How do you know whether a channel is truly driving growth, or just capturing conversions that would have happened anyway?

That is the question behind everything people call "incrementality."

Every good measurement setup tries to capture it. Only a few actually can.

That’s why this week, we’ll look into why treating measurement as a process, and how a geolift test actually pins down what a channel is worth.

Let’s get into it:

Incrementality is the most important thing to get right in measurement.

Answering: Is my channel truly driving growth, or just capturing demand that was already there?

And even the biggest brands underestimate what it actually takes.

Because at a larger scale, incrementality isn't a number you can read off.

It's a continuous series of experiments you have to run.

The channels you most want to understand are the ones you can't really click-track. New channels, awareness channels, anything upper-funnel.

A fully deterministic MTA model is blind to all of them.

And even for click-based channels like brand search, that’s not really suited to tell you their incremental impact.

So there are two methods that approach to measure a channel’s impact:

The most important point here for both of them:

To learn how incremental a channel really is, you have to be willing to make mistakes.

You have to "waste" money on purpose.

Turn a channel down, in a controlled way, and sit with a worse result for a few weeks. Because that's the only way the data shows you what the channel was actually worth.

It forces one real decision. Do you want the best outcome today, or twelve months from now?

You can't have both.

Get comfortable with that, and the tools start working.

Geolift testing is a test. It's in the name.

And Marketing Mix Modelling needs variance to model anything at all.

Spend the same share in the same places every month, and no model on earth can tell you what actually worked.

You have to move things around on purpose.

So if you really want to find out the incremental impact of your marketing channels, there are two things to do:

Btw, I recently recorded a full 45-min course on marketing measurement, where I also cover MMM and Incrementality. If you’re interested you can watch it here for free: Measurement Course.

So if the answer is "run experiments," what does one actually look like?

The cleanest option is Geolift testing.

The idea: you use a geography as your control. You change spend on a channel in one set of regions, keep another set untouched as a baseline, and measure the difference.

Done well, a geo test answers the questions a click-based model never can:

"Turn something down in Region A and watch what happens" sounds trivial.

It isn't.

Picking comparable regions, sizing the test so the result is readable, controlling for seasonality and outside noise, running it long enough to trust the number. That's where most tests fall apart and hand you a confident answer that's simply wrong.

Over the last few months we've started running geolift tests with a handful of customers at Klar, and designing them properly is a crucial part of the work.

I recorded a full walkthrough of how to design, run and read these tests: