Incrementality.

Incrementality, properly understood

Incrementality is measured through a holdout or control experiment: withhold a randomized slice of the eligible audience from a campaign, then compare the conversion or booking rate of the exposed group against the holdout group — the difference is incremental lift, and whatever the holdout group converted anyway represents the baseline that would have happened regardless of the campaign. This requires a properly randomized holdout, enough sample size for statistical power, and a clean measurement window on both sides. The data itself is split across two systems: the ad platform knows who was exposed, and the product or finance system knows what actually converted — reconciling the two, not reading either alone, is what produces the incrementality number.

Getting statistical power for an incrementality test is harder in smaller Gulf markets like Bahrain or Kuwait than at Saudi or UAE scale, since the sample sizes available are simply smaller — a holdout design that works cleanly at large scale can be underpowered in a smaller market and produce a noisy, inconclusive read. Ramadan also shifts baseline conversion so dramatically that an incrementality test run outside Ramadan tells you little about what will happen during it — seasonal windows generally need their own dedicated test, not an extrapolation from an off-season result.

Last-click attribution systematically overstates incrementality, because it credits every exposed converter, including the people who would have converted anyway — that gap between attributed and incremental conversions is the entire reason the discipline of incrementality testing exists. Conflating the two is one of the biggest reasons marketing budget drifts toward channels that simply catch people at the bottom of a funnel they were already moving through, rather than channels that actually create new demand. Geo-based holdouts carry their own risk too: media spillover between test and control regions (people traveling, ads served across a border) can quietly contaminate the comparison if the regions aren't drawn carefully.

Use incrementality results to periodically recalibrate platform-reported attribution and to sense-check or reweight a Marketing Mix Model's channel coefficients — the two methods validate each other, and neither should be trusted in isolation over the long run.

Incrementality testing is resource-intensive — it requires holding back spend from a control group, which has a real opportunity cost, and it takes time to reach statistical significance — so most teams can't run a formal holdout on every channel continuously. A common compromise is running periodic incrementality tests (quarterly or per major campaign) on the channels with the largest budget or the most attribution ambiguity, and using those results to calibrate trust in that channel's platform-reported numbers for the periods in between, rather than treating every dollar as requiring its own live experiment.