PostHog

Server metrics tell you the service is healthy. They cannot tell you that eighty percent of people abandon the second step of signup, or that the feature you shipped last month has four users.

PostHog covers that side: product analytics, funnels, session replay, feature flags, experiments, and surveys in one place. It’s open source, can be self-hosted, and the cloud free tier — a million events a month, plus allowances for replays and flags — is large enough that many small products never pay.

Install link

  npm install posthog-js
  

  import posthog from 'posthog-js';

posthog.init(process.env.NEXT_PUBLIC_POSTHOG_KEY, {
  api_host: 'https://eu.i.posthog.com',       // or us.i.posthog.com
  person_profiles: 'identified_only',          // cheaper, and usually what you want
});

posthog.identify(user.id, { email: user.email, plan: user.plan });

posthog.capture('project_created', { template: 'blank', source: 'onboarding' });
  

Server-side events use posthog-node and the same event names — which matters, because the events that count money usually happen on the server.

Name events like you’ll read them in six months link

The most common failure isn’t technical. It’s ending up with clicked_button_2, Click Button, and button-click all meaning the same thing.

• object_verb, past tense, snake_case: project_created, invite_sent, subscription_upgraded.
• Properties instead of new events. One project_created with a template property beats four event names.
• Track outcomes, not clicks. “Signed up” is a fact about your business; “clicked the blue button” is a fact about a button that will be redesigned.

Write the list of events down before you instrument, and keep it to a dozen at first.

The four features worth using in order link

1. Funnels. Pick the three steps that matter — landed, signed up, did the core thing — and look at where people fall out. This is the highest-value hour you’ll spend here.

2. Session replay. Watch five recordings of people failing at that step. It is uncomfortable and it answers the question faster than any dashboard. Mask inputs by default, and treat replay as personal data.

3. Feature flags. Ship dark, enable for yourself, then a percentage. A flag is also the fastest rollback that exists — faster than a redeploy.

     if (posthog.isFeatureEnabled('new-checkout')) { /* ... */ }
     

4. Experiments, once the first three are habits — flags plus a metric and the statistics done for you.

Privacy, which is not optional link

You’re collecting behaviour tied to identified people. Under GDPR that needs a lawful basis, a mention in your privacy policy, and a consent path for tracking that isn’t strictly necessary. Two settings do most of the work: choose the EU cloud region if your users are European, and mask inputs in session replay so passwords and card fields never leave the browser. Self-hosting is the option when the data genuinely cannot leave your infrastructure.

Where it fits alongside the rest link

Sentry tells you the code threw. Prometheus tells you the system is saturated. PostHog tells you the feature isn’t being used. Three different questions, three tools, and only the last one changes what you build next.

Next link

When something breaks, you want to hear it from a tool, not a user → Sentry