At 9:47 on a Tuesday night, a homeowner calls because the furnace won't come on. Without anyone to answer, the call goes to voicemail, and there's a good chance they call the next company on the list. With an AI receptionist, the call is answered on the first ring. It confirms who's calling, asks whether anyone smells gas or feels unwell, offers tomorrow 8 to 10 or 10 to 12, books the one they pick, and the job is on your board before your dispatcher's alarm goes off.
What can an AI receptionist handle, and what should it hand off?
The useful line is between routine calls, which follow a pattern, and calls that need judgment or carry risk.
| Call type | AI receptionist | Why |
|---|---|---|
| Existing customer booking a repair or tune-up | Handle | Clear pattern, real windows available |
| Reschedule or confirm an appointment | Handle | Low risk, saves office time |
| Hours, service area, from-prices you publish | Handle | Fixed answers |
| Brand-new caller wanting service | Handle with guards, or pass to staff | Address, area and callback number need checking |
| Gas smell, CO alarm, flooding, sparks, smoke | Hand off after safety instructions | Life safety, not a booking |
| Quote for a new system | Book an estimate visit | Needs a site visit, not a phone price |
| Complaint or billing dispute | Hand off | Needs a person with authority |
| Caller asks for a person | Hand off, always | Nobody should be trapped in a bot |
AI receptionist vs answering service vs voicemail vs CSR
| AI receptionist | Human answering service | In-house CSR | |
|---|---|---|---|
| Hours | Any time, as configured | Any time, by contract | Business hours, plus on-call |
| Books real windows | Yes, if connected to the schedule | Usually takes a message | Yes |
| Judgment on unusual calls | Limited; should hand off | Moderate | High |
| Cost model | Per minute of use | Per minute or per call | Wages and benefits |
| Knows the customer's history | Yes, if connected and the caller is verified | Rarely | Yes |
Voicemail is the baseline all three are measured against. The CSR page covers the in-house role. For the after-hours side of the decision, see what the data says about after-hours calls.
What makes an AI receptionist safe to use?
AI models can be persuasive and wrong at the same time. The safeguards that matter most are the ones enforced by the system around the model, not by instructions in a prompt. Our article on why prompts are not policies explains the difference.
What to check before you trust one with your phones
- Emergency words like gas, carbon monoxide and flooding trigger safety instructions before the AI responds
- Pressing 0 or asking for a person always transfers or takes a message
- Callers verify an address detail before anything on file is read back
- It can't claim a booking that wasn't actually written to the schedule
- It only offers windows a qualified tech can actually reach
- Prices are limited to the ones you publish
- Every call is recorded or transcribed, with an outcome you can review
- A kill switch sends calls back to your team immediately
What are the legal rules?
Inbound answering isn't banned or specially licensed at the federal level, but several rules touch it:
- Outbound AI calls. In February 2024 the FCC unanimously ruled that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act, so they're held to the same consent rules as other robocalls.1 That matters if an AI system calls leads back, not for answering inbound calls.
- Disclosure. In August 2024 the FCC also proposed requiring callers using AI-generated voices to disclose that at the start of outbound calls.2 Disclosing on inbound calls too is simply good practice.
- Recording and transcripts. Most AI receptionists transcribe calls. State laws differ on whether one party's or every party's consent is needed, so play a notice that fits the strictest state you take calls from.
This is general information, not legal advice.
How do you measure an AI receptionist?
Judge it the way you'd judge a CSR: on outcomes, not minutes.
Metric
AI receptionist booking rate
The share of opportunity calls answered by the AI that end with a real appointment on the schedule.
Appointments written by the AI ÷ opportunity calls it answered × 100
Where the numbers come from: Call log with outcomes, matched to appointments
Cost per booked call puts the per-minute price in context:
Formula
| Illustrative month | Value |
|---|---|
| Calls answered by the AI | 200 |
| Average call length | 3 minutes |
| Cost per minute (example) | $0.15 |
| Total cost | $90 |
| Calls that booked | 100 (50%) |
| Cost per booked call | $0.90 |
Compare that with the booking rate and average ticket of the same call types when your team answers. For a fuller cost comparison against hiring, see AI agent vs a part-time CSR.
How the TradesWork AI receptionist works
The TradesWork AI receptionist is in early access: it needs a phone number on TradesWork and the Twilio ConversationRelay voice service switched on for your account, and AI turned on in settings. You choose after hours only, after hours plus calls the office doesn't answer, or every call. Callers confirm a house number or ZIP before anything on file is read back. Gas, CO, flooding, sparks and smoke are matched in code before the AI model runs, so the caller hears safety instructions first, then a warm transfer to your on-call person with a whispered summary. Pressing 0 or asking for a person always transfers.
Booking claims are checked against the schedule, and windows come from the same capacity engine as the dispatch board. New callers can book only with a real caller ID, a ZIP inside your service area and a confirmed address; otherwise the call becomes a booking request for the office. Our own estimate is about $0.14 per minute on the faster model and about $0.20 on the most capable one. It doesn't make outbound calls, runs one receptionist per company, and its calls can be graded on the same call scorecard as your CSRs.
Footnotes
-
FCC Makes AI-Generated Voices in Robocalls Illegal, Federal Communications Commission, news release, February 8, 2024. ↩
-
Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, FCC 24-84, Federal Communications Commission, notice of proposed rulemaking, adopted August 7, 2024. ↩
