How to build an always-on incrementality testing calendar

How to build an always-on incrementality testing calendar

The question: one-off lift tests answer one question and expire. How do you turn incrementality from a project into a program that continuously feeds your models?

The build, step by step:

— Inventory your channels and rank them by 'credit fragility' — how much you suspect last-click or platform reporting overstates each. Test the most fragile first.
— Assign each a test method: geo holdouts for channels without user-level tracking (TV, podcasts, broad social), ghost/conversion-lift for auction-based platforms.
— Stagger tests on a rolling quarterly calendar so you're never dark on all channels at once and never running so many that markets contaminate each other.
— Maintain a results ledger: channel, method, measured lift, confidence interval, date. Lift decays — last year's number is a stale prior.
— Feed each result back as a calibration point for your MMM or DDA.

The nuance: incrementality is not a constant. It shifts with saturation, seasonality, creative fatigue, and competitor pressure. A program re-measures; a one-off pretends the world stands still.

Bottom line for practitioners: schedule incrementality the way you schedule financial close — recurring, calendared, and feeding everything downstream. Continuous calibration is what separates a model you trust from a model you hope is right.