Analytics & Tracking

How to Run a Geo Lift Test for eCommerce Advertising

A geo lift test compares regions with and without ads to measure true incremental sales. Learn how to design, run and read a geo test step by step.

TL;DR
  • A geo lift test turns ads off (or up) in some regions and compares their sales with similar regions where nothing changed.
  • It measures what a channel really adds to total sales, without relying on platform attribution or user-level tracking.
  • Good tests need comparable regions, enough sales volume, a clear duration and store data as the source of truth.

What is a geo lift test?

A geo experiment, often called a geo lift test, is a controlled experiment where the “unit” is a geographic region rather than a person. You change advertising in some regions (the test group) and keep it unchanged in others (the control group). Then you compare total sales between the two groups against what history predicted.

Because it uses total sales by region, a geo test does not depend on cookies, clicks or platform attribution. That makes it one of the most reliable ways to measure incrementality, especially as tracking becomes less complete.

For an overview of incrementality methods, see incrementality testing.

When should you use a geo test?

Geo tests are useful when:

  • You want to know if a channel (Meta, YouTube, branded search, Pinterest) adds real sales.
  • Platform reports disagree with each other or with GA4.
  • You are deciding whether to scale a channel significantly.
  • A channel is hard to track at user level, such as video, TV or podcasts.

They are less suitable when your sales volume is very small, when you sell only in a few regions or when regional differences are extreme.

Step 1: Define the question

Be specific. Examples:

  • “How many incremental orders does Meta prospecting add per month?”
  • “What happens to total sales if we pause brand search?”
  • “Does adding YouTube to our mix increase new customer orders?”

Decide the main metric up front: total orders, revenue, new customer orders or gross profit.

Step 2: Choose the regions

In the U.S., common units are states or designated market areas (DMAs). Requirements:

  • Each region must be targetable and excludable in the ad platform.
  • You must be able to see sales by region in your store data (usually by shipping address).
  • Test and control groups should have similar sales patterns over the past months.

Pick regions so the test group covers a meaningful share of sales, often 10% to 30%, but not so much that you risk too much revenue if the channel matters.

Step 3: Check the pre-test fit

Before the test, compare weekly sales for test and control regions over at least 8 to 12 weeks. They should move together closely. If one group is much more volatile, adjust the region selection.

Open-source tools such as Meta’s GeoLift package can help choose regions and estimate how big an effect the test can detect, based on your historical data. This is the geo version of a minimum detectable effect calculation.

Step 4: Run the test

Common designs:

Design What changes in test regions Good for
Holdout (go dark) Channel paused Measuring what a running channel adds
Scale up Spend increased Testing whether more spend pays
New channel Channel launched Testing a new platform

Guidelines:

  • Run for at least 3 to 4 weeks, often longer for products with long buying cycles.
  • Do not change other marketing differently between test and control during the test.
  • Avoid major holidays or sales unless that is what you want to test.
  • Document start and end dates and any unexpected events.

Step 5: Read the results

Compare actual sales in test regions with expected sales based on the control regions. The difference is the estimated lift.

Worked example (illustrative numbers): a brand pauses Meta ads in test states covering 20% of its sales for four weeks.

Metric Test regions (Meta paused) Expected based on control
Orders 3,420 3,800
Difference minus 380 orders
Lift attributed to Meta 380 ÷ 3,800 = 10% of orders in those regions
Meta spend saved in test regions $24,000

Meta caused about 380 orders in the test regions over four weeks, at about $63 of spend per incremental order. Meta’s own reporting for the same regions and period showed about 900 attributed purchases. The brand uses the test result to calibrate its view of Meta’s reported numbers and sets its targets accordingly.

What mistakes should you avoid?

  • Too short or too small: the effect gets lost in normal weekly noise.
  • Poorly matched regions: differences unrelated to the test look like lift.
  • Leakage: ads still reaching test regions through broad targeting or other campaigns.
  • Changing other channels at the same time: email or price changes in only one group distort results.
  • Using platform data as the outcome: use store orders by region.

How should you use the results?

  • Calibrate platform attribution: if Meta reports 2.4 times more conversions than the test shows, factor that into how you read its reports.
  • Feed results into marketing mix modeling if you use it.
  • Re-test periodically, because results change as your mix, creative and market change.

Checklist: geo lift test

  • [ ] Clear question and main metric chosen before the test
  • [ ] Regions targetable in the ad platform
  • [ ] Store sales available by region
  • [ ] Test and control regions matched on historical sales
  • [ ] Detectable effect estimated before launch
  • [ ] Test runs at least 3 to 4 weeks, avoiding unusual periods
  • [ ] No other marketing changes applied to only one group
  • [ ] Results based on store orders, not platform conversions
  • [ ] Platform attribution compared with test results
  • [ ] Re-test planned within 6 to 12 months

FAQ

How long should a geo lift test run?

Usually at least 3 to 4 weeks, and longer for products with long consideration periods or low weekly sales. The goal is to collect enough data for the effect to stand out from normal variation.

Can small stores run geo tests?

It is harder, because small regional sales numbers make results noisy. Smaller stores can test bigger changes, such as pausing a channel completely, or group regions into larger units.

Is a geo test better than a platform lift study?

Both have value. Platform lift studies use the platform’s own randomization at user level. Geo tests use your store data and work across any channel. Using both gives a fuller picture.

Turn this into a plan for your store

The Free Profit Audit scans your store, sizes your biggest profit leaks and builds a 90-day plan.

Free Profit Audit