All posts

· 6 min read

How to write a knowledge base your AI receptionist can actually use

An AI receptionist is only as good as what you tell it. A practical guide to writing services, prices, hours and policies it can answer from — and keeping them current.

An AI receptionist answers questions the same way a new hire does: from whatever you gave it to work with. Hand a new hire a crumpled price list from two years ago and a vague “just use common sense” and you will get vague, out-of-date answers. Hand them a clear, current, one-page brief and they sound like they have worked for you for years. The AI is no different.

The good news is that writing a usable knowledge base is not a big job. Most service businesses can do it properly in an afternoon, because the raw material already exists — it is in your head, your booking system and the answers you give on the phone every day. This playbook covers what to include, how to phrase it, what to leave out, and how to keep it from going stale.

Start with the questions callers actually ask

Before you write anything, list the ten questions you hear most on the phone. For most service businesses the list is boringly predictable: what do you charge, are you open Saturday, do you come to my suburb, how long does the job take, do I need to do anything before you arrive, can I get a quote first, do you take card. Those ten questions are your table of contents. Everything else is secondary.

Then answer each one the way you would say it out loud — one caller, one question, plain words. “A standard service is $180 and takes about an hour” beats a paragraph of qualifications. If an answer genuinely depends on something, say what it depends on: “Rewiring is quoted after an inspection, because it depends on the size and age of the property.” That is still a complete answer. “It varies” is not.

Services and prices: be specific, then stop

List each service you actually want booked, with a short plain-English description and a price or a price range. If you cannot give a number, give the process: “fixed quote after a ten-minute assessment, no charge for the assessment.” Callers do not need certainty as much as they need a straight answer about what happens next.

Name your services the way customers say them, not the way your invoice template does. A caller asks about “a leaking tap,” not “minor domestic plumbing maintenance.” If your industry uses jargon, include the plain version alongside it so the AI can match what the caller says to what you offer.

Resist the urge to list everything you have ever done. If you technically can do a job but do not want more of it, leave it out or state the boundary directly: “We don’t service commercial kitchens.” A knowledge base is a menu, not a memoir.

Hours, service area and policies

Write your hours precisely, including the exceptions: weekend hours, public holidays, the last booking slot of the day. “Open until 5pm, last appointment 4pm” prevents a caller being booked into a slot you will resent. If you offer emergency or after-hours work at a different rate, say so and say the rate.

For service area, name suburbs or regions rather than a radius. Callers say suburb names; nobody knows whether they live thirty kilometres from your depot. Include the awkward edges — the suburbs you will do for a callout fee, and the ones you simply do not go to.

Policies are where a receptionist — human or AI — earns their keep, because these are the answers owners hate repeating: deposits, cancellation notice, payment methods, whether someone needs to be home, what happens if it rains. Write each one as a short rule with the reason attached. “We ask for 24 hours’ notice to cancel, because that slot can usually be refilled with a day’s warning” sounds fair. A bare “24-hour cancellation policy” sounds like fine print.

What to leave out

Leave out anything you would not want said verbatim to a customer. The knowledge base is a script source, not an internal wiki — no margins, no supplier gripes, no “we usually discount if they push.” If a discount exists only when you personally approve it, it does not belong in the knowledge base; it belongs in a human conversation, and the right move is for the call to be escalated to you.

Leave out answers you are not sure of. A wrong price stated confidently is worse than “I’ll have someone confirm that and call you back.” It is fine — sensible, even — for some questions to end in a handover to a human. Decide which ones, and say so explicitly: complex quotes, complaints, anything involving a dispute.

Keep it current, and let it answer the phone

A knowledge base decays the same way a price list does. Put a fifteen-minute review in your calendar for the first Monday of each month, and update it immediately whenever a price changes, hours change, or you catch an answer that made you wince. The test is simple: if you overheard the answer given to a real caller, would you cringe? If yes, fix the source, not the symptom.

This is exactly the document Estric runs on. Estric answers your existing phone number around the clock with a natural voice, answers questions straight from the knowledge base you write, books appointments during the call with an SMS confirmation, and hands off to a human when a question falls outside it. Every call is recorded and transcribed in one dashboard, so the questions you did not anticipate become next month’s knowledge base updates. Write the brief once, and it gets used on every call.

Put Estric on your phone

Estric answers every call, books the appointment, and texts the customer back — 24/7, on the number you already have.

Get started