the founder's playbook

You can run a whole startup on one tool for a long time

We run defract's entire growth operation on PostHog — and not just product analytics. The whole marketing side lives there too: every channel's data pulled into one place (LinkedIn, Cloudflare, search, our own site), plus a CRM, user feedback, experiments, error tracking, and an AI agent that reads all of it. Here's the real setup, what to check every day, and the honest list of what PostHog does not replace.

Most early startups drown in tool sprawl before they have users to justify it. When we started defract we did the usual thing: product data in one tool, web analytics in Google Analytics, heatmaps in Microsoft Clarity, errors in Sentry, a CRM like HubSpot, plus a survey widget and a feature-flag service. A stack of subscriptions, a stack of logins, and more time spent managing tools than getting anything out of them. You need almost none of it. One well-instrumented PostHog is your control room — your HQ — for a long time.

To be clear up front, so this reads as a playbook and not a pitch: PostHog is not your payment processor, your auth provider, your transactional email, or your app's backend. You still need those. But for the entire job of understanding your users, measuring your marketing, and deciding what to do next, you can run on one tool well past your first hundred users. This is how we do it.

1. Instrument a spine, not everything

The first mistake is capturing hundreds of events and drowning in noise. Capture a small, high-signal spine you can reason about: the moment someone signs up, the moment they do the one thing that is your product's core action, and the handful of steps in between. For us the core action is a user launching their first agent — that single event is our activation definition, and almost every dashboard hangs off it.

Name events for the user's action, not your code (signed_up, project_created, first_agent_launched — not btn_click_7). Autocapture will grab clicks and pageviews for free; add named events only where a decision depends on them. You can always add more later. You cannot easily un-pollute a project full of junk.

2. Build the five dashboards that actually matter

Ignore vanity. Five views answer nearly every question a founder has:

What "good" looks like is directional, not absolute: activation trending up, a retention curve that flattens instead of hitting zero, engagement concentrating on your core action rather than scattering. You are tuning slopes, not chasing a magic number.

3. The daily check — the routine, not the dashboard

A dashboard you open once a month is decoration. The value is the habit. This is the actual loop we run every morning.

Run it daily and you feel the product's pulse in a way no weekly report gives you.

4. Use it as your CRM — for a while

You do not need a sales CRM to know your users. In PostHog, every user is a person with properties, and you can write your own. We enrich each person with custom fields — who they are, their company, their status, where they came from, notes — and now our "CRM" is the same place as our analytics, so a person's entire behavioral history sits next to their profile. Filter, build cohorts ("activated founders in the last 30 days"), and act.

This holds comfortably through early sales and design-partner conversations. You outgrow it when you need pipeline stages, deal tracking, and a sales team's workflow — then add a real CRM. Until then, a second tool is overhead you do not need.

5. Delete the tools it replaces

The reason one tool is enough is that PostHog is not just analytics. Each of these is a SaaS subscription you can skip at the start:

One bill, one login, and — the real win — all of it joined to the same user. An error links to the replay of the session that threw it. A survey response sits next to that user's behavior. That join is impossible when the data lives in separate products.

6. Make it your marketing HQ too — pull every channel in

This is where one HQ pays off most. Product analytics is only half the job; the other half is marketing, and PostHog is just as good a home for it. Anything you can send over HTTP becomes an event through the capture API, so every channel's numbers land in the same place as your product usage.

What we pull into a single project:

The real unlock is the join. Because marketing data and product data live in one place, keyed to the same person, you can answer the question a founder actually cares about — not "how many impressions," but "which LinkedIn post brought the users who installed, activated, and came back." One tool, one query, the full funnel from first touch to retained user. That is a question you cannot answer when reach lives in one tool, clicks in another, and retention in a third.

7. Wire it to your AI agents — the part almost nobody does yet

This is the 2026 upgrade. PostHog exposes an MCP server and a SQL interface (HogQL), which means an AI agent can read and write your entire analytics stack in plain language.

Point an agent at PostHog's MCP server and you can ask "how did activation move this week, and which cohort dropped" in plain language and get an answer from your real data — no dashboard-building, no SQL by hand. Give it the query API and it writes HogQL directly for anything ad hoc. Here is what our agent actually does, on its own:

Your analytics stop being something you visit and become something that works while you sleep.

8. Know when to graduate — the honest limits

A playbook that pretends one tool does everything forever is lying to you. PostHog is your HQ for understanding and deciding. It is not:

Naming these is the point. The claim is not "you will never need another tool." It is "you can go remarkably far — well past your first users and first dollars — before you need a second one." Add tools when a real limit forces you to, not because a landing page told you a startup needs eight of them on day one.

The honest part

Tool sprawl feels like progress — every new subscription feels like leveling up. It is usually the opposite: scattered context, higher cost, and no single place that answers "how are we doing." Pick one HQ, instrument it well, build the habit of the daily check, and wire it to an agent so it works while you sleep. You will make sharper decisions with one tool used properly than with eight used shallowly.

One database. One source of truth. One login. One MCP connection. That consolidation is the actual hack — it creates focus and speed, for you and for your agents.

This is the actual stack behind defract — a desktop app that runs your whole development process on parallel Claude Code agents (story → design → review → ship). The playbook above is free. So is the tool.

Build the thing worth measuring.

defract runs your whole process on parallel Claude Code agents — local-first, on your own Claude subscription.

Written from the actual stack we run defract's growth on. PostHog is an independent product; this playbook is not sponsored — it is just how we work.