Over the years we have experimented with a variety of solutions for A/B and multivariate testing solutions.

Google Optimize was our go-to for a long time. It was capable, it integrated cleanly with Google Analytics, and it had a free tier generous enough to cover real testing programs. Then Google discontinued it in late 2023. Optimizely is still one of the most powerful platforms on the market, but the pricing has climbed to a point where it’s out of reach for most of the clients we work with. CrazyEgg is affordable and easy to start with, but it runs thin once you want to do more than basic testing. VWO is excellent, and if a client tests constantly it can be worth every dollar. But for clients with more intermittent testing needs the cost is hard to justify.

That’s the gap that sent us looking for something else, and it’s where PostHog came in.

What PostHog actually is

The first thing to understand is that PostHog is not a testing tool with a few extras bolted on. It’s a full product analytics platform, and A/B testing is one piece of it.

In a single tool you get product analytics, session replay, feature flags, surveys, and error tracking, alongside experiments. For a client, that matters in two ways. First, you’re not stitching together a separate testing tool, a separate analytics tool, and a separate session-recording tool and hoping they agree with each other. The data lives in one place. Second, because testing sits inside a broader platform, an experiment result isn’t a dead end. You can filter any report, funnel, or session recording by which test variant a visitor saw, which makes it much easier to understand why a test won or lost, not just that it did.

So even for a client who comes to PostHog purely to run a headline test, there’s a lot more available the moment they want it.

How testing works

PostHog’s testing model rests on two connected ideas: feature flags and experiments.

A feature flag is a switch. It decides which version of something a given visitor sees. An experiment sits on top of that switch. When you create an experiment, PostHog creates a feature flag to go with it, splits your traffic across the variants, and handles the statistics that tell you which version is winning.

Setting up an experiment involves running through a short guided wizard with three steps. You name the experiment and set its feature flag key. You configure the variants and how traffic splits between them, a control plus one or more test versions. Then you choose the metrics that define success, such as a signup, a click, or a purchase. Save it as a draft, review it, and launch when you’re ready. PostHog then reports results and tells you when a variant has reached statistical significance, so you’re not guessing about when a test is done.

The pros of PostHog

A very generous free tier. This is the headline for most of our clients. PostHog’s free tier includes up to one million feature flag requests a month, along with product analytics, session replay, surveys, and more. PostHog reports that more than 90 percent of companies use the platform without paying anything. For a client running occasional tests, the free tier is all that is needed.

Unlimited experiment metrics. You’re not forced to pick a single success metric and ignore everything else. You can track a primary goal and layer on as many secondary metrics as you want, which means a test can answer more than one question at a time.

It is developer-friendly. Our developers like PostHog. They don’t feel like they’re fighting it the way they do with other platforms.  Here are a few reasons why:

• The feature flags make it obvious which experiment is currently loaded, which makes the JavaScript for the visual changes easier to write.
• All the test code lives on the site itself, so there’s no bouncing back and forth between the site and the PostHog dashboard to confirm everything is in place.
• Debugging is straightforward thanks to PostHog’s JavaScript methods for quickly switching between feature flags.
• Above all, PostHog gets out of the way. It doesn’t force its dashboards or other tools into the process of building an A/B change. The separation between setting up the experiment and its metrics on one side, and writing the code that changes the site on the other, is clean and deliberate.

It’s less likely to break your site. This one matters to us because we’ve been burned before. Tools like VWO and Optimizely typically make their changes with a visual editor that injects JavaScript to rewrite the page after it has already loaded. That’s what causes the “flicker” you sometimes see, where the original version of a page flashes on screen for a moment before the test version replaces it. This approach is also what tends to break things on modern, dynamic sites by introducing JavaScript conflicts.

There’s a capable AI assistant built in. Most testing platforms have added some flavor of AI by now, but PostHog’s AI feels more woven into the product than most. You can describe what you want in plain language and its AI will set up feature flags and experiments, build reports, write queries, and dig into your data. It works inside the actual product rather than sitting off to the side as a separate chatbot. It doesn’t remove the need for a developer to build the on-page change, but it lowers the barrier to getting an experiment configured and understanding the results.

The cons of PostHog

No platform is perfect, and PostHog has its drawbacks.

It’s code-first, and that comes with a learning curve. A non-developer can set up the experiment itself entirely in PostHog’s interface: naming it, defining the variants, and choosing the metrics that count as success. What needs a developer is the code that makes the actual visual change on the page. PostHog does have a no-code visual editor for building those changes without code, but it’s still in beta and PostHog has openly signaled it isn’t a priority going forward, so the reliable path is to have a developer wire the change up. Either way, there’s a learning curve before the flag-and-experiment model feels natural. For clients with development support this is a non-issue, and it’s how we work anyway. For a team hoping to run tests entirely on their own with no engineering help, it’s a genuine limitation to weigh.

The company and the interface are a little quirky. PostHog has a distinct personality, and the interface reflects it. It’s not the buttoned-up enterprise feel of some competitors, and there’s a short adjustment period before it clicks. I personally have grown to really like it.

It lacks some of the bells and whistles. PostHog covers more than you might expect, including heatmaps and session replay, but dedicated conversion-optimization suites still have polish it doesn’t match. You won’t find a mature drag-and-drop visual editor, the deep audience-targeting and personalization rules those tools offer, or the wider marketing suite that a platform like Optimizely bundles, such as content management and campaign tooling. If your program leans on one of those specifically, check that PostHog covers it before you switch.

The verdict

PostHog isn’t the right fit for every team, and the code-first approach to implementing tests may be a turn off for some. But for the situation most of our clients are in, wanting a capable, well-supported testing platform without a price tag that only makes sense for constant testing, it’s a strong choice. The free tier is very generous, the experiment features are more than enough for most projects, and the teams building the tests actually enjoy using it.

If you’re looking for a cost-effective way to test and improve your site PostHog is well worth a look.