Incrementality testing

COMING SOON: The true lift behind every channel and campaign

A geo holdout test reveals what’s really driving your revenue, so your decisions are backed by proof.

Book a demo Pricing plans

What incrementality gives you

Evidence you can act on

Know whether a channel is truly driving results, or capturing demand that was already there.

Confident budget moves

Spend more on the channels that earn it, and less on the ones that don’t.

Credit where demand starts

Recognise the channels that build demand, not only the ones that capture it at the moment of purchase.

A real-world check on top of your attribution

Clicks and impressions tell you a channel was part of the journey. They cannot tell you, on their own, whether it caused the conversion. Incrementality answers one plain question. If you paused a channel, would your results change?

How a test works

A guided process, from feasibility check to measuring impact. You set the geo exclusions in your ad platforms, we handle everything else.

1

Checking feasibility

Before recommending a holdout, we analyse historical regional data to determine whether a test can produce a reliable answer. We identify suitable regions, estimate the minimum detectable effect, the smallest lift the test could reliably pick up, and confirm that the expected channel impact is large enough to measure confidently.

Example: London holdout

We choose a branded search holdout in London. We assess whether its conversions can be closely modelled from other comparable regions, and whether the expected impact is large enough to detect reliably.

2

Running the holdout

Once a suitable region has been identified, you run a 30-day geo holdout by setting the selected channel’s spend to zero there while keeping it active elsewhere. This creates a clean comparison between the held-out region and the rest of the country, isolating the channel’s true impact.

Example: London holdout

You pause branded search in London for 30 days while keeping it active across the rest of the UK.

3

Monitoring the test

Throughout the holdout, we monitor spend, impressions and performance for signs of contamination. If unexpected events affect the comparison, we can recommend adjustments to keep the results trustworthy.

Example: London holdout

We track London’s spend and impressions against the rest of the UK throughout the test, ready to flag anything unusual before it affects the read.

4

Estimating the incremental effect

After the holdout, we estimate what would have happened if ads had remained active using a synthetic control. The difference between expected and actual performance becomes the estimated incremental effect. For channels close to the purchase we measure conversions, for channels that build demand earlier we measure traffic and demand.

Example: London holdout

For branded search, we look at conversions and organic traffic.

Conversions: 1,240 expected → 1,225 actual

Organic sessions: 18,500 expected → 21,900 actual

5

Reading the result

We evaluate the estimated effect, confidence interval and statistical significance to understand whether the channel created measurable incremental impact, and how those findings should influence future measurement and budget decisions.

Example: London holdout

In this test, conversions remained broadly stable while organic traffic increased significantly. This suggests many users shifted from paid branded clicks to organic listings, so that budget can be redirected to channels that bring in demand you would not have captured otherwise.

Put your budget where it grows your business

Incrementality shows you which channels create real demand, so you can move spend towards what works and grow with confidence.

Book a demo
Marketing triangulation

One measurement system you can trust

MTA captures the click path. UMM adds the view layer on top. Incrementality validates both, grounding the models in real evidence from what actually changes when you pause a channel, then feeding its findings back to recalibrate UMM. All three work together, each one strengthening the others.

Download the guide Attribution models

Frequently asked questions

What do I need to do to run a test?

How do you know a test will produce a clear result before we run it?

What happens while the test is running?

How long does a test run?

What do the results tell me?

Is this a user-level A/B test?