Statsig review: features, pricing, and AI capabilities

Platform overview

Statsig's optimization core is still Autotune, a Bayesian multi-armed bandit that shifts traffic toward the best-performing variant, plus contextual bandits that rank variants per user from attributes.

Since December 2025, AI has spread across the experiment workflow: summaries that make an explicit ship recommendation, a hypothesis advisor that critiques hypotheses during setup, contextual AI descriptions, AI knowledge-bank search and AI stale-gate cleanup.

Agent surfaces have grown fastest. There's an OAuth-authenticated MCP server usable from Codex, Cursor and Claude Code, a Statsig ChatGPT app, and an agent-skills repository. MCP coverage spans metrics tools, segments and layers, parameter store, autotune creation, version history and read-only access modes.

Statsig also ships an AI Evals product for offline and online LLM evaluation alongside the experimentation platform.