How to Set Up an AI Receptionist: 7 Steps, Templates and a 15-Call Test Script

Most setup guides stop at "add your business details and click go live". That gets a receptionist answering the phone. It does not get one that books the right service at the right time, refuses to make up a price, and passes an upset caller to a person before they hang up.

This guide is the full job, in the order we would do it with any vendor. Each step has something you can copy: a knowledge template, booking and handoff rule tables, a 15-call test script and a launch checklist. Where it helps, we show how the step is done on DialerBee's AI receptionist, and we say plainly where you would use your phone company or another tool instead.

What You Need Before You Start

• Last month's call list from your phone system or carrier, including missed and after-hours calls.
• Your price list and service menu, with durations.
• Your policies: cancellation, deposits, late arrivals, parking, what to bring.
• The calendar you book from, and who owns it.
• One person who will take handed-over calls and own the weekly review.

If you are not sure an AI receptionist suits your calls yet, the AI receptionist readiness assessment takes a few minutes and works for any vendor.

Step 1: Decide the Receptionist's Job

Write one sentence that says what the receptionist is for. For example: "Answer every call to the main number when the front desk is busy or closed, book, move and cancel appointments, answer questions about services and prices, and pass anything else to a person." Everything later in the setup is checked against that sentence.

Then answer three questions.

Which calls?

Sort last month's calls into types and count them. A worked example for a fictional salon with 400 calls a month:

Here 340 of 400 calls (85%) are routine. Start with those three types and nothing else. It is easier to add a call type after two clean weeks than to win back callers after a bad first week.

Which hours?

Pick one of three coverage models: overflow (the AI answers when your team is busy or does not pick up), after hours (the AI answers when you are closed), or always (the AI answers first and hands over). Overflow plus after hours is the most common starting point, because your team keeps the calls they are already answering.

Which languages?

List the languages your callers actually use, from staff memory or call notes, and test each one in Step 5. DialerBee's receptionist speaks 11 languages: English, Arabic, Spanish, French, German, Italian, Turkish, Hindi, Urdu, Portuguese and Indonesian. Arabic is handled as dialects, including Gulf, Levantine and Egyptian, so a caller in Dubai and a caller in Amman both hear something that sounds local.

Step 2: Write the Knowledge (FAQ, Prices, Policies)

This is how you "train" an AI receptionist. You are not teaching a model. You are writing the facts it is allowed to say. Three rules make the difference:

1. Write answers, not marketing. "We are open Monday to Friday 9:00 to 18:00 and Saturday 9:00 to 13:00, closed Sunday" beats a paragraph about your passion for service.
2. Keep prices in a list, not in sentences. One service, one duration, one price per line. Prices buried in prose are where receptionists, human or AI, get things wrong.
3. Say what you do not do. "We do not do home visits" stops the receptionist guessing.

Copy this template into a document and fill in the right-hand column. The middle column is an example for a fictional salon.

Find your top 20 questions in your call list, your inbox and your WhatsApp chats, not in your head. The questions customers actually ask are rarely the ones on your website's FAQ page.

On DialerBee, you set this up in Agent Studio. You give the agent a persona (name, tone and pace; the voice is Layla, a female voice), upload your documents to its knowledge base, and it answers from them by semantic search, so a caller who phrases a question differently still gets the right passage. Prices go in your service catalogue, and only prices in the catalogue can be spoken: if a price is not there, the agent does not invent one. You can start from an industry template rather than a blank agent. See Agent Studio for the full list of settings.

Step 3: Define Booking Rules and Handoff Rules

Two short rule sets decide most of what can go wrong on a call. Write them down and agree them with whoever runs the front desk.

Booking rules

On DialerBee, the agent checks free-busy on the connected calendar before offering a time, then books, reschedules or cancels during the call. Google Calendar and Microsoft 365 sync at organisation scope and to each staff member's own calendar. The service catalogue carries each service's duration and resources, and capacity rules refuse a booking when a service is full instead of double-booking it. The agent only tells the caller a booking is made once the system has confirmed it, and a confirmation can go out by SMS or WhatsApp after the call. More on the booking side: appointments and reminders.

Handoff rules

An AI receptionist is useful because it knows its limits. Decide which calls go to a person, who that person is, and what happens if nobody picks up.

