Onboarding that lands: the session where AI actually gives you time back

Run your current onboarding against these four and be honest about which one has no owner. In most companies it is the fourth, because the first three feel like hospitality and the fourth feels like conflict.

• Working relationships. Not "meet the team" - the specific eight people whose goodwill or knowledge unblocks this job. Name them, say why each one, and put the introduction in someone's diary.
• The context nobody writes down. Why the queue is split the way it is, which customer you never promise a same-day fix to, which system is held together by one person's patience. This is the stuff that takes new joiners four months to learn by embarrassment.
• The first real contribution. Something shipped, in production, that mattered. Not a training module. People decide whether they belong somewhere the first time their work is used.
• The early feedback loop. A written check at 30 and at 60, so that a mismatch is a conversation in week nine rather than a surprise in month seven. This is the kindest part of onboarding and the one that gets dropped first.

One new joiner is a relationship. Twenty-two in a month is a system, and systems fail differently. The three failures you will meet at Kestrel:

• Manager capacity, not HR capacity. The bottleneck is never the induction slot. It is the six managers each losing a day a week to context transfer. Anything you generate should reduce their load, not yours.
• Cohort envy. Twenty-two joiners talk to each other. If four get a real 90-day plan and eighteen get a laptop, everyone finds out by week two, and it reads as favouritism rather than as chaos.
• The same question, twenty-two times. When the same question arrives repeatedly, that is not a people problem. That is a missing document, and it is the cheapest thing you will fix all quarter.

Here is the reuse point, and it is the most valuable idea in this session. The role scorecard you wrote in b2 to hire against already says what good looks like in this job. Day 90 success is not a new question - it is the first checkpoint of the same definition. Which means the 30/60/90 plan is not authored, it is derived: outcomes from the scorecard become milestones, capabilities become the things the joiner practises, and the interview evidence you gathered in b5 tells you which milestone this particular person will find hardest.

Companies that write the job description and the onboarding plan in separate documents, months apart, by different people, end up hiring for one thing and onboarding toward another. The joiner feels the gap first and cannot name it. They just say the role was "not quite what was described".

Managers do not resist onboarding. They arrive on Monday with no idea what they are supposed to do, so they do the warm generic thing and hope. The pack removes that gap: what to prepare before day 1, who to introduce and why, what to say in the first conversation, what to check at day 30, and the two sentences that set expectations honestly.

Every joiner asks roughly the same fifteen questions in their first month: how leave works, when the first pay date lands, what expenses need approval, how probation is assessed, who to tell when they are sick, what happens if they want to change teams. Answering fifteen questions twenty-two times is a hundred and thirty conversations you do not have room for.

So you build an answer base from your own policies. But hold this rule hard: the answer base answers from your documents or it says it does not know. An onboarding assistant that invents a plausible leave entitlement, or gets the probation notice period slightly wrong, has done more damage than the slow human answer it replaced - because the joiner believed it and acted on it.

• The welcome. Somebody has left a job they knew for a job they do not. Being expected, by name, by someone who looks pleased about it, is the whole thing. A generated welcome email that says "we are thrilled to have you on board" is worse than no email, because it announces that nobody spent thirty seconds.
• The here-is-what-nobody-tells-you conversation. How decisions really get made, who to ask when the official channel is slow, which meeting matters. You cannot write this down, because writing it down would be a political act. It is transferred by a person who trusts the new joiner enough to say it.
• The check-in where somebody admits they are struggling. That admission only happens if the person asking has already made it safe. No form gets it. A form gets "all good, thanks", at day 30, from someone who resigns at day 110.

• A person decides, and the record shows who. Onboarding is mostly not decisions - until day 90, when somebody decides whether probation passes. That is a decision about a person's job, it is Dara O.'s to make, and the reason gets written the same day. The generated plan is the evidence base for it, never the verdict.
• Rate the job, not the polish. Milestones describe work: tickets closed, runbook published, queue area owned. Not "settling in well", not "great cultural fit", not "seems enthusiastic". Those are the same undefined criteria that broke the screen in b4, wearing a friendlier coat.
• People data is not prompt fodder. The plan is a document about a role, so it needs no personal data at all. You do not need the joiner's CV, address, visa status, health information or accommodation request in a workspace to write what day 60 looks like. If a prompt seems to want it, the prompt is wrong.