ASAt its peak, the digital media operation I ran for Domino's India managed more than ₹7 crore a month across paid search, social, app, programmatic and CRM. At that scale, something changes quietly: the dashboards keep reporting confidence, but the decisions they support get riskier. Every platform attributes conversions to itself. Every channel looks marginally positive. And yet total order growth slows while the sum of "credited" contributions keeps rising.
The attribution illusion
Platform attribution answers one question: which touchpoints were present? It does not answer the question that matters: which touchpoints caused the order? In QSR, a large share of "performance" conversions are demand that already existed — a hungry customer searching "pizza near me" or opening the app. Media that intercepts them reports a flattering ROAS. Media that created the craving in the first place often reports nothing at all.
This is why budgets shifted purely on reported ROAS systematically starve demand creation and over-fund interception. For a few weeks the blended numbers look excellent. Then growth plateaus, because the engine that created demand was defunded.
If every channel claims credit for the same order, the sum of your platform reports will always flatter reality.
Geo-holdouts as the tie-breaker
The fix was not a better dashboard. It was a better question: what would have happened without this media? Geo-incrementality testing answers it by comparing regions where media activity changed with matched regions where it did not.
- Start with the channels carrying the most spend, not the ones that are easiest to test.
- Pre-register the success metric and the threshold before launch, so results cannot be argued away afterwards.
- Run tests long enough to cover full weekly and monthly cycles — QSR demand is daypart- and payday-sensitive.
- Treat platform lift studies as directional input, never as the verdict.
At Domino's, this discipline reshaped budgets across search, social and app campaigns. Some "hero" campaigns turned out to intercept demand that was already coming; some underrated ones were genuinely creating it. Reallocating on that basis supported +17% YoY digital order growth and a +25% lift in mobile app conversion over the period.
What we would tell any brand spending above ₹1Cr a month
You do not need a data science team to start. You need agreement on what counts as proof, and the courage to hold out a small slice of spend to earn it.
- Agree on one north-star metric — incremental orders, not clicks or platform ROAS.
- Map which campaigns are designed to create demand versus intercept it, and judge each on its own job.
- Run one holdout test per quarter per major channel, and let the learnings compound.
- Report both views to leadership: what platforms claim, and what tests prove.
Incrementality is not an academic exercise. At scale, it is the difference between renting demand and building it.
Figures in this article are drawn from the author's prior professional experience at Domino's Pizza India and are presented as learnings, not as client results.