Incrementality & MMM

Geo-lift experiments#

1. Under Experiments → New, pick the metric (installs, conversions, revenue), the treatment regions (countries or regions) and the pre-period (≥ 4 weeks).
2. WhichClick builds a synthetic control from the remaining regions (weighted to match the treatment's pre-period trend) and reports the calibration fit.
3. Run the campaign change in the treatment regions for the test window.
4. The result page shows observed vs. counterfactual, incremental lift with a confidence interval (block bootstrap), incremental CPI / ROAS and a recommendation.

Experiments read from the daily rollups, so any metric available in reports can be tested; spend from cost sync is used for iCPI.

Marketing-mix model#

MMM fits a ridge regression of the target metric on channel spend with:

• Adstock (geometric carry-over, decay per channel searched on a grid),
• Saturation (Hill transform),
• controls for seasonality (weekly / yearly Fourier terms), holidays and TV airings.

Outputs: contribution per channel over time, response curves, marginal ROAS, and a budget-reallocation suggestion under a fixed total. Runs are stored (MmmRun) with parameters and fit statistics so they can be compared.