On DialerBee, the handoff goes to your own team in the same system: inbound queues that route by FIFO, round-robin or longest-idle, the agent desktop, and the same customer history for AI and human calls. If a queue does not answer within its ring timeout, the caller hears your own announcement and a nominated manager gets an SMS naming the caller. See missed-call recovery, and how to set up an inbound queue and route a number for the queue itself.

Step 4: Choose and Connect the Phone Number

The receptionist needs a number that rings it. There are two ways to get one, with any vendor:

• Use a number the vendor provides, then forward your existing business line to it (Step 6).
• Bring your own: if your number already arrives over a SIP trunk or a PBX, route it to the receptionist directly, with no forwarding.

On DialerBee the two routes work like this:

For the self-serve route, follow how to connect your SIP trunk first, then add the number in Inbound DIDs. Route Type offers Queue, Agent, Campaign, IVR and AI Agent; pointing a number at the AI agent is the same action as pointing it at a queue, and Bulk Update changes the routing of many numbers at once.

One honest limit: DialerBee routes a number to the AI agent whenever that route is set. If you want the AI to answer only after hours or only on overflow, do that with forwarding rules on your existing line or your PBX (Step 6), and keep your staff's number as the first destination.

Step 5: Test With Scripted Calls

Do not test by chatting to it until it feels fine. Use a script, have two or three people make the calls, use a real calendar, and score each call pass or fail against a written expectation. These 15 calls cover what goes wrong most often.

After each round, check three things, not just the call: the calendar, the confirmation message, and the transcript. When a call fails, fix the source (the document, the price list or the rule), then run the failed call again. Aim for 15 out of 15 before launch, and treat calls 7, 9 and 11 as must-pass.

On DialerBee, you can place test calls from Agent Studio and try the agent in a test simulator before any real caller reaches it. Every call leaves a result with the recording, the transcript and the outcome, so you score the test against what happened rather than what you remember. Changes are saved as agent versions you publish, and you can roll back a version that made things worse.

Hear what a scripted test sounds like

This recording covers three rows of the script: opening hours, a price question and a request for someone to call back. AI voice agent demo — simulated caller: the replies come from DialerBee's production voice agent, the caller is a synthetic voice and the clinic is fictional.

Listen for the things your script checks: does it give the hours exactly, does the price come from the business's own list, and does it take the callback details and say what happens next? Read the transcript, and two more demo calls, on the AI voice agent page.

Step 6: Go Live With Call Forwarding

If your business number stays with your current phone company, you go live by forwarding it to the receptionist's number. Forwarding is general telecom practice, set on your phone line, not in the AI product. There are three useful modes:

On most GSM mobile networks, ##002# cancels all forwarding; in North America, landlines commonly use *72 to switch forwarding on and *73 to switch it off (codes as listed in Wikipedia's call forwarding article, checked October 2026). Codes and support vary by carrier and country, and business lines on a PBX use the PBX's own forwarding rules, so confirm with your provider. Forwarded calls may also be charged by your carrier.

A safe launch sequence:

1. Day 1: switch on no-answer and busy forwarding only, during opening hours, with staff told what to expect.
2. Days 2 to 7: read every transcript daily and fix the knowledge the same day.
3. Week 2: add after-hours forwarding (set "always" when you close, cancel it when you open, or use your PBX's time rules).
4. After two clean weeks: consider AI-first answering or a new call type.

Call the business number yourself from an outside phone after every change. Forwarding that was set but never tested is the most common reason a "live" receptionist never receives a call.

Step 7: Review Transcripts Every Week

An AI receptionist does not get better by itself. It gets better because someone reads its calls and fixes what it was given. Book 30 minutes a week and work through the same list:

1. Read every handed-over call. Was the handoff right? Could a better answer in the knowledge have finished the call?
2. Read every call that ended without an outcome: no booking, no answer, caller hung up.
3. Collect new questions and add a one-line answer for each.
4. Check the calendar against what the receptionist said it booked.
5. Check prices after any change to your price list.
6. Re-run the test script after every change, at least calls 6, 7, 9 and 11.

On DialerBee, every AI call leaves the recording, the transcript and the outcome, pushed to your CRM and visible in the same customer history as human calls, and every call is scored automatically, so the review covers all calls rather than a sample. A/B experiments let you run two versions of an agent against real traffic and compare them, supervisors can listen live and take over a call, and kill switches stop one agent or all AI calling from your own console if something goes wrong.

