On this page
Meridian GeoX
Meridian GeoX
Meridian GeoX is Google's open-source geographic incrementality testing framework, designed to measure the causal lift of advertising spend using geographic market-level experiments rather than user-level tracking. GeoX is integrated with Google Meridian (Google's open-source MMM) so geo-experiment results feed directly into the MMM as Bayesian calibration priors. It reached general availability globally on September 9, 2026. (ppc.land, 2026-09; developers.google.com, 2026-09-09)
Key facts (as-of 2026-11-02)
- General availability: GeoX reached GA globally on September 9, 2026 (ppc.land, 2026-09-09). Meridian 2.0.0 was released September 2, 2026.
- Open-source: free Python package; published on GitHub alongside the Meridian MMM repo. (ppc.land, 2026-09)
- Primary estimator: Time-Based Regression (TBR) — a time-series regression approach, distinct from Meta GeoLift's Augmented Synthetic Control (ASC). (bellstatistics.com, 2026)
- JAX backend: computationally faster than the prior trimmed_match/matched_markets repos; Google vendor claims >94% faster design generation and >31% budget savings for large advertisers vs. unnamed open-source alternatives (unverified, no sample sizes stated — confidence: low). (ppc.land, 2026-09)
- Python 3.11+ required; daily time-series data required (weekly not supported). (developers.google.com, 2026-09-09)
Default test parameters (as-of 2026-09-09)
Per Google's developer documentation (developers.google.com/meridian, 2026-09-09):
| Parameter | Default |
|---|---|
| Test type | Holdback |
| Estimator | TBR (Time-Based Regression) |
| Assignment | Stratified |
| Treatment cells | 1 |
| Alpha | 0.10 |
| Power | 0.80 |
| Test direction | Two-sided |
| Candidate designs generated | 10 |
| Candidate market splits searched | 100,000 |
| Minimum geo units | 10 (50+ recommended) |
Test design types
GeoX supports three test design types (ppc.land, 2026-09):
- Holdback: a new campaign launches only in treatment regions; control regions receive nothing
- Go-dark: an existing channel pauses only in treatment regions; control continues
- Heavy-up: spend scales up in treatment regions vs. unchanged control
MMM integration
- GeoX output feeds Google Meridian as Bayesian priors; Meridian recommends which channels would benefit from testing, then acts as the budget allocation layer on top of geo experiment results. (ppc.land, 2026-09)
- When translating GeoX results into Meridian priors: convert experiment ROI point estimate + standard error into Bayesian prior mean and SD; widen the confidence interval when the test window is short relative to years of MMM data; if the experiment measures online sales only but MMM tracks total sales, shift the prior's mean upward to account for offline effect. (Google Analytics YouTube channel, 2026-04-01, via pulse.bot — third-party summary; confidence: low)
Vendor landscape context
Per Sellforte vendor comparison (self-authored, AI-scored, September 2026 — low confidence; Sellforte is a competing vendor): GeoX scored 5.8/32, lowest of all vendors evaluated, primarily due to no hosted interface, managed onboarding, or enterprise support. NMS Editorial (2026-05-05) recommends running GeoX alongside an independent vendor for one quarter as a credible cross-check tool.
Structural concerns
Per NMS Editorial (2026-05-05): a structural concern with GeoX — Google built a tool that may grade Google's own advertising lift; while the code is auditable, routing through Meridian, Google Cloud (Meridian Studio) centralises measurement with one vendor. Amazon's MMM API also reached GA in 2026, meaning both Google and Amazon simultaneously operate as vendors of MMM tooling that evaluates their own ads. (as-of 2026-05-05; Amazon GA claim not independently verified)
Key terms
| Term | Meaning |
|---|---|
| TBR | Time-Based Regression — GeoX's primary effect estimator; contrasts with ASC used by Meta GeoLift |
| Holdback | Test design where new campaign launches only in treatment; control receives nothing |
| Go-dark | Test design where existing campaign pauses only in treatment; control continues |
| Heavy-up | Test design where spend scales up in treatment; control unchanged |
| JAX | Google's array computation library used for GeoX's computational backend |
| Meridian Studio | Google Cloud enterprise product wrapping Meridian/GeoX (pricing and GA date not confirmed as-of 2026-11-02) |