An experimentation platform is the tool a product team uses to put code behind feature flags, split users into variants and measure the result against product and revenue metrics.
For server-side and feature tests in 2026, GrowthBook and Statsig are the best-value starting points, Eppo (now Datadog Experiments) suits data teams with a warehouse, PostHog suits teams that want analytics in the same tool, and Optimizely, Amplitude Experiment and LaunchDarkly fit companies already paying for that vendor's wider platform.
We have not used these tools ourselves. This roundup draws on each vendor's pricing pages and docs, checked in November 2026, plus the product tests in our SaaS test database that show what these platforms are for.
Client-side vs server-side vs warehouse-native testing
Most marketing A/B tools work client-side: a script in the browser rewrites the page after it loads.
That's fine for headlines and layouts, but it can't reliably change pricing logic, trial length, onboarding flows or anything else decided on your servers.
Server-side testing moves the split into your code.
An SDK decides which variant each user gets, often through a feature flag, so the change ships with no flicker and can touch any logic.
The cost is engineering time for every test.
Warehouse-native platforms go one step further. They read metrics from your own data warehouse (Snowflake, BigQuery, Redshift, Databricks) instead of copying events into the vendor's cloud.
Results then match the revenue and retention numbers your finance and data teams already trust.
The tests that need this setup are the ones that change the product itself. A few from our database:
- AURUM added a floating onboarding checklist to a trial and saw +187% activation.
- Patrick McKenzie cut a free-trial limit from 15 to 8 cards, and paid conversion went from 1.04% to 1.89%.
- Server Density moved to packaged pricing tiers and got +114% revenue.
- Acuity Scheduling replaced a free plan with a 14-day Pro trial for +268% paid sign-ups.
Each one changes billing or app logic and is judged on a metric well past the click.
The downstream metric matters: a wellness app found a 14-day trial beat a 7-day one on trial starts but lost on paid conversions (Daphne Tideman's test), and Patrick McKenzie's 2012 redesign lifted trial sign-ups +10% while purchases stayed flat.
A platform that can measure revenue per user, not just sign-ups, catches results like these.
| Platform | Architecture | Free option | Pricing model | Best for |
|---|---|---|---|---|
| Statsig | Cloud; warehouse-native on Enterprise | Developer plan, 2M events/month | Metered events; Pro $150/month | Product teams that want flags, tests and analytics on a budget |
| Optimizely Feature Experimentation | Server-side SDKs, plus a separate web product | Rollouts: flags and one A/B test | Sales-led, no public prices | Enterprises testing on the web and in code |
| Eppo (Datadog Experiments) | Warehouse-native | Datadog's 14-day suite trial | Not published; sales-led | Data teams with a warehouse, and Datadog customers |
| GrowthBook | Warehouse-native; cloud or self-hosted | Starter (3 users) and open source | Per seat; Pro $40/seat/month | Engineering-led teams that want open source |
| PostHog | Cloud (self-host unsupported) | 1M flag requests/month | Usage-based per product | Startups wanting analytics, replay and tests in one tool |
| Amplitude Experiment | Cloud, tied to Amplitude Analytics | Free plan, 1 active experiment | Event-based; Growth and Enterprise custom | Teams already on Amplitude Analytics |
| LaunchDarkly | Feature-flag platform with experiments built in | Developer plan, 100K experimentation MAU | Per service connection plus client-side MAU | Engineering orgs already using LaunchDarkly flags |
Prices are as of November 2026; follow the links in each section for current terms.
Statsig
Statsig combines feature gates, experiments, product analytics and session replay.
As of November 2026, Statsig's pricing page lists a free Developer plan with 2M events a month and unlimited flag checks, a Pro plan at $150 a month with 5M events included and $0.05 per 1K events after, and custom Enterprise contracts.
Its statistics are the draw. Statsig offers CUPED variance reduction and sequential testing, and the pricing page puts CUPED on paid tiers. Warehouse Native is Enterprise-only.
The downside is ownership. OpenAI bought Statsig in 2025, and on May 5, 2026, Amplitude said it would "take on Statsig's brand and customers" and keep developing the platform.
The founding team stayed at OpenAI, so buyers should ask about the roadmap.
Not for: marketers who want a visual editor. More in our Statsig review.
Optimizely Feature Experimentation
Optimizely's docs describe Feature Experimentation as "a feature flagging and experimentation platform" with open-source SDKs for websites, apps and APIs.
The free Rollouts plan includes flags and "the ability to run one A/B test," which is enough to try the SDKs and not much more.
Its Stats Engine uses sequential testing, and the newer results page adds frequentist, Bayesian and CUPED options.
The main reason to pick it is having marketers on Web Experimentation and engineers on Feature Experimentation under one vendor.
As of November 2026, Optimizely publishes no prices.
Its plans page says every plan is "individually packaged" and routes you to sales.
Not for: startups or teams that want to sign up and pay by card. See our Optimizely review.
Eppo (now Datadog Experiments)
Eppo was built warehouse-native from the start. Datadog acquired it in 2025, and geteppo.com now reads "Eppo is now Datadog Experiments."