AI Receptionist Launch Checklist

• One-sentence job description written and agreed
• Call types for launch chosen from last month's real calls
• Coverage model chosen: overflow, after hours or always
• Languages listed, each with a native-speaker tester
• Knowledge document filled in, including "things we do not do" and "what to say when it does not know"
• Price list complete, one service per line
• Calendar connected; service durations and resources set
• Booking rules written: window, details collected, full-slot behaviour, changes, confirmation
• Handoff rules written, with a named person and a no-answer fallback
• Phone number connected and routed; outside test call answered
• 15-call test script passed, including the unlisted price, unsupported question, transfer and complaint calls
• Forwarding set and tested from an outside phone
• Staff briefed on what the receptionist handles and what it passes to them
• Recording and retention settings checked against your own policy
• Weekly 30-minute review booked, with an owner
• You know how to switch it off

What Does It Cost to Run?

Setup is mostly your time. Running cost depends on the vendor's pricing model. On DialerBee, the AI receptionist is the AI voice agent at $0.60 per started connected minute on managed AI, added to a Professional plan at $129 per agent per month, and carrier minutes are billed by your carrier. With your own model keys, a flat monthly fee replaces the per-minute rate. For the salon above, 340 routine calls at 2 minutes each is 680 minutes, or $408, plus one $129 seat: $537 a month before carrier minutes. That is an example, not a quote. For other vendors' prices and hidden fees, read how much an AI receptionist costs, and see the pricing page for ours.

If you are still deciding between an AI receptionist, an IVR menu and a human answering service, this comparison covers the trade-offs.

Frequently Asked Questions

How long does it take to set up an AI receptionist?

The software part can take an afternoon. The work that decides quality takes longer: writing the knowledge, agreeing booking and handoff rules with your team, and running test calls. For a small business, plan about a week from first draft to live calls, then a few weekly reviews before you widen what it handles.

How do I train an AI receptionist?

You do not train a model; you give the receptionist the right material. Write short, factual answers to the questions callers actually ask, keep prices in a price list rather than in prose, set clear booking and handoff rules, then test with scripted calls and fix the source document whenever an answer is wrong.

Can I keep my existing business number?

Usually, yes. Either forward your existing line to the AI receptionist's number when you are busy, do not answer or are closed, or, if your number already arrives over a SIP trunk or PBX, route that number straight to the AI. On DialerBee, pointing a number at the AI agent is the same action as pointing it at a queue.

Do I need my own SIP trunk to use DialerBee's AI receptionist?

For self-serve sign-up and the 14-day free trial, yes: you connect your own SIP trunk or PBX and your carrier bills its minutes. If you do not have one, book a demo and DialerBee sets up the phone number for you.

What should an AI receptionist not handle?

Complaints and upset callers, emergencies, payment disputes, medical, legal or financial advice, emotionally sensitive calls, and any caller who asks for a person. Write these down as handoff rules before launch so the receptionist passes them on instead of trying to finish them.

How do I test an AI receptionist before going live?

Run a written test script of at least 15 calls covering bookings, changes, cancellations, prices, a question it should not answer, a request for a person, an interruption and a second language. Make the bookings against a real calendar, then check the calendar, the recordings and the transcripts rather than trusting how the call felt.

What if the AI receptionist gives a wrong answer?

Find the call, read the transcript, and fix the cause: usually a missing or ambiguous line in your knowledge document, or a missing price in your price list. On DialerBee you change the document or the catalogue, publish a new agent version, and can roll back if the change makes things worse.

Can an AI receptionist answer in more than one language?

Many can. DialerBee's speaks 11 languages, and handles Arabic as dialects, including Gulf, Levantine and Egyptian. Test each language you need with a native speaker, in the accent your callers actually use, before you rely on it.

The Bottom Line

Setting up an AI receptionist is less about the software than about four documents: what it is for, what it knows, when it books and when it hands over. Write those, test them with a script, go live on forwarding one mode at a time, and review the calls every week. If you want a receptionist that books into your real calendar, quotes only your prices and hands callers to your team in the same system, book a demo and we will set up the number for you, or start a 14-day free trial with your own SIP trunk or PBX. See everything it does on the AI receptionist page.