Datadog Experiments review: features, pricing, and AI capabilities

Platform overview

Datadog Experiments went GA in April 2026. Randomization comes from Datadog Feature Flags; analysis runs on RUM, Product Analytics, or warehouse metrics in Snowflake, BigQuery, Redshift or Databricks.

Stats options are sequential frequentist, fixed-sample frequentist and Bayesian. CUPED variance reduction is on by default, and multiple testing correction is optional. Protocols save reusable templates covering metrics, randomization, duration and analysis plan. Built-in diagnostics flag sample ratio mismatch, unreliable or zero-data metrics, and exposure imbalance between variants.

The main AI surface is agentic. The Datadog MCP Server's experiments toolset lets AI clients list, create, start, conclude and cancel experiments, pull diagnostics and metric definitions, and segment results by any assignment property.

Built on Datadog's 2025 acquisition of Eppo; the standalone Eppo product is being folded in. [Updated 2026-09-15: removed prior Eppo-era claims about Contextual Bandits, Auto-Generate Explore Charts and automated Slack alerts. None appear in current Datadog Experiments docs.]