PostHog Experiments review

What it is

PostHog is an all-in-one product platform founded in 2020 by James Hawkins and Tim Glaser, who went through Y Combinator's W20 batch. The codebase is open source. Experiments is one product inside it, next to product analytics, session replay, feature flags, surveys, error tracking and a data warehouse. You pay for experiments through feature flag requests rather than a separate meter.

How experiments work

Creating an experiment creates a feature flag. You pick primary and secondary metrics, target cohorts or locations, and launch the variants. Your code checks the flag and renders the right version, on the client or on the server. PostHog supports A/B, A/B/n, holdout, fake door and redirect tests. Any event you already capture can become a funnel, mean or ratio metric, and you can add metrics from data warehouse tables.

For simple site changes there are no-code web experiments, built in the PostHog toolbar, where you edit text, CSS or HTML. PostHog labels them beta. It says they work well for simple layout changes and server-rendered apps like Next.js, and that single-page apps built with React, Vue or Angular can overwrite the changes when they re-render.

Stats

PostHog offers a Bayesian engine and a frequentist engine. The Bayesian engine reports results as probabilities, for example a 96% chance that variant B increases conversion rate. You can turn on CUPED to reduce noise using each user's behavior before exposure.

Revenue

PostHog's Stripe connector syncs charges, customers, invoices, products, subscriptions and more into its data warehouse. PostHog describes webhook syncing as the cheapest mode to run and the only one that reliably captures changes to existing rows. To use that data in an experiment, you add a data warehouse metric and set a timestamp field, a warehouse join key such as the Stripe customer ID, and the PostHog field it should match. Then you can measure mean revenue per variant. For revenue analytics, Chargebee, Polar and RevenueCat are listed as coming soon.

AI

PostHog AI can create experiments from plain English. It sets split percentages and goal metrics, creates the flag with the right variants, configures targeting, then summarizes results and suggests whether to ship, extend or stop. The PostHog MCP server lets coding agents such as Claude Code, Cursor or Codex create, launch and end experiments from the editor. Neither writes the variant code or copy, so you still decide what to test and build it.

Who it suits

It suits product engineers and startups who already send events to PostHog and want experiments without another vendor or SDK. It suits marketing teams less well, because the no-code editor is in beta and revenue metrics need warehouse joins.

What to watch

Flag checks count as flag requests, and the $feature_flag_called events PostHog uses for exposure count as analytics events. Each has its own free allowance and price, so set a billing limit per product, which PostHog supports.