In short: an AI receptionist is worth it when the revenue in your missed calls exceeds its monthly cost — and for appointment-based businesses that miss even a handful of calls a week, it usually does, comfortably. The arithmetic takes five minutes: missed calls per month × the fraction that were real enquiries × your conversion rate × average job value, compared against the subscription. If that number isn’t clearly bigger than the cost, don’t buy one. Here’s how to run it honestly — including the situations where the answer is genuinely no.
I run AI4SMB, an Australian AI automation agency, so I have an obvious interest in you answering “yes”. Which is exactly why this post includes the “no” cases — an AI receptionist sold to a business that doesn’t miss valuable calls becomes a cancelled subscription and a story about AI being overhyped. Neither helps me.
What’s a missed call actually worth?
Work through it with real numbers from your own business, not vendor examples:
- Missed calls per month. Your phone system or mobile carrier logs these — after-hours calls, rings that hit voicemail, abandoned holds. Most owners guess low; pull the actual log. If you run a PBX like 3CX, the report takes two minutes.
- How many were real enquiries? Not all of them — strip the telemarketers and wrong numbers. Half is a common honest estimate; use your own.
- Your conversion rate on answered enquiries. If you normally win, say, 40% of genuine enquiries you actually speak to, use that. A caller who reached voicemail converts far lower — for urgent trades and new-patient calls, often near zero, because they simply ring the next listing.
- Average job or patient value. For a tradie, the average invoice. For a dental practice, first-visit value at minimum — lifetime value if you’re being realistic, since a recovered new patient stays for years.
Multiply those together and you have the monthly revenue currently leaking through your phone. Compare it with what an AI receptionist costs — our honest cost breakdown covers the range and what drives it. For most trades and clinics the leak is a multiple of the subscription, which is why the category exists.
Worked example (deliberately conservative)
A sole-trader electrician misses 20 calls a month. Ten are genuine enquiries. Reaching voicemail, maybe one leaves a message and converts. If an AI answers instead and even a third of those ten book — that’s roughly three extra jobs a month. At a few hundred dollars average invoice, the recovered work pays for a typical subscription several times over — and that’s before counting the after-hours emergency that’s worth triple the day rate.
Run the same shape with your numbers. If it doesn’t clear the bar by a comfortable margin, that’s your answer.
When an AI receptionist is NOT worth it
Honesty cuts both ways. Skip it if:
- You don’t actually miss calls. A staffed front desk with genuine coverage and no after-hours demand has nothing to recover. Check the call log before assuming.
- Your callers aren’t revenue. If the phone is mostly suppliers and existing-customer admin, recovered calls don’t convert to dollars. The maths above will tell you this.
- Your call flow is genuinely complex and low-volume. Ten calls a week that each need deep human judgement is a job for a human and a good voicemail script, not automation.
- You’d set it and forget it. An unmonitored AI answering badly costs more than voicemail — in reputation. If nobody will review transcripts, don’t deploy one. (This is why we only run them monitored, like infrastructure.)
Beyond the receptionist: measuring AI ROI properly
The same discipline applies to any AI investment: define the metric before you buy (missed-call recovery, admin hours saved, response time), measure the baseline first, and review against it after 60–90 days. The Australian government publishes a sensible framework for this — see measuring AI return on investment on ai.gov.au — and it matches how we run our own AI Opportunity Audit: baseline first, ROI estimate per opportunity, then build only what clears the bar.
We hold ourselves to the same standard in public: this site’s own performance is measured and published, win or lose, on the experiment page.
If you want the arithmetic done against your real call log, that’s a five-minute part of an audit call — and if your numbers say “not worth it”, we’ll tell you that too.