When a marketing team tells me they “can’t run an incrementality test because we can’t withhold ads from individual users,” I usually have the same answer: you don’t have to. You can hold back entire regions instead. That is the core idea behind geo lift testing, and it is one of the most underused measurement techniques in SaaS marketing.
I have run geo experiments for companies that could never have managed user-level holdouts — privacy constraints, platform limitations, or simply not enough volume per user. Geo lift testing sidesteps all of that by treating geography as the unit of randomization. In this guide I will explain how it works, when to use it, and how to avoid the design mistakes that quietly invalidate results.
What Is Geo Lift Testing?
Geo lift testing is a method for measuring the causal impact of marketing by splitting geographic regions into a test group and a control group. The test regions receive your campaign; the control regions get nothing. You then compare conversion outcomes between the two groups to isolate the incremental lift your marketing actually produced.
It belongs to the broader family of incrementality testing methods. The difference is the unit of randomization. A standard holdout test withholds ads from a random sample of individual people. A geo lift test withholds them from entire places — cities, states, or designated market areas.
“In God we trust. All others must bring data.”
— W. Edwards Deming, statistician and quality-management pioneer
Why Geo Testing Instead of User-Level Holdouts?
User-level holdouts are the gold standard when you can run them cleanly, but real-world constraints often make them impractical. Geo testing solves several problems at once.
| Problem with user-level holdouts | How geo testing helps |
|---|---|
| Privacy rules limit individual tracking | Works on aggregate regional data — no user-level identifiers needed |
| Cross-device and cross-channel leakage | A whole region is either exposed or not, so leakage is contained |
| Platforms don’t support clean holdout splits | You control exposure simply by where you do or don’t run campaigns |
| Offline conversions are hard to tie to users | Regional sales totals capture both online and offline impact |
This makes geo lift testing especially valuable for measuring upper-funnel and brand activity, where the path from impression to conversion is long and messy. It is also the practical choice for any channel where you cannot reliably exclude individual users.
How a Geo Lift Test Works, Step by Step
Step 1: Choose Comparable Regions
The foundation of a valid test is picking regions that behave similarly. You want test and control markets that have matching baseline conversion trends before the experiment starts. If your test regions were already growing faster than your control regions, you will mistake that pre-existing trend for campaign lift.
I look at historical conversion data and select region pairs whose trends move in parallel. The more closely they tracked each other before the test, the more trustworthy the result.
Step 2: Establish a Baseline Period
Before turning anything on, measure both groups for a stable baseline window — usually a few weeks. This baseline lets you confirm the regions really do behave alike and gives you a reference point for what “no change” looks like.
Step 3: Run the Campaign in Test Regions Only
Launch your campaign exclusively in the test regions while keeping control regions dark. Geo targeting in Google Ads, Meta, and most ad platforms makes this straightforward. Run it long enough to cover your typical sales cycle plus a buffer.
Step 4: Measure the Lift
After the test window, compare the change in conversions between test and control regions, adjusted for their baseline difference. The gap is your incremental lift — the conversions attributable to the campaign rather than to background trends. Open-source tools like Google’s CausalImpact and Meta’s GeoLift library handle the statistical modeling.
Common Geo Lift Design Mistakes
Geo tests fail quietly. The numbers come out, they look plausible, and nobody realizes the design was broken. These are the mistakes I see most.
- Mismatched regions. Pairing a fast-growing metro with a stagnant rural market guarantees a misleading result. Match on baseline trend, not just size.
- Too few regions. Splitting into just two regions leaves you exposed to one local anomaly — a weather event, a competitor promotion, a single big deal. Use many regions on each side.
- Spillover between regions. If your test and control markets share TV coverage, commuters, or media overlap, exposure leaks across the boundary and dilutes the measured lift.
- Window too short. Ending the test before your sales cycle completes underestimates lift, because conversions are still in flight.
- Ignoring seasonality. Running test and control across different calendar periods, instead of simultaneously, lets seasonal swings masquerade as campaign effects.
Tip: The biggest single safeguard is running test and control at the same time. Geography handles the “who,” but simultaneity handles the “when.” Skip it and seasonality will quietly corrupt your numbers.
When Geo Lift Testing Is the Right Choice
Geo testing is not the answer to every measurement question. It shines in specific situations:
- Measuring brand or upper-funnel campaigns where individual attribution is hopeless.
- Validating channels with heavy offline conversion components, like demos that close over the phone.
- Operating under privacy constraints that rule out user-level tracking.
- Testing channels you cannot split at the user level, such as connected TV, podcasts, or out-of-home.
For on-property optimization — landing pages, emails, onboarding — you are better off with simpler experiments. I cover that distinction in detail in my guide on measuring true marketing lift.
Geo Testing and Your Attribution Model
Geo lift results are one of the best reality checks for your attribution setup. Your attribution model assigns credit based on observed touchpoints, but it cannot prove causation on its own. A geo test can.
When a channel’s geo-measured lift comes in far below the credit your attribution model gives it, that is a signal the model is over-crediting an easy-to-track channel. Use the geo result to recalibrate, and your day-to-day attribution reporting becomes more honest.
Frequently Asked Questions
How many regions do I need for a geo lift test?
More is better. Tests with only two or four regions are fragile because a single local event can swing the result. Aim for at least a dozen comparable regions split across test and control groups so that random local noise averages out and the measured lift reflects your campaign rather than one anomaly.
How long should a geo lift test run?
Run it for at least one full sales cycle plus a buffer, which for many SaaS businesses means six to twelve weeks. Ending early underestimates lift because conversions are still completing. Include a baseline period beforehand to confirm your test and control regions truly behave alike.
Can small companies run geo lift tests?
Yes, and geo testing is often more accessible than user-level holdouts for smaller advertisers. Because you withhold spend by region rather than from individuals, you avoid privacy and platform limitations. You still need enough total conversions across regions to reach statistical significance, so very low-volume businesses may struggle.
What tools support geo lift analysis?
Open-source options include Google’s CausalImpact and Meta’s GeoLift library, both of which model the counterfactual statistically. Several ad platforms also offer built-in geo lift or conversion lift studies. The analysis itself is straightforward once you have clean regional conversion data and well-matched test and control markets.
The Bottom Line
Geo lift testing gives you a credible way to measure causal impact when user-level holdouts are off the table. It trades the precision of individual randomization for the practicality of regional randomization — and for brand, offline, and privacy-constrained channels, that is a trade worth making.
Get the region matching right, run test and control simultaneously, and give the experiment enough time and volume. Do that, and you will have something most marketing teams never get: real proof of what your campaigns actually caused, not just what your dashboards happened to record.
Lukas Reinhardt
10+ years building analytics stacks for SaaS companies. Every tool reviewed here is personally tested — no sponsored content, no affiliate bias.
More about me →