The site highlights CUPED and "CUPED++" variance reduction and connections to Snowflake, BigQuery, Redshift and Databricks.
Datadog's docs list four analysis methods (sequential frequentist, fixed-sample frequentist, sequential hybrid and Bayesian) and metric sources including Datadog Feature Flags, Real User Monitoring, Product Analytics and your own warehouse.
The product page also stresses checks for sample ratio mismatch and watching latency and errors next to results.
Pricing for Experiments isn't shown on Datadog's pricing page as of November 2026; you start with a demo or the 14-day Datadog trial.
Not for: teams without a warehouse or a data person.
The Datadog tie-in is a plus for existing customers and extra weight for everyone else. See our Eppo review.
GrowthBook
GrowthBook is open source and warehouse-native, and it can run in its cloud or on your own servers.
As of November 2026, GrowthBook's pricing has a free cloud Starter plan for up to 3 users with unlimited flags, experiments and traffic; Pro at $40 per seat per month for up to 30 users; and custom Enterprise.
The self-hosted open-source version is free with community support.
Billing by seat instead of by event is the big difference.
A high-traffic app pays the same as a small one, which makes GrowthBook cheap at scale for small teams.
It connects to Snowflake, BigQuery, Redshift, Databricks and ClickHouse, and its stats docs describe a Bayesian default with frequentist, CUPED and sequential options.
Downsides: seat pricing gets expensive if dozens of PMs and analysts need access, and SSO, approval workflows and holdouts sit on Enterprise.
Self-hosting also means you run the service. See our GrowthBook review.
PostHog
PostHog bundles product analytics, session replay, feature flags, experiments and surveys behind one SDK.
As of November 2026, PostHog's pricing gives every account 1M analytics events and 1M feature flag requests a month free, with usage-based pricing above that. Experiments are billed as flag requests.
Every experiment runs on a feature flag, and the experiments docs let you measure funnels, revenue per user, ratios or retention with a Bayesian or frequentist engine.
Having the replay and the funnel for a variant in the same project is the main reason teams pick it.
Downsides: the no-code web experiments suit only simple changes, so it's weak for marketing-page tests.
Self-hosting is officially unsupported. See our PostHog review.
Amplitude Experiment
Amplitude Experiment is the testing product inside Amplitude's analytics platform.
Its docs describe two modes: feature experiments that call the Experiment SDK or REST API from your code, and web experiments built in a visual editor.
It defaults to sequential testing, with a T-test as the alternative.
The appeal is that experiment metrics come straight from the events and cohorts you already track in Amplitude.
Since May 2026 Amplitude also runs Statsig, so its experimentation lineup may change; the two products are still separate for now.
The limits are tight on lower plans. As of November 2026, Amplitude's pricing page shows Free and Plus with unlimited feature flags but one active experiment and no A/B/n tests.
Growth and Enterprise are custom-priced, include two active experiments, and sell more experiments, holdouts and bandits as add-ons.
Not for: teams not already on Amplitude Analytics.
LaunchDarkly
LaunchDarkly is first a feature flag platform, and its experimentation docs describe attaching metrics to flags you already use.
Results show frequentist confidence intervals or Bayesian credible intervals, and metrics can come from a warehouse such as Snowflake.
As of November 2026, LaunchDarkly's pricing includes experimentation on every plan.
The free Developer plan has unlimited seats and 100K experimentation MAU a month.
Foundation costs $10 per service connection a month (billed yearly) plus $8.33 per 1,000 client-side MAU a month, and Enterprise is custom.
The pricing page lists CUPED, sequential testing and bandits across plans.
Downsides: client-side MAU pricing can climb fast on consumer apps, and the docs say experimentation isn't available on LaunchDarkly's EU instance.
For flag tools, the main LaunchDarkly alternatives are Statsig, GrowthBook and PostHog, which all bundle flags and experiments with free tiers.
Which experimentation platform to pick for your team
- Startup with engineers and no data team: PostHog or Statsig. Both are free to start, and both give you analytics next to the test.
- Data team with a warehouse: GrowthBook for open source and seat pricing, or Eppo (Datadog Experiments) if you want a sales-led product with strong stats and already use Datadog.
- Already on Amplitude or LaunchDarkly: start with the experimentation built into the tool you pay for, and check the plan limits first.
- Enterprise with marketers and engineers testing: Optimizely, which covers web and feature tests under one contract.
- Mostly marketing-site tests with some server-side work: a hybrid A/B tool. Convert's Full Stack is included on its Pro plan, from $420 a month as of November 2026, with SDKs for JavaScript, PHP, Python, Ruby, Android and iOS. VWO Feature Experimentation adds server-side SDKs to VWO's visual testing.
Whatever you choose, server-side tests on product metrics usually need more users than a headline test, so size them with our sample size calculator before you build.
For marketing-site tools, see our A/B testing tools roundup.
We may earn a commission if you sign up for Convert or VWO through our links; the other tools on this page don't pay us.