The Overnight Expert
Your best employee took 18 months to learn the job; an AI system can take in everything written down about it in minutes. That written half is what AI retrieval handles well. The judgment half stays with your people, and the design has to know the difference.
Your best employee took 18 months to learn the job. An AI system can take in everything written down about it in 18 minutes, and that is only part of what they know.
The written part is what AI retrieval does well; the rest is judgment, and it stays with people. Design for both and a new hire gets answers on day one without interrupting anyone senior. The 18 minutes is illustrative; the time depends on how much you have written down.
What 18 Minutes Buys You
Give an AI system your policies, manuals, FAQs, past decisions and client notes, and it can find any passage in them by meaning as well as by keyword. Ask about your refund policy or your escalation procedure and it surfaces the right document and clause.
That is valuable: years of knowledge sitting in folders nobody opens and wikis nobody reads become available to anyone who needs it. The system does not understand your business, though. It can find what you have written down about it, which is a different thing.
What It Cannot Learn From Documents
Illustrative: how a firm's knowledge might split. The shares show the argument, not a measurement.
The largest share is facts, processes, policies, templates and decision logs, and retrieval handles it well. The highlighted share is where AI projects go wrong: edge cases, exceptions and the “technically yes, but no” situations an experienced employee handles by instinct. “Our policy says 30 days, but this client is our largest account and the delay was our fault, so we make an exception.” No document holds that reasoning.
The last share needs a person in the room: relationships, and knowing which client is quietly unhappy before it becomes a complaint. Treating the written share as the whole is how trust in these systems breaks.
Retrieval Supports the Decision
The AI answers every client question on its own: it retrieves the policy, writes a reply and sends it. It works until it meets an edge case or makes a confident error, and the client pays for it.
The AI puts every relevant document in front of a person: the policy, the client history, similar past cases, the exceptions that applied. The person reads it in seconds, makes the call and replies.
The AI does the retrieval, at any hour, and the person makes the judgment.
Onboarding: Where It Pays First
In most onboarding, the bottleneck is how often the senior team can afford to be interrupted.
New hire starts
Day one, with a list of questions
Asks the AI
Any process, policy, acronym or precedent that is written down
Gets a sourced answer
With the source named, so they can check it
Productive sooner
Seniors are interrupted only for what is not written down
When Someone Leaves
When an expert hands in their notice, the usual response is “we need to document everything they know before they go”, and two weeks is not enough. Retrieval helps here only if writing things down was already a habit.
AskBase, our knowledge engine, is available now and does the retrieval half: it answers from your own documents in Slack and a web app, scoped to who is asking, and says what it searched when it finds nothing. It publishes no accuracy percentage. Instead it grades every answer high, medium or low, so the person reading knows how much checking it needs. The judgment half stays with your people.
Start with the questions new hires ask in their first month and the documents that answer them. Where an answer lives only in someone’s head, get it written down while that person is still here.
What does your next hire need to ask on day one?
Book Free Assessment