Shamanth Rao

Incrementality models based on MMMs are a massive part of marketing measurement, especially on iOS, where measurement is fundamentally broken. But they are not a silver bullet.

 Use incrementality models alongside your traditional “deterministic” attribution (or SKAN on iOS). Both provide a different side of the picture. This allows you to identify where to re-test, double down or approach things differently.

Be intentional about the “experiments” you run and the impact you measure from these. For example: if you want to measure the impact of a channel like Facebook or Taboola, make a big or significant enough change in the CPA campaign and try not to make other changes.

You must be mindful that a D1 ROAS but poor retention doesn’t mean: your ad is great, but your game is not. The wrong ad may attract the wrong people!

MMM models do not naturally account for creatives, so bake creative changes into your incrementality experimentation cadence. For example: if a creative concept is promising in early testing (e.g., on Android), set it up along with its variants in a brand new campaign with a significant budget to measure the impact.

Algorithms don’t like local apps. When you try to advertise for an app in a specific location, CPMs go through the roof. These local apps might be able to explore offline advertising (bus billboards, radio, etc.).

To avoid concluding that something is not performing when it could be due to the platform still being in the learning phase (e.g., Meta, Google UAC, programmatic), give advertising platforms at least a week after changes.

You’re really fighting for people’s attention. If you have a product showing metrics that make you believe in it, it makes sense to re-engage users (via email, paid ads, etc.) on D0 (session 2), D1, D3, first-weekend post-install, and then before each Live Op.