How to evaluate an AI receptionist for a service business

Most service businesses do not lose jobs because their work is poor. They lose jobs because the phone rang at 6:40 pm, nobody picked up, and the caller booked the next company on the list. An AI receptionist exists to close that gap. Whether it actually does depends on decisions you make before it answers its first call.

This is the checklist we use when scoping one for a client.

1. Decide what "good" sounds like before you hear a demo

Write down, in plain language, what your best front-desk person does on a typical call: how they greet, which three questions they always ask, what they say when the schedule is full, when they put someone through to the owner. That document is the specification. Any vendor demo should be judged against it, not against how impressive the voice sounds.

If you cannot describe the job, the agent cannot do it. That is also a sign the job needs defining anyway.

2. Separate "answer" from "act"

An agent that only answers questions is a voicemail with better manners. The value is in actions: creating the job in your CRM or field-service tool, booking the slot in the real calendar, sending the confirmation text, flagging the emergency to the on-call technician.

Ask every vendor exactly which systems the agent can write to, through which integration, and what happens when that system is down. "We integrate with everything" is not an answer. "We create the job in Jobber through its API, and if the API fails we text the office and log the call" is.

3. Insist on explicit hand-off rules

The best agents know when to stop. Define the rules in advance:

• Keywords that trigger an immediate transfer (gas smell, flooding, chest pain, legal threat).
• Situations where the agent takes a message instead of guessing (pricing disputes, complaints, anything about an existing invoice).
• What the caller hears while being transferred, and what happens if nobody picks up.

Then test them. Call the agent and say the words. If it keeps chatting, it is not ready.

4. Check the logs, not the dashboard

Every conversation and every action should be recorded: transcript, which fields were captured, which systems were updated, and whether a human was involved. Review a week of transcripts before you trust the agent with volume. You will find phrasing to fix and questions you forgot to include.

A vendor who cannot show you raw transcripts is asking you to trust a summary of their own product.

5. Get the compliance basics right for your market

Rules differ across the markets we serve:

• In the United States, call recording consent is set state by state; several require all parties to consent.
• In Canada, PIPEDA and provincial laws apply to the personal data captured on the call.
• In the UK and Australia, recording and data-protection disclosure requirements apply, and marketing follow-ups by SMS need consent.
• In the UAE and Kuwait, data protection laws are newer and outbound calling is restricted; check before launching any outbound campaign.

The agent should identify itself as automated where required and always offer a route to a person. Pretending to be human is bad for trust and, in some places, unlawful.

6. Measure the things that pay

Vanity metrics for an AI receptionist are "calls handled" and "minutes saved". The numbers that matter are:

• Answer rate outside office hours.
• Booking rate for calls that should become appointments.
• Hand-off rate, and whether hand-offs were correct.
• Revenue from jobs that would previously have gone to voicemail.

Set these up before launch so the pilot can be judged honestly.

7. Price the whole thing

Expect three cost lines: implementation (the scripting, integrations and testing), platform and model usage (per minute or per call), and telephony. Ask for all three in writing and ask what happens to cost if call volume doubles. A cheap monthly fee with opaque per-minute charges can surprise you in peak season.

What a good pilot looks like

Two to four weeks, a limited window (for example evenings and weekends), every transcript reviewed, hand-off rules tested deliberately, and one clear comparison: bookings and answered calls in the pilot window versus the same window the month before. If the numbers are good, widen the window. If they are not, you have learned cheaply.

If you want a second opinion on a vendor proposal or a scoped pilot for your own business, request a consultation. We will tell you whether an agent makes sense for your call volume before we talk about building one.