Incrementality without geo-lift: the micro-conversion method.

Geo-lift incrementality tests are the gold standard. They are also expensive, long, and not feasible for most sub-scale brands. A cheaper method has quietly become standard at growth teams we work with: micro-conversion holdouts.

The method

Instead of turning off a channel in whole DMAs for two weeks (classic geo-lift), you hold out a percentage of your audience at the ad platform level for a continuous period, then measure the delta in a micro-conversion upstream of purchase.

Example. You run Meta retargeting. You create a 10% holdout audience that never sees retargeting ads for 6 weeks. Users enter the holdout via a deterministic hash, so the groups are comparable. You measure: add-to-cart rate, return-visit rate, branded search volume (via Google Search Console query filter on brand terms) during the test window, per group.

Why it works

Purchase is a rare event and needs high power to detect lift. Micro-conversions (add-to-cart, return visit, search) are 20 to 200x more frequent. You can detect meaningful lift in 2 to 4 weeks with holdouts as small as 5% of audience.

Where it fails

• If your micro-conversion is not predictive of purchase, you get a confident answer to the wrong question. Validate the predictive relationship first.

• Ad-platform holdouts can leak. Meta's holdout is at the ad account level and does not prevent impressions from other ad accounts in the same business manager. Verify with the platform's lift-study reporting.

• Seasonality and competitor activity confound short windows. Six weeks is usually enough; two weeks is usually not.

Real result from a client

Subscription DTC brand, Meta retargeting budget of $180k/month. Platform-reported ROAS: 6.4. Six-week micro-conversion holdout on 10% of retargeting audience. Branded search volume delta: -2% in holdout. Add-to-cart delta: -4%. Direct purchase delta: -11% (low power, but suggestive).

Interpretation: the incremental ROAS was roughly 2.1, not 6.4. Most of the reported ROAS was non-incremental (would have bought anyway). Budget was cut by 35%, spent on creative iteration instead. Brand-level revenue held flat.

What this does not replace

For board-level confidence on channel-level incrementality, MMM + geo-lift is still the right stack. Micro-conversion holdouts are the fast feedback loop in between.

Sources: internal client test on retargeting spend, Q1 2025; Meta Marketing Science GeoLift; internal methodology docs.