A geo-based incrementality test compares outcomes in areas that see your ads with similar areas that do not, so you can measure the lift your marketing is actually creating.
I will partner with you from test design to final results, turning experimental outcomes into clear “go/no-go” decisions that maximise your budget’s true incremental impact.
Book a test planning session
What you’ll gain
A geo-based test gives you a clearer ground truth by separating genuine ad-driven lift from conversions that would have happened anyway.
Scale channels and tactics based on evidence, pause or redesign activity that adds little incremental value, and reduce wasted spend.
These results can also help validate future marketing mix modelling.
When to test and why
Tests are most useful when you need a clear causal read on a specific decision, especially in situations like these:
Validate new or underused channels before scaling
De-risk pausing a suspected low-value channel with a controlled holdout
Challenge platform attribution when the numbers do not match reality
These work as stand-alone scenarios but can also feed into marketing mix modelling when available to reflect more recent causal evidence.
How I work
Solutions are tailored to your business goals, martech stack and data maturity.
FAQ
Digital attribution is useful for tactical in-platform adjustment, but it often over-credits specific channels because it assigns credit based on observed user clicks or views. Tests establish causality, especially for upper-funnel or cross-channel questions, and they complement marketing mix modelling when available.
Most tests run for 3 to 6 weeks, plus a short cooldown period, so you can usually get results within a quarter. Duration depends on baseline volume, spend intensity, and the size of the expected effect.
Marketing mix modelling is broad and historical, ideal for planning and long-term optimisation. Testing is narrow and immediate, ideal for specific questions or never before used channels. Together they cover more ground.
We typically need outcome data (sales or conversions) by region or store, plus media delivery data for the same areas, and any known drivers that may influence results, such as promotions, stock issues, or major trading events.
The key requirement is that outcomes and media can be reliably mapped to the same geographic units. That can come from your existing setup or from a lookup table built from postcodes, store IDs, or other location markers.
If gaps exist, we will set expectations clearly and adjust the test design where needed.
Geo-based tests (geo-lift & geo-holdout) are the most frequently requested because they answer many common budget and channel questions. We also support other test mechanics when they are a better fit for the decision at hand, such as audience or cohort holdouts, time-based holdouts (switchback tests), and platform/retail media experiments where exposure can be controlled (e.g. ghost ads, sponsored products).
Use testing for specific decisions. Run a geo experiment when a specific decision needs a causal read that MMM cannot yet provide, for example a new or lightly used channel, flat spend that limits variation, or to verify surprising platform lift.
Like marketing mix modelling, geo and holdout designs do not need user-level data. They work well when cookies or customer-level identifiers are limited, provided we can track outcomes by region or cohort.
Yes. Data from these scenarios can be used to improve the accuracy of MMM frameworks. Specifically, they serve as calibration inputs for Meta’s Robyn and as Bayesian priors for Google’s Meridian.
When a result is inconclusive, we report the specific margins of uncertainty, allowing to adjust test design and determine if higher-intensity is required to achieve statistical significance. An inconclusive result often highlights limits in scale, data quality, or test power. That insight still protects budget by avoiding confident but unsupported scale-up, and informs what needs to change before testing again.
No. I work with open-source frameworks like GeoLift & CausalImpact whenever feasible, keeping work portable, transparent and easy to maintain & scale. This means no vendor lock-in, transparent code, and extensibility as your internal capability grows.
