Is an AI Receptionist Worth It? The 30-Day ROI Check for Small Shops

By Dan Miller · Last updated November 11, 2026 · Prices checked November 10, 2026

The first bill arrives about a month after you switch the phones on. $79, $149, $270 — whatever your plan costs, the same question follows it: did that money come back? Most owners answer by feeling. The month felt busy, so it must be working. Or the month felt quiet, so it must be a waste. Both answers are guesses, and both are usually wrong.

This is a 30-day check you can run in one evening, from numbers you already have: your call log, your calendar, and your invoices. It gives you one number that decides almost everything — your cost per booked job — and a clear keep, fix, or cancel verdict. No new software, no spreadsheet degree.

The short version

• Pull five numbers from your call log, your calendar, and your invoices — the worksheet below shows where each one lives.
• Cost per booked job is the number that decides: monthly bill divided by the jobs the AI booked. Under 5% of your average ticket means keep it.
• Calls answered but few bookings means fix the setup, not cancel yet — the three fixes below cover most cases.
• Zero jobs booked in 30 days, even after the fixes, means cancel it.

The 5 numbers that decide it

Five numbers, none of them hard to find. Pull them for one full month and write them down before you judge anything.

Number	Where to find it	Healthy sign
1. Calls answered	Your AI tool’s call log, total calls vs. answered calls	95% or better. Below 90% is usually a setup problem, not a tool problem.
2. Jobs booked	Your calendar or CRM, counting only real appointments	At least one most weeks. A booking rate of 15–25% of answered calls is a fair rule of thumb for home services.
3. Revenue from booked jobs	Your invoices for the jobs in row 2	Anything that dwarfs the monthly bill — see the worked examples below.
4. Cost per booked job	Monthly bill divided by jobs booked	Under 5% of your average ticket.
5. After-hours bookings	Call log filtered to outside business hours	Any number above zero is found money — before the AI, most of these went to voicemail.

Two warnings before you start. First, count a job only when it lands on the calendar. A caller who says “call me back Tuesday” is not a booking. Second, use a full 30 days. A quiet week tells you nothing — one slow week and one storm week average each other out.

Two shops, worked out in full

Shop one — a solo plumber. He takes about 120 calls a month. Roughly a third are repeat callers, so about 78 different people — inside Goodcall’s Starter plan at $79 a month for 100 unique callers, with unlimited minutes. In his first month the AI books 18 jobs. Cost per booked job: $79 ÷ 18 = $4.39. His average job is $450, so those 18 bookings are worth $8,100 on the calendar. Even if only six of those eighteen would have been lost without the AI — the rest would have called back or left a message — that is $2,700 recovered against a $79 bill. Before the AI, Invoca’s 2026 data says around 48% of calls to businesses like his went unanswered; we worked out what that costs a plumber in detail.

Shop two — a three-truck HVAC shop. It takes 150 calls a month and sits on Smith.ai’s Pro 150-call step at $270 a month. The AI books 34 jobs. Cost per booked job: $270 ÷ 34 = $7.94. The average ticket is $620, so $7.94 is about 1.3% of one job — the AI pays for itself if it books a single extra job every month, and it booked 34. Any booking that came in after hours is one the shop could not have taken at all before, and with 82% of callers saying they would try a competitor after an unanswered call (CallRail, 2025), few of them would have waited until morning.

Notice what the chart shows: the bigger shop pays more than three times as much per month, but its cost per booked job is still under $8. The monthly price is the wrong number to argue about. Cost per booked job is the number that decides.

Keep, fix, or cancel

Once you have your five numbers, the verdict almost writes itself.

Verdict	What your numbers show	What to do
KEEP	Cost per booked job under 5% of your average ticket, and at least one job booked most weeks	Leave it alone. Re-run the check once a quarter.
FIX	Calls answered, but bookings low or cost per job above 5%	Work through the three fixes below before you spend another month.
CANCEL	Zero jobs booked in 30 days after the fixes, or cost per job above 10% of your ticket two months running	Cancel. The tool is not solving the problem you have — or it is the wrong tool for your volume.

Three fixes to try before you cancel

Most “the AI does not work” stories are one of these three, and all three are free to fix.

1. Shorten the greeting. If the AI asks for a name, an address, the full story, and a preferred time before it offers a booking, callers hang up halfway. Keep it to three questions: name, service, time. Every extra question before the calendar is another chance to lose the caller.

2. Check the AI can actually book. On some tools booking starts on a higher tier — Rosie only books from its $149 Scale plan, Dialzara from its $99 Pro plan. If your plan takes messages but cannot put a job on the calendar, your bookings column will stay empty no matter how good the answering is. Either move up a tier or switch tools. Do not blame the AI for a plan limit.

3. Fix the transfers. If the AI offers to transfer urgent calls to a tech who never picks up, callers get bounced twice and give up. Set transfer hours that match real working hours, and give the AI a fallback for the rest: take a message and send a text-back instead of ringing a dead line.

Answer 3 questions → your verdict

Do not start with the verdict — start with the questions. Question 1: Did the AI book at least one job most weeks? If yes, go to Question 2. If no — zero bookings in a full month — work through the three fixes above, then run one more 30-day check. Still zero? Cancel. Question 2: Is your cost per booked job under 5% of your average ticket? If yes, keep it and stop shopping for alternatives. If no, go to Question 3. Question 3: Is the problem the plan or the setup? A plan that cannot book, or a greeting that asks too much, is a setup problem — fix it. A plan that books fine but costs more than 10% of your ticket is a pricing problem — drop to a cheaper tier or a per-caller plan that fits your volume, and check the Price Index before you switch. If the AI shape itself is wrong for you, the human answering-service route is priced in our answering-service comparison.

FAQs

How long should I wait before judging it?
A full 30 days at normal volume. A quiet week is not a verdict, and neither is one storm week. If your first month included a holiday or a shutdown, run the check on the next normal month instead.

What if most of my calls are existing customers?
The check still works, but read it differently. Repeat callers book at a higher rate, so your cost per booked job will look good — the number to watch is time saved. If the AI handles the “what time are you coming” calls, that is hours back every week even before the new bookings.

My plan is cheap — does that skew the check?
No. The check uses your actual bill, so a $79 plan that books nothing still fails and a $349 plan that books forty jobs still passes. Cheap only matters if the plan cannot book — see the fixes above.

What counts as a “booked job”?
A real appointment on the calendar with a name, a service, and a time. A callback promise, a message taken, or a quote request is not a booking. Count only confirmed appointments, or the check flatters the tool.

Run your own numbers first

Your call volume decides which plan is cheapest — the same tool costs three different shops three different amounts.

Calculate my true cost →

Prices checked November 10, 2026: Goodcall Starter $79 for 100 unique callers (Growth $129/500, Scale $299/2,000); Smith.ai AI Receptionist Pro steps $150/75 calls, $270/150 calls, $500/300 calls; Rosie Scale $149 for 1,000 minutes; Dialzara Pro $99 for 220 minutes. Call statistics: Invoca (2026), CallRail (2025), Quo (2026) — see our Statistics page. The worked examples use stated assumptions, not any single shop’s real books.

Affiliate disclosure: some links on this site are affiliate links. If you buy through them we may earn a commission at no extra cost to you.