How it works
Three passes, in order:
- Find out what actually happens. The AI searches whatever it can reach — task trackers (Asana, Monday, Wrike…), docs (Notion, Drive…), and your email and calendar around your last 2–3 hires. It pays special attention to reminders and chase-ups ("still waiting on…"), because every reminder marks a step that doesn't happen on its own. It also asks you to paste in any onboarding plans you already use.
- Interview you. Questions one at a time, ten maximum, only about what the evidence couldn't answer — who owns an unclear step, whether a one-off is standard, and conflicts between what the evidence shows and what you said.
- Reconstruct and write. First it shows you the process as a compact timeline (day -14: contract sent, day 1: orientation…) so you can check it at a glance. Then one grouped round of remaining questions, then the document.
What you need to give it
- Access to email, calendar, task tools, or docs — or, without access, pasted email threads around recent hires plus any existing checklists.
- Your existing onboarding plans or templates, if any.
- Answers during the interview.
What you get back
One document, saved as processes/onboarding.md (or printed in the chat if the AI can't write files), with exactly these sections:
- Overview — the phases from offer to day 90, in three sentences.
- A checklist per owner — HR, hiring manager, buddy, IT, new hire — each step with its timing and trigger.
- Variants — real differences per role or country, if any.
- Tools and documents — what you use, with links to the masters (never copied content).
- Check-in cadence, known failure points (the steps the evidence shows slipping), and a data note on which steps touch sensitive personal data.
How to run it
Copy the whole file from onboarding-docs.md and paste it into a new AI conversation. Then answer its questions.
Good to know
It documents the current state only — no improvements folded in, no steps invented, no "common HR practice" used to fill gaps. Anything backed only by memory is marked (unverified) so you know exactly what to double-check. Sensitive data (IDs, bank details, salaries) never goes into the document — only the steps that handle it.