AI onboarding and training that answers from your own playbook
Turn handbooks, SOPs and recorded sessions into a role-based learning path, a quiz and an assistant that answers "how do we do this here" with a citation, so new hires stop waiting for the one person who knows.
Ramp-up takes months because the knowledge lives in people
A new hire's first quarter is mostly discovery: which system holds what, who approves what, how this team actually does the thing the handbook describes in general terms. The answers exist, scattered across a wiki, a shared drive, old chat threads and the heads of two experienced colleagues who are busy.
Generic onboarding content does not fix that, because the questions are specific to your organisation. An assistant grounded in your own material does, provided it cites what it says, knows when content is stale and routes the questions it cannot answer to a person.
How it works
From scattered documents to a guided first ninety days
The assistant builds the path and answers the questions; managers and experts stay in charge of what is taught. Unanswered questions become the content backlog.
1. Collect what you already have
Handbooks, SOPs, product docs and recorded sessions are ingested with their permissions intact. Recordings are transcribed so a two-year-old team walkthrough becomes searchable text.
2. Build the path per role
For each role the assistant drafts a sequenced path with short quizzes, drawn only from approved sources. The manager edits it once; later hires in that role inherit it.
3. Answer, cite, learn
In the chat tool the new hire asks in plain language and gets an answer with its source and date. Questions with no good source are logged for an expert, and the correction updates the content.
What you get
A system your people team can run
Permissioned content pipeline
Connectors for your wiki, document store, LMS and recordings, with transcription for audio and video, on the architecture of our internal copilot.
Role-based paths and quizzes
Generated from your material, approved by managers, versioned when content changes. Quiz results show where a cohort is struggling before anyone asks.
Assistant in your chat tool
Cited, dated answers where people already work, with a one-click "ask a person" route. Built with the same front end as our AI chatbots.
Progress and gap dashboard
Time to complete each path, quiz scores by topic, most-asked questions and the unanswered ones, so the content backlog writes itself.
HRIS and LMS integration
Role and start date flow in; completion and quiz results flow back. We have built this against our own ZekoHR platform and against third-party HR systems through their APIs.
Guardrails we build in
Accuracy thresholds: answers are only given when retrieval finds a sufficiently relevant, approved source; otherwise the assistant says so and routes to a person. Policy topics link to the canonical page instead of paraphrasing. Human-in-the-loop: managers approve every path and experts review flagged answers. Audit trail: question, sources, answer and any correction are logged, which also shows which content is doing the work. PII handling: the assistant sees role and team, never pay or performance data; permissions are enforced at retrieval, and conversations are retained only as long as the dashboard needs. This follows our AI data privacy controls and the wider AI security and governance practice.
Typical rollout
One role, one cohort, then everyone
Audit (2 weeks)
Pick the first role, inventory its sources, decide what is canonical, and connect the HRIS for role and start-date data.
Build (3–4 weeks)
Content pipeline, transcription, the first path with quizzes, and the assistant in your chat tool. Managers review the draft path.
Pilot cohort (2–3 weeks)
A handful of new starters use it with experts watching the flagged-answer queue. Unanswered questions become content tasks.
Widen and hand over (1–2 weeks)
Further roles added on the same pipeline, dashboard live, the people team trained to maintain paths and sources.
Illustrative example
What a first release usually looks like
Illustrative scenario, not a client case
A growing services firm hires customer success managers in batches and relies on two senior staff to train them. The first path is built from the handbook, a dozen SOPs and recordings of the last three training sessions. In the pilot, new starters ask the assistant the practical questions, which systems, which templates, who signs off a credit, and get cited answers; the two seniors see only the questions the content could not answer, about a fifth of the total at first, and fix the sources. By the second cohort that share is noticeably smaller, the quiz shows where the handbook is unclear, and the seniors are back on client work. The same pipeline is then reused for the support team, which shares most of the sources.
Both, deliberately. The assistant answers questions with citations from your own material, and it also builds a learning path for each role and tracks progress. Most clients keep their LMS for compliance training and let the assistant handle the day-to-day "how do we do this here" questions.
What content do we need to have ready?
Less than you expect. A handbook, a few SOPs, product documentation and some recorded team sessions are enough to start. The assistant exposes gaps quickly because it logs the questions it cannot answer, which is also the fastest way to build a content backlog. See knowledge management for fixing the sources.
How do you stop it teaching new hires the wrong thing?
Every answer cites its source and carries a freshness date. Subject-matter experts review flagged answers and corrections feed back into the content, not just the model. Where policy matters, such as HR, security or compliance, the assistant links to the canonical page rather than paraphrasing it.
Can it see employee records?
Only what the role needs. The assistant reads the HRIS for role, team and start date to personalise the path, never salary or performance data. Permissions mirror your existing systems and are enforced at retrieval time, in line with our AI data privacy practice.
How long before the first cohort uses it?
Six to ten weeks for a first role. Two weeks for content audit and HRIS integration, three to four to build the pipeline, paths and quizzes, and two to run a pilot cohort with managers and experts in the loop before widening to other roles.
Ready to make AI work in production?
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.