Attribution tells you which touchpoint got the credit. Incrementality testing tells you whether the ad caused the sale at all. Those are not the same question, and confusing them is how budgets get wasted with a clear conscience.

Why attribution can't answer it

Every attribution model, first click, last click, or data-driven, starts from conversions that already happened and divides the credit. It never asks the counterfactual: would this person have bought anyway.

That's why retargeting and brand search always look brilliant in attribution. They sit next to conversions that were going to happen regardless, and the model rewards them for being in the room.

What incrementality measures instead

Incrementality testing measures cause with a control group. You withhold ads from a randomly chosen slice of your audience or a set of regions, then compare their conversions to the exposed group.

The gap between the two is the lift: the conversions that only happened because of the ads. Everything else was going to happen anyway, and no honest measurement should give the ads credit for it.

The main ways to run it

  • Conversion lift studies: a user-level randomized holdout, run on the ad platform itself.
  • Geo experiments: turn a channel on or off by region and compare against matched control regions.
  • Ghost ads and PSA tests: log the ad the control group would have seen, so the comparison is clean.

They differ in precision and cost, but they share the one thing attribution lacks: a control group.

What it costs you

Incrementality isn't free. A holdout means deliberately not advertising to some people, which feels like leaving money on the table, and tests need enough volume and time to reach significance.

But that cost buys the one number attribution can't fake: what your ads actually add. A cheap wrong answer is more expensive than a test.

How it fits with your models

Incrementality isn't a replacement for attribution or MMM, it's the calibration layer under both. You run a test, get the true lift for a channel, then use it to correct what the models tell you the rest of the time.

Without that anchor, your MMM and your attribution are two opinions. With it, they're opinions checked against a fact.

Attribution will always tell you where the credit landed, and incrementality is the honest answer to whether the credit was earned. Run the test, anchor your models to it, and you stop paying for conversions that were coming anyway.

Want a stronger data analyst role or a raise? Grab the FREE Product Analyst Playbook and get the exact roadmap to your next offer.