AI receptionist for small business: how it works, prices and limits

An AI receptionist is a voice agent connected to your business phone number. It picks up when you can't, answers common questions from information you give it, collects the caller's details, books an appointment or sends you a summary, and hands off to a human when it should. For most small businesses the software costs between about US$30 and US$300 a month, plus per-minute or per-call charges past the included volume. It works well for routine calls and badly for anything emotional, unusual or legally sensitive.

Here is how the call flow works, what to put in the script, what you will actually pay, and where the limits are.

How a call flows

Think of it as a short decision tree that runs on every call.

1. The call arrives. Your number either rings your team first and forwards to the AI after a few rings, or goes straight to the AI after hours. Most businesses start with "after hours and overflow only".
2. Greeting and disclosure. The agent says who it is and, if you record, that the call is recorded.
3. Intent. It works out why the person is calling: new quote, existing job, billing, emergency, something else.
4. Routine path. For questions it can answer from your approved facts (hours, service area, what you do and don't do, how pricing works), it answers.
5. Qualification. For a new enquiry it asks your three to five qualifying questions: what, where, when, budget range if relevant.
6. Action. It books a slot in your calendar, creates a contact in your CRM, or promises a callback within a set time.
7. Escalation. For emergencies, upset callers or anything outside its rules, it transfers to a person or takes a message flagged as urgent.
8. Wrap-up. You get a text or email with a summary and the recording or transcript. The caller can get a confirmation text.

The quality of the outcome depends far more on steps 4 to 7, the rules you write, than on which voice the vendor uses.

A sample script for a service business

This is the kind of brief you would load into the agent. Adapt it to your trade.

Role: You answer calls for Lavoie Renovations, a renovation contractor. You are an automated assistant and you say so if asked. Greeting: "Thanks for calling Lavoie Renovations. I'm the automated assistant. This call is recorded so we can follow up accurately. How can I help?" What we do: kitchen and bathroom renovations, basements, flooring. What we don't do: roofing, electrical-only jobs, insurance claims. Service area: list of cities. Outside it, say we can't help and suggest they check a directory. Pricing: never quote a price. Say every quote follows a free on-site visit. For new projects, ask: type of project, address or city, ideal start date, best callback number and email. Booking: offer two available site-visit slots from the calendar. Confirm by text. Transfer immediately if: water is actively leaking, the caller mentions injury, the caller asks for a human twice, or the caller is a current client with a complaint. Never: promise dates, discuss other clients, give legal or safety advice.

Notice how much of the script is about what the agent must not do. That is where most of the work goes.

What it costs

Prices below come from vendor pricing pages in October 2026. Most are in US dollars, so add the exchange rate if you pay from a Canadian account.

What actually drives your bill:

• Call volume and length. A clinic with 400 short calls a month and a contractor with 60 long ones need different plans.
• Billing unit. Per minute, per call and per unique caller give very different totals for the same business. Run last month's call log through each model before choosing.
• Integrations. Calendar and CRM booking often sit on higher tiers.
• Set-up. Writing the script, connecting the calendar and testing is the part people underestimate. Done in-house it takes a few days; an agency typically charges a one-time fee for it.

A quick test of value: take the number of calls you miss in a month, multiply by the share that are new business, and by what a new customer is worth. If one recovered job a month pays for the year, the decision is easy. If most of your missed calls are existing clients wanting updates, a better voicemail and a callback routine may do the job for less.

What it can't handle well

Be honest about these limits before you switch it on.

• Emotion. Angry, grieving or panicking callers need a person. Set clear transfer rules.
• Real emergencies. For anything involving safety, the agent should transfer or tell the caller to contact emergency services, never troubleshoot.
• Complex pricing. If every quote depends on a site visit, don't let the agent improvise numbers.
• Accents, noise and bad lines. Recognition has improved a lot, but job-site noise and spotty mobile connections still cause misunderstandings. Read transcripts weekly in the first month.
• Made-up answers. Language models can state wrong information with confidence. In Canada, the BC Civil Resolution Tribunal held Air Canada liable in 2024 when its website chatbot gave a customer wrong information about bereavement fares, and rejected the argument that the bot was responsible for its own words. Your AI speaks for your business.
• Callers who refuse bots. Some people hang up. Offer "say 'person' at any time" and make it work.

Privacy and disclosure in Canada

If the agent records or transcribes calls, Canadian privacy law applies. The Office of the Privacy Commissioner says businesses must tell callers a call is being recorded, explain the purposes clearly (not just "quality assurance" if you also use it for marketing), and offer an alternative when recording isn't essential to the service. Callers can ask to access their recordings, and you need safeguards and a retention limit.

In Quebec, Law 25 adds a duty to inform people when a decision about them is made exclusively by automated processing. A receptionist that only books appointments is unlikely to trigger it, but one that accepts or rejects applicants on its own could. Keep a human in the loop for decisions that matter, and check your vendor's data location and retention settings.

When an AI receptionist is the wrong choice

• You get fewer than a handful of calls a week. Forward to your mobile and use a good voicemail.
• Most callers are existing clients with complex, ongoing files.
• Your business depends on a warm, personal first impression that a person delivers better, such as high-end services where every call is a sales conversation.
• Nobody on your team will review transcripts and adjust the script. An unattended agent drifts out of date as soon as your services or prices change.

Setting one up without regret

1. Start with after-hours and overflow only.
2. Write the "never" and "transfer" rules first.
3. Connect booking to a calendar with buffers so it can't double-book you.
4. Send every summary to one shared inbox or CRM pipeline.
5. Review 20 calls in week one and 20 more in week four, then fix the script.
6. Track booked appointments from the agent separately so you can judge it on results, not on how human it sounds.

Where PNE fits

Our AI growth systems connect the assistant to the parts that make it useful: a CRM pipeline, calendar booking and follow-up by text and email. For Kaltico, a construction and consulting firm, we set up an AI assistant that answers first enquiries and FAQs and books discovery calls around the clock, feeding a CRM that follows up automatically. You can see the full project in the Kaltico case study.

Sources

• AI Receptionist pricing — Smith.ai
• Pricing — Goodcall
• Pricing — Dialzara
• Pricing — Retell AI
• Pricing — Ruby
• HighLevel AI Products Pricing — HighLevel Support Portal
• Recording of Customer Telephone Calls — Office of the Privacy Commissioner of Canada
• Airline ordered to compensate a BC man because its chatbot provided inaccurate information — Mondaq / Dentons