Write the hypothesis
Each test states what changes, who sees it, which metric should move and how much movement would change your plans.
4.9 / 5PostHog A/B testing that settles the argument.
PostHog A/B testing planned, built and read by one team, so every experiment answers the question it was run for.
Selected work, in motion.
The experiences behind the journeys teams measure.Tests that settle nothing.
Experiments are cheap to start in PostHog. An answer you can act on takes more planning.
BeforeAfter
Tests stopped the moment one variant looks aheadSample size and run time agreed before launch
A win on clicks, a loss on revenue nobody noticedSecondary and guardrail metrics watched beside the goal
Variants pasted into the page by handEvery variant served from a PostHog feature flag
Results screenshotted into Slack and forgottenA written readout with the decision and the next test
What's included
Hypothesis to readout.
We run the whole experiment, from the idea on paper to the decision in your backlog.
Build on feature flags
Variants ship behind a PostHog multivariate flag, so targeting, rollout and rollback happen in PostHog without a new deploy.
Set guardrail metrics
Revenue, retention or error rates tracked beside the goal metric, so a winning variant cannot quietly damage something else.
Read the results plainly
A short readout: what happened, how sure we are, what to ship and which test the result suggests next.
Security, by consideration.
Trust is part of the build.
For your PostHog project, we consider event collection, personal data, workspace access and ingestion endpoints as part of one connected build.
See our approachConnected systems. Considered safeguards.
Changes you can trace.
GitHub workflows, peer review and dependency checks bring decisions into the release process.
Protected in transit.
Certificate configuration, HTTPS and secure headers help protect the path between your users and your systems.
Intelligence with limits.
Scoped tool access, validated inputs and human approval where automated actions carry consequences.
One test, done properly.
Three steps per experiment, repeated as your backlog grows.
- 01
Design
Hypothesis, audience, goal metric, guardrails and the traffic needed to detect the change you care about.
- Experiment brief
- Sample size estimate
- Metric definitions
- 02
Build and launch
Variants built behind a flag, checked on every device, then launched to the agreed split.
- Variant build
- Flag configuration
- QA checklist
- 03
Read and decide
We wait for the planned sample, then write up the result and the decision.
- Results readout
- Ship or revert call
- Next hypothesis
Built for what’s next.
Built to driveMore memberships.
Built to driveBetter-qualified leads.
Built to driveMore qualified candidates.
Built to driveMore enquiries.
Built to driveMore qualified demos.
Built to driveMore workplace demos.
Built to driveMore business accounts.
Built to driveMore free trials.
Built to driveMore building owners.
Built to driveMore free sign-ups.
Private Villas Mexico
Built to driveMore direct bookings.
Built to driveMore committed players.
Built to driveMore returning customers.
Built to driveMore meaningful demos.
Built to driveBetter talent connections.
Built to driveMore course creators.
Built to driveMore product discovery.
Built to driveMore sign-ups.
Trusted across our practice
Working together. In their words.
Independent reviews. Long-term working relationships.
“After a website (re)launch, they are also always accessible and happy to help. I highly recommend everyone who is looking for a new and more stable website in Webflow, to work with Stega!”
“Stega Creative team not only understood our needs and vision for the websites, but they also came up with improvements.”
“We’ve been working with Stega & Gabriel for the last two years. Cannot recommend them enough. They’re responsive, professional & an asset.”
PostHog experiment questions.
A few things worth discussing before your PostHog A/B testing takes shape.
How does PostHog A/B testing work?
PostHog A/B testing splits users with a feature flag and compares a goal metric between the variants. You define the variants, the metric and the audience in PostHog. Your code reads the flag to show each user their variant, and PostHog calculates which variant performs better.
How much traffic do I need for a PostHog experiment?
The traffic a PostHog experiment needs depends on your current conversion rate and the smallest change worth detecting. PostHog estimates the sample size when you set up the test. Low-traffic products should test bigger changes on busier steps, since small tweaks may never reach a clear result.
Can PostHog replace Optimizely for A/B testing?
PostHog can replace Optimizely for product and code-based A/B testing, especially if you already use PostHog for analytics and feature flags. Optimizely is stronger when marketers need a visual editor, personalisation at scale or a stats engine built for high-traffic programmes. We work with both and recommend based on who runs the tests.
How long should a PostHog experiment run?
A PostHog experiment should run until it reaches the sample size planned before launch, and for at least one full week to cover weekday and weekend behaviour. Stopping early because one variant looks ahead is the most common way to ship a false win.
Revamp today.The 
way.
Your ambition. Our collective craft.
Strategy, design and technology, working as one.