AI for Onboarding New Employees: Three Hires, Three Kits, and What the AI Got Wrong

Onboarding is mostly a knowledge problem. A new person needs to know who owns what, how things are done here, which documents to trust, and what to do first. In most small companies that knowledge comes from whichever founder has time that week, and it shows. Gallup found that only 12% of employees strongly agree their organization does a great job of onboarding new employees.

Between late August and late September we brought on three people: a part-time bookkeeper, a fractional CFO and an outbound sales lead. Each one got an onboarding kit that our AI wrote from what was already in our company brain, the repository where our email, call transcripts and docs live.

What a kit looks like

Each hire got three documents, written for three different readers. The CFO’s role was the most complex, so the CFO’s kit was the longest:

The welcome note is short on purpose. The sales lead’s reading list says it directly: “the Welcome Note is genuinely enough to start. The rest is there when you need it.” The context pack is the deep reference. The SOP isn’t for the new hire at all. It’s the checklist for whoever creates their accounts, so access gets set up the same way every time and lands in our access registry.

The kit starts with who owns what

The first thing the bookkeeper’s pack covers is a one-paragraph map of who decides, who owns each tool, who approves payments, who does the work day to day and who is the bookkeeper’s contact. It’s the question every new person has and nobody thinks to answer, because everyone already there knows.

The sales lead’s pack went further and spelled out what the role does not do. It doesn’t build the software, because engineering owns that. It doesn’t write the founder’s personal-brand content, because someone else owns that. A list of what’s not your job prevents more first-month confusion than a list of what is.

Where the AI caught stale docs before the new hire did

The bookkeeper’s and CFO’s packs open by listing the documents they were built from. The bookkeeper’s came from six: our finance manual, a build plan, a reconciliation, a tax memo, a budget changelog and a correction letter from a previous accountant.

While pulling those together, the AI noticed that our finance manual listed 17 account names in prose, while the file our tools actually use had 13 since a cleanup in June. It didn’t quietly pick one. The pack told the bookkeeper the prose list was stale and should be ignored, and made reconciling the live chart of accounts against our source of truth her first review item. If her first week had started from the manual, she’d have been categorizing against accounts that no longer existed.

The CFO’s pack did the same thing at a larger scale. It flagged a 24-month projection tab as built on an old team roster and an old cost per engineer, said which parts of its shape we still believed, and called it the obvious first thing to rebuild. That’s the conversation you’d want to have in week one, and without the flag it would have taken a month to surface.

A person writing onboarding docs copies from the docs they remember. The AI read all six and noticed when two disagreed. Confidently wrong docs were the worst of the five places our knowledge hid, because an AI repeats a wrong doc as confidently as a right one. Asking it to compare sources and say which one wins is the fix.

The first week is real work

The bookkeeper’s welcome note gave her a two-week plan built around “five known issues,” all real problems in our books: marketplace payouts recorded as refunds instead of revenue, contractor payments flipping between two accounts, $4,000 to $8,000 a month sitting in Uncategorized, customer refunds parked inside income, and platform fees filed as office expenses. She started on actual work on day one, with a clear note on which questions to ask our operations lead instead of guessing.

The note also had one rule I’d put in every onboarding kit: “Anything unusual: flag it, don’t fix it silently.” A new person in a system they don’t fully understand will find things that look wrong. Some of them are wrong. Some are deliberate in ways that aren’t written down yet. Flagging tells you which docs are missing.

What the AI got wrong

The bookkeeper’s kit was written on August 31 from our plan for how her access would work. She started on September 1, and by the end of that day a person had changed 31 lines across the kit and the access registry, because the plan and reality had drifted.

The plan said she’d get a shared drive folder. In practice she got access to individual files in someone’s personal drive. The plan said she’d be a single-channel guest in our chat. Our operations lead made her a full member instead, which on our plan means a paid seat. A few account invitations moved from “pending” to “due now” once her contracts were signed.

The AI wrote the plan down correctly. The plan changed when real people did the setup. That’s the normal state of onboarding, and the reason the SOP lives in the same repository as the access registry: the person doing the setup updates both in one pass, and the next kit starts from what actually happened.

After day one

A kit gets someone through the first week. After that, the questions get specific: what did we promise this client about hosting, which version of the payment process is current, who has the login for this tool. Those are the questions a new person usually asks a founder, one interruption at a time.

That’s where the brain itself takes over. Our company rulings live in one shared file that the AI loads before it writes anything, so a new person’s AI follows the same rules mine does from their first session (how that file works). Answers come back with a link to the email or transcript they came from, so a new hire can check the answer instead of trusting it.

If you want to try this

You don’t need our setup to start. Take your last hire and write down the ten questions they asked in their first two weeks. For each one, find where the answer lives today. The ones that live in a person’s head are your first documents. The ones that live in two places are your first cleanup.

If you’d rather have it built, I set up the brain these kits come from for one team: email, docs, calls, spreadsheets and calendar, in accounts you own. It’s from $3,997 for a team of up to 20, live in 14 days or all your money back. You can try the demo first to see what asking one of these feels like.