Incrementality for Apps

Incrementality for apps measures what your ads truly caused, the installs and actions that would not have happened without them, using holdout or lift tests rather than trusting attribution's word for it.

You have a report that proudly credits your ads with a pile of installs, and some large share of those people were coming anyway. They already knew your app, they searched for it, they were going to download it with or without your money. Attribution happily takes credit for all of them, because its job is to assign the last touch, not to ask whether the touch mattered. Incrementality is the discipline that asks the harder question: if you had shown these ads to no one, how many of those installs would you have gotten regardless?

You answer it by running a test with a holdout. You withhold your ads from a randomly chosen group, or from whole regions in a geo-based test, and you compare that group against the people who did see the ads. The difference between the two is the lift, the genuinely extra installs and actions your spend produced. Conversion lift tests and geo tests are the everyday tools here, and they turn a flattering attribution number into an honest one you can actually defend when you decide where the next tranche of rupees goes.

The honest snag is scale. Real incrementality tests need meaningful volume to reach a trustworthy result, and at small spend the signal is too noisy to read, so a clean user-level holdout is often out of reach. That is why geo tests become the practical fallback, splitting by region when you cannot split cleanly by person. It is more work than reading a dashboard, but it is the only real counter to attribution's built-in generosity.

Attribution tells you who to thank; incrementality tells you what you actually changed.