On 18 November a founder's post about his PostHog bill passed 300,000 views on X. PostHog's co-founder replied that the spike came from autocapture. I went through 19 public complaints about PostHog bills from the last two years, and the same few causes kept coming up. Here they are, with the fix PostHog's own docs give for each.

// Detect dark theme var iframe = document.getElementById('tweet-2122878581999055011-386'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2122878581999055011&theme=dark" }

PostHog says most companies pay nothing, and the bill only bites when you track far more than you meant to.

1. Autocapture

Autocapture records clicks, inputs and form submits on every page by default. On a busy app that turns into a lot of events nobody looks at.

Fix: set an allowlist so it only captures what you use, or turn it off and capture the events you care about yourself.

posthog.init('<ph_project_token>', { api_host: 'https://us.i.posthog.com', autocapture: false, }) 

2. Identified events

PostHog says anonymous events can be up to 4x cheaper than identified ones. If every visitor gets a person profile, you pay the higher rate on everything.

Fix: only identify users when you need to, and call identify() once per session. The same goes for group().

3. Pageviews and page leaves

PostHog captures these automatically. If you don't use them on every page, you can switch them off and capture them only where they matter.

posthog.init('<ph_project_token>', { api_host: 'https://us.i.posthog.com', capture_pageview: false, capture_pageleave: false, }) 

Bounce rate and a few other features depend on these, so check before you cut.

4. Session replay

Recording every session adds up fast. In the replay settings you can sample a percentage of sessions, record only on certain URLs or events, and drop sessions shorter than a minimum duration. Recording 100% of checkout and 10% of everything else is a common pattern.

5. Feature flag requests

PostHog bills flags per request to the /flags endpoint, and server-side local evaluation polls for flag definitions every 30 seconds by default. Per PostHog's docs, a poll costs 1 request when nothing changed and 10 when definitions changed or on older SDK versions without ETag support. That's for each server instance, so 20 instances cost 20 times as much.

Fix: upgrade to an SDK version that sends ETags, lengthen the polling interval, share definitions across instances, and don't use local evaluation on edge or Lambda functions. Old staging or demo environments can also keep calling /flags long after you've forgotten them.

6. No billing limit

PostHog lets you set a billing limit per product. Ingestion stops when you hit it, so a spike can't run away with your month. If you only have alerts, set a limit.

Finding your own culprit

PostHog has a billable usage dashboard template that breaks your bill down by event, SDK and product. Start there.

I wanted the same breakdown in dollars, with the fix spelled out, so I built Frumo. It's a free, open-source command-line tool. You give it a read-only PostHog key and it lists the events, replays and flags driving your bill, estimates what you'd save, and gives you the fix to paste in. It runs on your machine, and your data only goes to PostHog.

Try it on sample data:

npx frumo-cli demo 

The code is on GitHub under the MIT licence. Frumo is built for PostHog and isn't affiliated with them.

If you run it on your own project, I'd like to hear what it gets wrong.