How to train an AI receptionist on your business in an afternoon

Training an AI receptionist means giving it your business facts (services, prices, hours, locations, policies), approving the answers it is allowed to give, and telling it what to hand to a person. It is not machine learning and there is no model to tune. It is a well organised list, and most small businesses can build it in an afternoon.

Start with the questions you already answer

Write down the twenty questions your front desk answers most. For most service businesses the list is boringly stable: hours, parking, the price of the most popular service, whether you take a particular insurance or payment method, how soon someone can get in, and what happens if they cancel. Pull the wording from real voicemails and texts rather than from memory, because customers ask in their own words and the AI needs to recognise those.

These twenty become your first approved FAQs. Everything else can wait.

Load facts, not prose

An AI receptionist answers well when it has short, factual entries to work from and badly when it has a pasted brochure. For each of these, one line each:

1. Services, with duration and price, or a starting price where the real one depends on a visit.
2. Locations, with address, phone, hours and timezone.
3. Policies: deposits, cancellation window, late arrival, refunds.
4. People: who does what, and who is not bookable online.
5. Practical details: parking, entrance, what to bring, forms to fill in.

If a fact changes with the person asking ("depends on your plan"), record the rule, not a guess.

Decide what it must not answer

This list matters more than the FAQs. Medical, legal and financial questions, anything that reads like a complaint, prices that depend on seeing the job, and anything with a deadline attached should go to a person. Write the handoff message yourself: "That is one for Dana. She will text you back this morning." A generic "let me connect you" sounds like a phone tree.

Set the tone in one paragraph

Tone is not a personality. It is a paragraph: warm, brief, first names, no exclamation marks, never says "absolutely". Then test it with five questions a real customer sent last week and read the drafts out loud. If a draft sounds like a press release, shorten the paragraph.

Run it supervised for a week

Keep approval mode on, so every reply is a draft a person approves or edits before it sends. The edits are the training. Each one shows a fact that was missing or a phrasing to fix. Watch the log of questions it could not answer; every entry there is an FAQ you have not written yet. After a week of clean drafts, let it send on its own for the simple categories and keep the rest supervised.

Keep it current

Assign one person. When a price changes, the knowledge entry changes the same day. Once a month, spend fifteen minutes on the gaps log. That is the whole maintenance plan.

Frequently asked questions

Do I have to write all the FAQs myself?

Mostly, yes. Some products draft FAQs from a website; in Frontiva you write them, with your industry template as a starting point and the gaps log showing what customers actually ask. Either way, approving is the part you cannot skip, because the AI will only answer from what you approved.

How long before it is useful?

An afternoon to load facts, then a week of supervised replies. Businesses that skip the supervised week spend longer fixing answers than they saved.

What happens when it gets something wrong?

In approval mode, you edit the draft and the customer never sees the mistake. Afterwards, fix the fact it used, not the reply. A corrected fact fixes every future answer; a corrected reply fixes one.

What Frontiva does here

Knowledge is where facts live in Frontiva: services, hours, staff, policies and promotions, seeded from your industry template, plus the FAQs your team approves, with a counted log of every question the agent could not answer. The agent answers only from what is approved, and approval mode holds every draft for a person until you widen its autonomy. For the background, start with what an AI receptionist is.