Should a robot answer your phone?

Voice agents got good enough to sell hard this year. Here's the honest version — where they work, where they'll cost you customers, and what to try first.

Why this is suddenly everywhere

AI voice agents crossed a quality line over the last year. They interrupt naturally, they handle someone changing their mind mid-sentence, and on a short call most people don't clock it. The vendor market has exploded to match, and the pitch is always the same: never miss a call again, for a fraction of a receptionist.

The underlying problem is real. For a lot of local businesses a missed call is simply a lost customer, and they miss plenty during the busy hours when the phone matters most.

Be sceptical of the adoption and ROI statistics in this category — nearly all of them are published by companies selling voice agents. The technology is genuinely better; the numbers are marketing.

Where it genuinely works

After hoursThe comparison isn't a human, it's voicemail. Almost anything beats voicemail, and this is where nearly all the value is.

The overflow callSecond and third simultaneous calls that currently ring out. Nobody was answering those anyway.

The same four questionsHours, location, parking, do you take my insurance. High volume, zero judgement, no downside to automating.

Booking a standard slotRoutine appointments into an open calendar. Works well when the rules are simple.

Where it will cost you

The failure mode isn't the agent sounding robotic. It's the agent being confidently unhelpful to somebody who needed a person — and that customer doesn't complain, they just don't call back.

Anyone upset. A complaint routed to a machine escalates. Every time.
Anything urgent. Burst pipe, tooth pain, legal deadline. If the call is an emergency, the answer is a human.
Your best customers. The people who spend the most usually expect to reach someone. Recognise their numbers and route them straight through.
Anything with real money attached. Quoting a complex job is a sales conversation, not a lookup.

The rule worth holding.

Never trap a caller. A human must always be one clear sentence away — "let me get someone" — and it must actually work. Systems that loop people are worse than the voicemail they replaced.

Try the cheap version first

Before you buy anything, run it after hours only, for two weeks. You keep every daytime call exactly as it is, and the only thing at risk is calls currently going to voicemail. That's a real test with almost no downside.

Then judge it on one number: of the calls it took, how many became a booking, a callback, or a useful message? Not "how many calls did it answer" — a vendor dashboard will happily show you a big answered-call number that contains no business.

Listen to ten recordings yourself. Not a summary. Ten actual calls. You'll know within twenty minutes whether it's ready for daytime.

What to ask a vendor

Can I hear recordings from a business like mine — not a demo?
Exactly how does a caller reach a human, and what happens if nobody picks up?
Does it write into my calendar and CRM, or do I retype it?
What's the price at my real call volume, including overage?
Can I export my call data if I leave?

If a demo is polished but they can't produce a real recording, that tells you where the product is.

Do this now

Check your phone system for how many calls went unanswered last month, and when.
If it's mostly after hours, that's your pilot — and it's low risk.
If it's during business hours, the honest answer might be a part-time person, not software.

Want someone to just do this with you?

The AI Once-Over maps where AI actually saves time or makes money in your business, then builds the first one with you.

See the AI Once-Over →