GrowthBook Starter plan: How to run experiments and feature flags for free

GrowthBook Starter gives a small team a real feature flag and experimentation workflow without charging for end-user traffic.

The current cloud Starter plan costs $0, requires no credit card, and supports up to 3 GrowthBook users in 1 project. It includes unlimited feature flags, unlimited experiments, and unlimited traffic. Teams can bring an existing data warehouse or use GrowthBook's managed warehouse path to begin collecting events.

That makes Starter useful for more than a demo. An engineer can control a production release with a flag, a product manager can define the hypothesis and decision, and a data teammate can analyze the result—if 3 seats and the base feature set fit the team.

This guide explains what is included, what is not, how to launch a first flag and experiment, and when cloud Pro or self-hosting becomes the better choice. Pricing and packaging can change, so verify the GrowthBook pricing page before making a purchase decision.

What the free Starter plan includes

As of July 2026, GrowthBook lists:

“Unlimited traffic” means GrowthBook does not price the core Starter plan by every user who encounters a flag or experiment. Data ingestion, CDN, managed warehouse, and other services can have separate included allowances or overage terms. The current free-plan page lists 1 million managed-warehouse events per month for Starter and 2 million for Pro. Confirm the live limits for the ingestion path you choose.

Independent directories corroborate the basic plan shape. G2's GrowthBook pricing listing shows the free 3-seat cloud tier with unlimited flags, experiments, and traffic, while TrustRadius lists both the $0 cloud Starter and $0 open-source self-hosted options. Official GrowthBook pages remain authoritative for current packaging.

What you can build on Starter

Feature flags

Use GrowthBook feature flags to:

• Hide incomplete code in production.
• Release to employees or an allowlist.
• Target attributes or segments.
• Roll out to a percentage of users.
• Return boolean, text, number, or JSON values.
• Keep a kill switch for a risky path.
• Assign a controlled experiment.

GrowthBook SDKs load feature definitions and evaluate rules in the application. This avoids a network call for every flag check. The application still needs an explicit fallback and tested behavior when configuration is unavailable.

Base experimentation

Starter includes unlimited A/B tests and feature flag experiments. Teams can define metrics, analyze control and treatment, inspect results, and use the same assignment for release and measurement.

GrowthBook experimentation supports code-based feature tests. The free-plan matrix should be checked for advanced methods and interfaces. Visual Editor, URL split tests, multi-arm bandits, Safe Rollouts, and some advanced statistics or workflow capabilities are listed on paid tiers.

Unlimited experiments does not mean every test is free to operate. Warehouse queries consume compute, managed event ingestion has allowances, and poorly designed tests consume traffic and team attention.

Existing warehouse metrics

GrowthBook's warehouse-native architecture can query data from supported systems instead of requiring all raw user behavior to move into a new analytics silo. Teams can reuse definitions for conversion, revenue, retention, latency, support contacts, or other outcomes.

The data source documentation covers connection and permission patterns. Use a restricted service account, validate generated queries, and understand warehouse compute before scheduling frequent refreshes.

Managed warehouse

A team without a warehouse can use GrowthBook's managed option and send events through supported collection paths. The current get-started page describes a ClickHouse-based managed warehouse and allows later migration to the team's own data system.

ClickHouse documentation provides background on the column-oriented database, but GrowthBook manages the service details for cloud users. Treat event naming, identity, and metric definitions as production data design even when infrastructure setup is simple.

What Starter does not replace

A production release process

A feature flag makes exposure reversible; it does not replace code review, automated testing, security review, observability, or rollback planning. Test both flag states and failure defaults before production.

Clean event data

An experiment engine cannot repair missing exposure, unstable identities, duplicate events, or ambiguous metrics. Run an A/A test when validating a new pipeline and reconcile counts with source data.

Enterprise governance

Starter is designed for individuals and small teams. Current Enterprise packaging includes capabilities such as SSO and SCIM, approval workflows, custom environments, advanced access control, exportable audit logs, higher service commitments, and dedicated support.

If production change governance is a requirement, compare the required paid tier rather than trying to approximate access control through a shared account. A seat represents an individual login; do not share credentials to stay under 3 users.

Paid advanced workflows

The current plan matrix lists capabilities such as the visual editor, multi-arm bandits, Safe Rollouts, advanced permissioning, dashboards, and premium support on Pro, with additional enterprise controls on Enterprise. Check the live matrix because capabilities can move as the product evolves.

