What is Incrementality?

How incrementality testing actually works

You need a control group. The cleanest version is a geo test, where you pick matched sets of cities, keep spending in one set, stop entirely in the other, and compare total revenue across several weeks to find the lift.

Meta and Google both offer built-in lift studies that withhold your ads from a randomly selected slice of the audience, which is easier to run than a geo test although you are trusting the platform to grade its own homework. The simplest test costs nothing but nerve. Switch a channel off for a fortnight. Watch total revenue.

Where teams get incrementality wrong

Running tests too short. Two weeks is the floor, and for considered purchases it needs to outlast your typical buying cycle, otherwise you are measuring noise and calling it a finding.

Second, testing during a period that represents nothing. A holdout across Diwali, an end-of-season sale or a major cricket tournament tells you about that fortnight and precisely nothing about the rest of the year.

Third, badly matched geos. Comparing Gurugram against a tier-three city produces a difference driven by the cities rather than the ads, so match on historical revenue trend, order volume and category mix, and confirm the two groups tracked each other for months beforehand.

What good looks like in India

Start with whichever channel you most suspect. Retargeting and branded search are the usual candidates, because both reliably report spectacular ROAS and both mostly capture demand that already existed before anybody spent anything.

If a two-week holdout on branded search barely moves total revenue, you have just learned something worth a great deal of money. For brands with a strong Delhi NCR base, geo testing works well because you can hold out comparable cities without touching the core market. Run one a quarter.