PROVIDER Β· Open-source self-hosted experimentation

GrowthBook

Emerging around 2020 in the US, GrowthBook is an open-source, self-hostable experimentation and feature flag platform computing metrics in your own warehouse so data stays in-house without per-traffic fees.

Testing & Optimization United States 2026.11.14
Testing & Optimization United States Open-source self-hosted experimentation
4.2pts

Pros

4 items
Open source and self-hostable with the core free and no per-traffic or experiment caps
Computes experiment metrics in your own warehouse so data need not leave your infrastructure
Feature flags target users, attributes, and percentages for progressive rollout
Both Bayesian and frequentist statistical engines to match team preference

Cons

4 items
Self-hosting requires maintaining the service, SDKs, and connections, costing more than SaaS
Analysis depends on your data and metric definitions, so dirty data yields wrong answers
Smaller docs and community than commercial platforms, with little Chinese material
Enterprise features such as SSO, audit, and support sit in paid editions

Product Information

growthbook
Category Testing & Optimization
Company GrowthBook
Service Region United States
Official Website https://www.growthbook.io/
Recommendation Score 4.2 pts
Published 2026-11-14

Detailed Notes

Overview

GrowthBook emerged around 2020 and is based in the United States, positioned as an open-source experimentation and feature flag platform with an open core you can self-host and paid hosting plus enterprise features. It analyses against your own data warehouse rather than ingesting all event data into the vendor.

For Testing & Optimization work the key advantages are keeping data in-house and avoiding per-traffic fees, since experiment metrics compute in your own database or warehouse, sidestepping third-party scripts and data residency problems. Teams with an existing data pipeline find that architecture a better fit than SaaS.

The limits are deployment and statistics. Self-hosting means maintaining the service, SDKs, and data connections, analysis quality depends on your data, and unclear metric definitions still produce wrong conclusions. Documentation and community are smaller than commercial platforms with little Chinese material.

Core Strengths

  • Open source and self-hostable with the core free and no per-traffic or experiment caps
  • Computes experiment metrics in your own warehouse so data need not leave your infrastructure
  • Feature flags target users, attributes, and percentages for progressive rollout
  • Both Bayesian and frequentist statistical engines to match team preference

Product Ecosystem

Experiments and statistics

A/B, multivariate, and staged experiments with switchable Bayesian or frequentist engines. Confirm metric definitions match data sources first, or the same experiment yields contradictory conclusions.

Feature flags and SDKs

Multi-language SDKs with rollout rules configured centrally. Self-hosted operations hinge on version upgrades and SDK compatibility, so pin versions and roll out gradually.

Drawbacks

  • Self-hosting requires maintaining the service, SDKs, and connections, costing more than SaaS
  • Analysis depends on your data and metric definitions, so dirty data yields wrong answers
  • Smaller docs and community than commercial platforms, with little Chinese material
  • Enterprise features such as SSO, audit, and support sit in paid editions

Use Cases

  • Teams with existing data pipelines (β˜…β˜…β˜…β˜…β˜…) - Metrics compute in your warehouse without data leaving
  • Engineering teams needing flags (β˜…β˜…β˜…β˜…β˜†) - Open source and self-hosted with no traffic fees
  • Budget-sensitive growth testing (β˜…β˜…β˜…β˜…) - Core is free, with cost in servers and labour
  • Marketing teams without engineers (β˜…β˜…) - Self-hosting and data connections set a high bar

Pricing

Plan Price Scope
Open-source self-hosted $0 Free core on your own servers
Cloud Per seat Managed with enterprise features
Enterprise Custom quote SSO, audit, and support

The open-source edition is free with cost in servers and operations, and cloud bills per seat. Include warehouse query cost, since computing metrics in your own warehouse incurs query charges that can exceed SaaS subscriptions on large tables.

FAQ

  • How is it different from Optimizely? GrowthBook is open source and self-hosted with data staying in your warehouse, while Optimizely is enterprise SaaS with more features and higher cost. Strict data residency requirements favour GrowthBook.

  • Which statistical engine? Bayesian for teams that peek continuously and decide at will, frequentist for strict hypothesis testing with fixed samples. Stay consistent, since mixing makes historical conclusions incomparable.

  • Can it query production directly? Use a read replica or warehouse rather than the primary. Metric queries are heavy and querying production degrades response times.

Visit GrowthBook Website β†’ https://www.growthbook.io/
Visit Website β†—