Use case assessment
Is an AI onboarding assistant worth building?
This is knowledge search with a smaller audience. Build that instead and point new joiners at it.
New joiners ask the same questions as everyone else, earlier and more often. A separate system duplicates infrastructure and splits the content maintenance.
What it usually means
An assistant answering new joiners' questions: policies, tools, processes, who does what.
Is it genuinely AI?
Yes — but it is the same AI as internal knowledge search, pointed at the same documents, for a subset of users.
New joiners are not asking different questions. They are asking questions everyone has, earlier and more often, because they have not yet learned who to ask. That is a discoverability problem, and it is the one knowledge search solves.
Why building it separately usually costs more
Two systems means two retrieval pipelines, two evaluation sets, and content maintained in two places — which in practice means maintained in one and stale in the other.
It also means the onboarding assistant is retired when the onboarding project ends, and nobody notices until a new joiner gets an answer about a tool that was decommissioned last year.
What the simpler version looks like
Internal knowledge search across your existing documentation, with an onboarding-specific entry point: a curated set of starting questions on the new joiner's first-week page. Same infrastructure, same content, different front door.
And before either: a checklist. A substantial share of onboarding questions are not questions at all, they are steps nobody wrote down.
What it costs to run
If it is a front door onto knowledge search, close to nothing beyond that system. If it is separate, the full cost of a second retrieval system for a fraction of the users.
Where it sits under the EU AI Act
Not an Annex III high-risk use. Article 50 disclosure applies to the conversational interface.
One thing to keep out: anything inferring how a new joiner is coping. Emotion inference in workplace contexts is a prohibited practice under the AI Act, in force since February 2025, and sentiment features get added to HR tooling routinely.
When it is worth building
As a scoped entry point onto internal knowledge search, once that exists.
Not as a standalone project. If onboarding is genuinely painful, the cause is usually that nothing is written down — and no retrieval system answers questions from documents that do not exist.
Common questions
Is an onboarding assistant different from internal knowledge search?
Not technically. New joiners ask the same questions as everyone else, earlier and more often, against the same documents. A separate system duplicates the retrieval pipeline and splits content maintenance, which usually means one copy goes stale.
What should you do before building an onboarding assistant?
Write the checklist. A substantial share of onboarding questions are not questions, they are steps nobody documented. Retrieval cannot answer from documents that do not exist.
Can an onboarding assistant track how new joiners are doing?
Not by inferring emotion. Emotion inference in workplace contexts is a prohibited practice under the EU AI Act, in force since February 2025, and sentiment features are routinely added to HR tooling without that being reconsidered.
Lifting AI adoption without retraining the model — where a capability sits deciding whether it gets used, rather than what it can do.