Analytics
Statsig - the practical guide.
Statsig is a unified platform for feature flags, experimentation, product analytics and session replay - built by ex-Facebook engineers and modelled on the experimentation stack used inside Meta. It's gained serious traction with PLG SaaS, fintech and consumer apps that want LaunchDarkly + Amplitude + Optimizely capabilities without paying three separate enterprise bills.
What Statsig does
The core covers feature gates (boolean and JSON flags with targeting rules), dynamic config, A/B and multivariate experiments with built-in statistical analysis, product analytics (events, funnels, retention, user journeys) and session replay. Experiments tie directly to product analytics so you can see the downstream impact of any rollout.
Warehouse-native mode lets large customers run analysis directly on Snowflake, BigQuery or Databricks without copying data into Statsig. Native integrations cover Segment, mParticle, Slack, GitHub, Linear and the modern dev and data stack, with SDKs across web, mobile, server and edge runtimes.
Who it's for
Product, engineering and growth teams at PLG SaaS, fintech, consumer apps and AI products who run experiments as a routine part of shipping. Particularly strong for engineering-led organisations that want statistical rigour and a unified platform rather than a stitched stack.
Pricing, in rough terms
Generous free tier covering up to 1 million events per month and unlimited feature flags - one of the most usable free tiers in the category. Paid tiers (Pro, Enterprise) are usage-based and quote-driven, with warehouse-native pricing structured separately. Most early-stage teams stay on the free tier well past initial traction.
When Statsig is the right fit
The right call when experimentation is a discipline rather than an occasional project, you want flags and analytics tied together, and you'd rather not run LaunchDarkly + Amplitude + Optimizely separately. Also a strong fit for AI product teams measuring model and prompt changes. A weaker fit for marketing-led optimisation on landing pages (VWO or Optimizely Web are more marketer-friendly) or for teams without engineering involvement in experiments.
Watch-outs
Experimentation done badly is worse than not done at all - invest in basic stats literacy and pre-registered hypotheses before scaling rollouts. The platform is engineering-flavoured; non-technical PMs may need help with setup and analysis. Free tier limits are generous but watch event volume on high-traffic apps, where the step up to paid can be steep.