LaunchDarkly review
- What it is
- LaunchDarkly is a feature management platform for engineering teams, with experimentation, observability and AI config tools built on its feature flags.
- Best for
- Engineering organizations that release through feature flags and want experiments on the same flags, from a free plan up to enterprise governance.
- Price
- Free plan, paid plans vary
- Free plan or trial
- Free plan
- Founded
- 2014, in Oakland, California
- Founders
- Edith Harbaugh and John Kodumal
- Headquarters
- Oakland, California
- Customers
- More than 5,500, including about a quarter of the Fortune 500, by its own count
- Warehouse-native experimentation
- Snowflake, BigQuery, Redshift, Databricks
What it is
Edith Harbaugh and John Kodumal founded LaunchDarkly in 2014 in Oakland, California, to give teams control over what happens after code is deployed. Its core is feature flags, which the pricing page now calls CodeControl, sold alongside AgentControl for configuring and evaluating AI prompts and models, and observability for logs, traces, errors and session replay. The company says it serves more than 5,500 customers, including about a quarter of the Fortune 500.
How experimentation works
An experiment sits on a flag or an AgentControl config. You create variations, build metrics, set the audience split, choose the statistical approach and success criteria, then start an iteration. Metrics come from events the SDKs send, such as clicks, page views and custom conversions, or from your warehouse. Snowflake, BigQuery, Redshift and Databricks have native experimentation integrations, and you can write custom SQL metrics. Warehouse-native experiments that use your own assignment source don't need a LaunchDarkly flag at all.
The pricing page lists Bayesian and frequentist analysis, CUPED, sequential testing, sample ratio mismatch detection, multi-armed bandits, holdouts and mutual exclusion on every plan, with experiment approvals on Enterprise only. There are 30 SDKs, split into 13 client-side, 13 server-side and 4 edge.
No visual editor, on purpose
LaunchDarkly doesn't offer a WYSIWYG editor. Its own blog argues these tools skip code review and security scans and can cause page flicker. So every variant is code your team writes, reviews and deploys behind a flag. That fits engineering teams and slows down marketers who want to try a new headline this afternoon.
Revenue
No payment tool appears in its experimentation integrations, which cover the four warehouses plus Segment, RudderStack, Snowplow, Tealium, Fivetran, Sentry and Sprig. The docs show how to build average revenue per user as a custom numeric metric that sums purchase values per user. So revenue tests work once your code or warehouse supplies the purchase amounts.
AI
If your account includes AI assist, it can draft the experiment hypothesis. AgentControl runs experiments on AI prompts and model configs, which suits teams shipping AI features. Neither writes variations of a page or feature for you.
Who it suits
LaunchDarkly fits engineering organizations that already release through flags and want experiments without a second vendor, including companies that need approvals, custom roles and SCIM. The free Developer plan also makes it usable for a small team that evaluates flags on the server. It's a poor fit for a team without developers, or for anyone who wants to test marketing pages quickly.
Before you commit
Model your bill on MAU, not seats. Client-side MAU and experimentation MAU both count unique users, and experimentation MAU counts users who were eligible for an experiment even if they never entered one.
Pros and cons
What LaunchDarkly does well
- Experiments run on the same flags you use for releases, so shipping a winner is a flag change.
- The free Developer plan includes 100K experimentation MAU a month and unlimited seats.
- The pricing page lists Bayesian and frequentist analysis, CUPED, sequential testing, SRM detection, bandits and holdouts on every plan.
- Warehouse-native experiments on Snowflake, BigQuery, Redshift and Databricks, plus custom SQL metrics.
- Approvals, scheduled changes, custom roles and SCIM on Enterprise for large engineering teams.
Where it falls short
- No visual editor. LaunchDarkly argues against WYSIWYG testing tools, so every variant is built in code behind a flag.
- Experimentation MAU counts every eligible user your SDKs evaluate in a month, whether or not they entered an experiment.
- No payment tool integration. Revenue comes from events your code sends or from warehouse tables.
- Client-side MAU pricing grows with traffic. 20,000 monthly users on Foundation is $166.60 a month billed yearly.
- AI assist can draft a hypothesis, but LaunchDarkly doesn't write variations for you.
Questions people ask
Is LaunchDarkly Experimentation included or an add-on?+
It's included. The pricing page lists A/B tests and experiments on the free Developer plan with 100K experimentation MAU a month, and experimentation features on Foundation and Enterprise too. Experiment approvals are Enterprise only.
How much does LaunchDarkly cost?+
Developer is free. Foundation is pay as you go at $10 per service connection a month after the first 5 and $8.33 per 1,000 client-side MAU a month billed yearly, with a 14-day trial. Enterprise and the Guardian add-on are custom priced.
Does LaunchDarkly have a visual editor?+
No. LaunchDarkly argues that WYSIWYG editors bypass code review and cause page flicker, so variants are built in code behind feature flags.
What stats does LaunchDarkly Experimentation use?+
Both Bayesian and frequentist analysis. The pricing page also lists CUPED variance reduction, sequential testing, sample ratio mismatch detection, multi-armed bandits and holdouts.
Can LaunchDarkly measure revenue?+
Yes, if you supply it. You can build a custom numeric metric such as average revenue per user from events your SDK sends, or use warehouse-native metrics from Snowflake, BigQuery, Redshift or Databricks. It doesn't connect to payment tools like Stripe directly. Outtest reads revenue from the payment tool and judges every test on revenue per visitor.
Sources
Checked on 28 September 2026 from LaunchDarkly's own pages. Prices and features change; check LaunchDarkly's site before you buy.
- launchdarkly.com/pricing/
- launchdarkly.com/docs/home/account/calculating-billing
- launchdarkly.com/docs/home/experimentation
- launchdarkly.com/docs/home/experimentation/create
- launchdarkly.com/docs/home/experimentation/types
- launchdarkly.com/docs/integrations/experimentation
- launchdarkly.com/docs/guides/metrics/example-metrics
- launchdarkly.com/blog/have-you-outgrown-your-wysiwyg-experimentation-tool/
- launchdarkly.com/about-us/
Let Outtest run your split tests
AI agents read your analytics and payments, find where you lose the most money, build the fix and test it. Every test is judged on revenue, not clicks. Plans from $29 a month.