Introducing GeoMatch

Today, we are excited to introduce GeoMatch, Cassandra’s new in-house tool for conducting geo-based incrementality experiments across various regions, markets, or DMAs.

Compared to open source alternatives, GeoMatch outperformed all major benchmarks.

• -50% budget required to reach stat. significance

• -30% in experiment time

• 30% less uncertainty with lift detected (tighter confidence intervals)

*Benchmark created a/b testing GeoMatch with Geolift on 30 different geo-experiments datasets

Why Geo-based testing?

Geo-based testing cuts through the noise with a simple, defensible idea: by A/B testing media spend on specific regions, are we able to detect a lift in the targeted regions?

From here, your incrementality assessment is straightforward. The change in your outcome variable compared to the change in budget is the true incremental drive of your media channel.

The result: You can confidently spend your money where the lift is detected, and this experimentation data is also essential to calibrate, trust and validate your MMM.

Why Geo Incrementality?

GeoMatch is Cassandra’s production‑scale platform for designing and analyzing geo‑incrementality experiments end‑to‑end—data ingestion, data cleaning, market matching, lift simulation, and incrementality measurement.

It’s a major upgrade over our previous Cassandra geo incrementality tool: faster guided setup, automated market pairing, built‑in data quality and power checks, and the ability to configure every parameter you may need.

Under the hood, GeoMatch builds upon the core synthetic control and market matching research to allow you to run experiments more cheaply than open source alternatives like Meta's Geolift.

Key Features that matter

Improved Fit

We built GeoMatch around producing strong counterfactuals. Using forward selection to pull in the most informative markets—and weighting to minimize pre‑treatment imbalance—GeoMatch routinely achieves tighter fit (lower L2 imbalance, MAPE) than stock open‑source solutions.

Comparing 30 datasets designed on open source solutions and GeoMatch, we were able to achieve around a 50% reduction in scaled L2 imbalance, and a confidence interval that is 30% tighter on average.

Reduced MDE

GeoMatch establishes a more precise counterfactual for each test market, effectively minimizing noise in the baseline. Our forward-selected market pool is intended to make sure we select the best control and treatment geos possible, which in turn decreases the standard error of the lift estimate. This means that less signal is required to achieve the same statistical power, giving you sensitive tests that work, while reducing the MDE by 50% on average.

Reduced Cost

Since GeoMatch constructs better fitting counterfactuals and can identify smaller lifts compared to previous solutions, this allows you the chance to (a) minimize the change in budget between test and control geographies, (b) limit the number of markets in the design, or (c) shorten the observation period—often achieving all three.

In fact, users can expect experiments that are around 50% cheaper on average compared to previous solutions. This reflects what we aimed for when developing GeoMatch: Empowering users to measure their media channels, while minimizing disruption to their business-as-usual.

All-in-One Planning

GeoMatch also features an Experiment Planner that organizes each study on a clear timeline—categorized as scheduled, in progress, or completed—allowing teams to monitor current activities, upcoming tasks, and anticipated results. This keeps your testing roadmap visible at a glance.

How  to use GeoMatch

1. You set the guardrails.
   Tell GeoMatch how long you can run the test, the budget you’re willing to move, and whether you plan to increase or decrease spend in test regions.

2. GeoMatch finds the best regions.
   The algorithm scans every eligible market, scores potential test–control pairs for similarity, and picks the combination that delivers the cleanest counterfactual within your time and budget limits.

3. It returns a ready‑to‑run design.
   You get a simple plan—which regions to adjust, which to hold steady, the recommended test length, and the expected outcome of the experiment.

4. Plan the experiment in your timeline
   One unified dashboard to plan your experiments over time

5. Launch the experiment, and get your analysis directly.
   Approve the design, start the experiment, and GeoMatch analyzes results so you can see incremental impact as soon as the experiment ends.

   
   

Do you want to try it?

Our Modelers are ready to show you how you can start measuring your media's incrementality using GeoMatch.

Click here to book a Demo