What the New 2026 Numbers Mean for Your ROAS

Incrementality testing ecommerce adoption just crossed a real threshold.

Per a July 2025 TransUnion survey reported by Digital Applied, 52 percent of US brand and agency marketers now run incrementality tests, up from niche status just two years earlier.

In retail media specifically, the ANA found 71 percent of advertisers now rank incrementality as their single most important KPI.

This is no longer an advanced technique reserved for data science teams…

It is becoming the baseline expectation for proving a channel actually earned its budget, and the gap it is uncovering is not small: platform-reported ROAS is running 20 to 60 percent above measured incremental lift.

Why the Number Moved This Fast

Three forces converged to push incrementality from a niche practice into a mainstream one. Signal loss from iOS tracking limits and cookie deprecation made attribution models less reliable at exactly the moment budgets got more scrutiny.

CFOs stopped accepting a platform’s own attributed number as sufficient proof, and started asking for a causal answer instead.

EMARKETER’s own reporting on the same TransUnion data frames it plainly: the approach has moved from niche practice to mainstream adoption, driven by tracking limitations and growing pressure to prove that ad budgets generate real business impact.

Retail media accelerated the shift further, since advertisers spending into Amazon, Walmart Connect, and similar networks wanted proof of net-new sales rather than credit for demand that already existed.

How Big the Gap Actually Is

The 20 to 60 percent range is not an outlier estimate.

Eightx’s research on incrementality testing puts measured incremental ROAS typically 30 to 60 percent below platform-reported ROAS, with the gap widest on brand search and retargeting, the two channel types most likely to intercept demand that would have converted anyway.

A worked example from Haus’s guide to incrementality testing makes the mechanism concrete: a brand spending $100,000 on Google Ads sees $300,000 in attributed revenue, a reported 3x ROAS.

A geo holdout on the same spend reveals the incremental lift is closer to $50,000, an incremental ROAS of 0.5x rather than 3x.

Platform attribution was not lying about the sale happening. It was crediting a sale that would have happened regardless of whether the ad ran.

What a Real Test Actually Looks Like

A geo holdout is the most common design, and the mechanics are straightforward even if the statistics behind it are not.

Split matched markets into treatment and control, keep ads running normally in treatment markets, and fully pause, not reduce, the channel being tested in control markets.

Metricuno’s guidance on running these tests recommends at least 6 to 8 matched market pairs, since fewer than that leaves confidence intervals too wide to distinguish a 15 percent incremental channel from a 40 percent one.

A pre-test period roughly equal in length to the test itself is worth running too, to confirm treatment and control markets were actually tracking together before the test started.

Four weeks is a reasonable minimum duration to account for weekly seasonality and let a real signal emerge from the noise.

The Barriers Are Real, but Getting Smaller

Adoption still outpaces maturity. A meaningful share of marketers cite accuracy concerns, difficulty applying incrementality across every channel and targeting method, and a lack of the right tooling as the reasons they have not scaled testing further.

Cost has come down, though. As we covered in why platforms shouldn’t grade their own homework, Google has lowered the minimum budget for its own incrementality experiments from roughly $100,000 to $5,000, which puts a real holdout test within reach of mid-market brands rather than only enterprise budgets.

Where Incrementality Fits in a Real Measurement Stack

Incrementality testing is not a replacement for attribution or marketing mix modeling, it is the calibration layer that keeps both honest.

We walk through how the three approaches complement each other in MMM vs attribution vs incrementality, but the short version is that a periodic holdout test tells you the truth about a channel’s real contribution, an always-on first-party attribution layer tells you what is happening day to day, and MMM extends that picture across channels a click can never see.

None of the three works well in isolation, and none of them is free from the platform’s own incentive to claim credit, which is exactly why an independent, click-based baseline matters as much between tests as during them.

If you want to see the gap between what Meta, Google, and TikTok report and what your spend is actually producing, independent of any single platform’s own math, book a live AdBeacon demo and we will show you the gap on your own account.

—-

FAQ

How many brands and agencies actually run incrementality testing?

About 52 percent of US brand and agency marketers, per a July 2025 TransUnion survey reported by EMARKETER, up from niche adoption just two years earlier.

How much does platform-reported ROAS typically overstate real performance?

Independent research puts measured incremental ROAS 20 to 60 percent below platform-reported ROAS, with the widest gaps on brand search and retargeting.

What is a geo holdout test?

A geo holdout splits matched markets into a treatment group, where ads run normally, and a control group, where the channel being tested is fully paused. Comparing revenue between the two groups isolates the ad-driven lift from demand that would have existed anyway.

How long should an incrementality test run?

Four weeks is a reasonable minimum to account for weekly seasonality and reach a statistically meaningful result, with longer durations recommended for smaller order volumes.

Why is incrementality especially important in retail media?

The ANA found 71 percent of retail media advertisers now rank incrementality as their top KPI, since retail media budgets are judged specifically on whether they drive net-new sales rather than capturing demand that already existed.