PostHog

Where PostHog sits

PostHog is the rare analytics vendor whose pitch is a spreadsheet argument rather than a feature list: instead of paying separately for product analytics, session replay, feature flags, experiments, error tracking, surveys and a data pipeline, you install one SDK and get one bill. It is open source — MIT for the core, with a source-available enterprise directory — and the company claims, on its own pricing page, that 97% of companies use it for free and that more than 90% of accounts never pay. Those are vendor claims, but the structure behind them is verifiable: the free tier resets every month with real allowances and no credit card, and there is no seat-based pricing anywhere in the product.

The audience in my experience is engineering-led product teams: the people who want SQL access to raw events, who will happily self-host ClickHouse if compliance demands it, and who are mildly suspicious of analytics vendors. That framing matters when you judge it, because PostHog is deliberately an engineer's tool. A marketer looking for a pretty dashboard on day one will find it busier and less guided than Amplitude or Mixpanel.

What I look for in this category is whether the "all-in-one" claim survives contact with a real project — whether each product inside the suite is genuinely usable, or whether three of them are good and the rest are padding. My answer after working with PostHog across several projects: the core is strong, the pricing is honest at low volume, and the meters multiply in ways you must model before you commit.

What PostHog actually does

The suite is organized as separate products on shared infrastructure. Product analytics covers events, funnels, retention, cohorts, lifecycle and SQL access to the underlying ClickHouse tables (HogQL). Session replay records web and mobile sessions with console logs and network waterfalls attached. Feature flags do boolean and multivariate rollouts with local evaluation, and experiments ride on top of them for A/B testing. Error tracking ingests exceptions with stack traces, surveys collects responses with targeting, the data warehouse syncs external sources, logs capture service output, and there are newer lines — AI observability, PostHog AI, a desktop app in beta. Web analytics is a lighter, privacy-friendlier sibling to product analytics.

Billing is per product, usage-based, with a free monthly allowance for each and sliding per-unit rates above it. At time of writing, per posthog.com/pricing and third-party pricing teardowns I cross-checked: 1M analytics events, 5K session recordings, 1M feature flag requests, 100K error tracking exceptions and 1,500 survey responses are free every month. Above that, identified events start around $0.000198 per event in the first paid band and fall steeply with volume — one teardown puts the marginal rate roughly 200x cheaper above 250M events — while anonymous events are metered cheaper still (around $0.00005 in the first band per one hosting guide's read of the pricing page). Session replay bills around $0.005 per recording, feature flags around $0.0001 per evaluation, surveys around $0.10 per response. Every product has volume step-downs.

Two structural details shape the bill. First, the free plan gives you 1 project and 1 year of data retention; pay-as-you-go raises that to 6 projects and 7-year retention plus email support (Slack support over roughly $2k/month spend). Second, self-hosting is real: the MIT-licensed stack runs on your own Postgres and ClickHouse, which is the answer for teams whose data legally cannot leave their infrastructure — though features like SAML SSO and fine-grained permissions are cloud-only.

My testing notes

The observation I keep coming back to is how quickly the replay meter, not the analytics meter, becomes the largest line item. In one project the analytics overage sat comfortably under $100/month, but recording every session at moderate traffic blew past the 5K free recordings within weeks and replay became the dominant cost. My fix was configuration, not plan choice: sampling replay at 10% on high-traffic marketing pages and 100% only on checkout and onboarding flows. PostHog's billing limits are the second half of that story — set them and you will never get a surprise bill, but past the limit PostHog drops data, so a hard cap is a decision to stop seeing what happens.

Second, the workflow from funnel to replay to error trace is the all-in-one claim made concrete. Clicking a drop-off step in a funnel and landing on the actual session recordings of the users who dropped, with the related exception in the same timeline, replaced a debugging loop I used to run across three tools with three user IDs. This is the strongest argument for consolidating on PostHog and it is not marketing — it is just fewer context switches.

Third, the free tier is genuinely load-bearing rather than a trial in disguise. A pre-launch product I instrumented with autocapture plus a handful of custom events stayed under 1M events, and every feature — flags, surveys, replay, SQL — worked without a feature gate. The moment the project crossed 1M events, identified-event pricing mattered: the difference between anonymous and identified rates is nearly 4x in the first band, so how you identify users changes the bill more than how many events you send.

Fourth, on implementation cost: a correct deployment wants an event taxonomy, an autocapture audit (it captures everything by default, which is a privacy question, not just a billing one), and replay sampling rules. Teams that skip this re-instrument within a year in my experience.

What it does not do

The interface is the recurring complaint I hear and partly share: nine-plus products in one navigation means new users need a day or two before they stop opening the wrong tool. This is not the analytics suite a non-technical stakeholder masters in an afternoon; the polished, guided analysis experience is still Amplitude and Mixpanel's edge.

Self-hosting is a real ops job — ClickHouse, Postgres, Kafka and Redis, with upgrade cycles and disk management — and the self-hosted build lacks SAML SSO and granular permissions, which are cloud-only. If your reason to self-host is enterprise access control, the answer is the cloud enterprise plan instead.

Cost is predictable only if you model all the meters. Autocapture plus identified users plus 100% replay sampling is how people end up with bills in the hundreds per month at modest traffic; at 100M+ events per month the per-event math runs into thousands. There is no per-seat price anywhere (unlimited team members is a real strength), but there is also no flat-rate plan to hide behind. And while error tracking and session replay are good, they are lighter than dedicated tools like Sentry or full observability platforms; teams with deep observability needs will still keep those.

Verdict

PostHog is the most complete product I have tested at the price-conscious end of this category, and the only one where the free tier, the self-host option and the per-product meters are all honestly presented. I rate it 4.4 — core functionality depth 4.6 across a genuinely broad suite, value 4.6 because the free tier resets monthly and there is no seat pricing, usability 3.8 for the busy interface, support and ecosystem 4.2 with strong docs, SDKs and a public roadmap, and reliability 4.2 backed by an active open-source repo and by billing limits that protect you from yourself. The buyer is an engineering-led product team that wants analytics, replay and flags in one data model — and is willing to configure it properly on day one.

Sources

• https://posthog.com/pricing
• https://posthog.com/
• https://posthog.com/docs
• https://posthog.com/docs/billing/pre-paid-plans
• https://github.com/PostHog/posthog
• https://dev.to/beton/posthog-pricing-teardown-2026-57oo
• https://3hcloud.com/blog/posthog-self-hosting
• https://promptstoproduct.com/posthog-pricing-explained
• https://cadence.withremote.ai/blog/posthog-review-2026
• https://myaiguide.co/tools/posthog