How to create your first free feature flag

1. Create a Starter account

Use the GrowthBook free signup. Name one project around the product or application, not a temporary experiment.

Invite only the people who need to manage configuration and analysis. Application end users do not consume seats.

2. Choose an SDK

Select the server, web, mobile, or edge SDK that owns the behavior. The GrowthBook SDK documentation includes installation and initialization guides.

Start server-side if both options fit. Sensitive targeting stays out of browser payloads, and the first integration can use a stable account or user identifier from a trusted context.

3. Define the flag contract

Record:

• Stable key.
• Owner.
• Purpose and type.
• Disabled and enabled behavior.
• Safe fallback.
• Eligible audience.
• Rollout stages.
• Cleanup date.

Use a small reversible feature, not billing, authorization, or an irreversible data migration.

4. Implement both states

Wrap SDK access in a domain-facing function and test control, treatment, and missing-configuration behavior. Deploy treatment code disabled.

5. Verify internal targeting

Target an employee account. Confirm the variation, configuration version, and application behavior. Test rollback while a normal request is active.

6. Ramp deliberately

Expand to a small percentage, monitor errors and latency, then proceed through defined stages. Starter supports percentage rollouts, while automated Safe Rollout behavior is a paid capability according to the current matrix.

How to run a free A/B test

1. Write the hypothesis

Name population, treatment, outcome, and mechanism:

For new workspace administrators, a setup checklist will increase successful data connection within 24 hours because it clarifies the remaining steps.

2. Define metrics

Choose one primary metric, a few guardrails, and diagnostic metrics. Use GrowthBook metric definitions against the managed or connected data source.

Specify unit, window, exclusions, and missing-data handling. Avoid choosing the winner from whichever metric looks positive.

3. Attach an experiment rule

Create a feature rule that assigns eligible units to control and treatment. Use a stable user or account identifier and log exposure when the treatment can affect behavior.

The feature flag experiment documentation explains the relationship between the flag variation and experiment analysis.

4. QA assignment and data

Test eligibility, both variants, repeat assignment, event values, exposure, metric joins, and rollback. Confirm that excluded traffic does not enter analysis.

5. Run to the planned decision point

Estimate sample and duration before launch. Use the statistical methods available on the selected plan and do not stop an ordinary fixed-horizon test because an early movement looks exciting.

6. Decide and clean up

Review data integrity, effect and uncertainty, practical significance, and guardrails. Ship, revert, or plan a follow-up. Remove the temporary branch after the decision.

Cloud Starter versus open-source self-hosting

The GrowthBook repository includes Docker Compose instructions and license details. A historical Hacker News launch discussion captures the underlying model: open-source software for teams that want to run it, plus SaaS for teams that do not want to operate it.

Self-hosting is not a way to make operations free. Budget infrastructure, backups, upgrades, vulnerability response, authentication, and on-call ownership.

When to upgrade

Move to Pro when:

• The team needs more than 3 users.
• Visual website editing or split URL tests are required.
• Multi-arm bandits or paid advanced analysis fit the program.
• Safe Rollouts and advanced permissioning matter.
• Custom dashboards and premium support add value.

The current price is $40 per seat per month for up to 50 users. Model seats and managed data separately.

Evaluate Enterprise when:

• SSO or SCIM is required.
• Production changes need formal approvals and advanced access control.
• Audit export and custom environments are required.
• Large datasets need specialized optimization.
• A contractual SLA, dedicated support, or training is necessary.

Upgrade because the operating requirement exists, not because experiment traffic grows. GrowthBook's seat-based model is designed so end-user exposure does not automatically move the core plan into a higher traffic tier.

Get useful evidence before paying

The best use of Starter is a complete trial, not a dashboard tour. Run one real flag through design, code, internal exposure, staged rollout, measurement, decision, and cleanup. Connect a metric source the team trusts and test failure behavior, including missing configuration and an emergency reversal after production exposure begins under realistic production workload conditions.

Community discussions about free A/B testing tools regularly mention GrowthBook's open-source path. Your own proof of concept should determine whether the workflow, data architecture, and collaboration model fit.

Create a free GrowthBook account, invite the 2 collaborators who complete the release-and-measurement loop, and use the first experiment to establish the process you would scale on a paid plan.