Statsig review

What Statsig is

Statsig bundles feature flags, experiments, product analytics, session replay and web analytics in one platform. Vijaye Raji, a former Facebook engineering leader, founded it in 2021 in Bellevue, Washington. OpenAI agreed to buy it for $1.1B in stock in September 2025 and made Raji its CTO of Applications. On May 5, 2026 Amplitude took on Statsig's brand, customers and platform, and said it will keep maintaining and developing Statsig across cloud and warehouse deployments. The original team stayed at OpenAI.

How experiments work

You install an SDK, wrap the change in a feature gate or experiment, and log the events you care about. Statsig records an exposure when your code checks a user against an experiment, then compares metrics between the groups. Server SDKs cover Node.js, Java, Python, Go, Ruby, C#, PHP, C++ and Rust, so you can test APIs, algorithms and infrastructure as well as UI.

The default analysis is frequentist with sequential testing, so p-values and confidence intervals stay valid if you check early. You can switch to SPRT, which allows unlimited peeking and can accept the null, or to a Bayesian mode that reports chance to beat control and expected loss. Pro adds CUPED, holdouts and multi-armed bandits.

For marketing sites there's a visual editor, which Statsig also calls Sidecar. It's a point-and-click tool for changing styles, text and calls to action, injecting scripts and setting up redirects on a live page, measured with the same stats engine.

Revenue and data

Statsig has no native Stripe connector. Revenue reaches it in three ways. You log purchase events from your own code, you import metrics from a data warehouse, or you use the RevenueCat integration, which sends billing and subscription events by webhook with no extra logging. For RevenueCat you have to set its appUserID to match your Statsig user ID. The Shopify guide has you write a custom pixel that forwards checkout_completed events with the order total. Warehouse Native, where Statsig analyzes data inside your own warehouse, is Enterprise only.

AI features

In January 2026 Statsig described a hypothesis advisor, AI summaries of results with a ship recommendation, natural language search across past experiments, and stale gate cleanup through GitHub pull requests. Its MCP server lets coding agents such as Claude Code, Codex or Cursor create gates and experiments. None of these write the variations. Your team still decides what to test and builds it.

Who it suits

Statsig fits product and engineering teams shipping changes inside an app who want flags and experiments in one place and the stats depth to trust the answer. It's a weaker fit for a small marketing team with no developer time, because anything beyond the visual editor needs code and event logging.

Before you commit

Ask how Amplitude plans to combine the Statsig and Amplitude Experiment roadmaps, and whether pricing will change at renewal. Check your volume against the 2M free and 5M Pro allowances, because exposures, logged events, imported metrics and custom metrics all count. And note that Pro covers one project per subscription.