Warehouse-native experimentation platform, now part of Datadog, for statistically rigorous A/B testing on the metrics you already trust.
Eppo is an experimentation and feature-flagging platform that runs its analysis directly on your data warehouse, so test results are computed from the same revenue and product tables your BI stack reports on. Teams define metrics once, ship variants behind flags, and get sequential statistics and confidence intervals instead of eyeballed dashboards. In the revenue stack it sits between product and analytics: growth and pricing teams use it to test paywalls, packaging, and price points with the same rigor they would apply to any product change. Datadog acquired the company, folding experimentation into its observability suite.
Which of the capability map's modules Eppo covers — each links to the module's own page, with every tool that supports it.
| Module | Phase | Depth | Note |
|---|---|---|---|
| Win the Deal | |||
| Pricing Experimentation (A/B, Shadow Rating) | Configure & Quote | Core | Price and packaging tests measured against warehouse-defined revenue metrics. |
The warehouse-native model is the real separator: because Eppo reads metrics from Snowflake, BigQuery, or Databricks rather than its own event pipeline, experiment results reconcile with finance-grade numbers — which matters a lot when the metric under test is revenue per user. Its statistics engine (sequential testing, CUPED variance reduction) is also more rigorous than the built-in stats of most flagging tools.
Flag tools answer who sees what; Eppo's center of gravity is measuring what happened next. It ships flags too, but the product is the analysis layer — warehouse-native metrics, sequential statistics, and experiment reviews that a data team will actually sign off on.
Yes — that is one of its strongest fits. Because metrics come from the warehouse, you can test a paywall or price change against actual billed revenue rather than a proxy event, and the stats engine handles the long, noisy horizons that revenue metrics usually have.