Answering services bill by the minute. That's the least of your problems.

Go find your last answering service invoice. Somewhere on it is a line that says minutes, and a number next to that line, and that number is what every comparison between a human answering service and an AI receptionist gets stuck on. The per minute rate is rigged in a mildly boring way, which we'll get to. The expensive part is happening somewhere else on the same call.

A minute is a strange thing to buy

An answering service sells you time in a bucket. A hundred minutes, five hundred, whatever the tier is called. You draw it down until it's gone, then you top up or eat the overage rate, which is always worse than the bundled rate. The headline price per minute gets quoted on the sales call. The shape of the deal underneath does not.

Four things to check on your own plan:

• Does every call round up to a whole minute, so an eight second wrong number bills the same as a real customer?
• Do unused minutes roll over, or evaporate on the first of the month?
• Is there a per call fee sitting underneath the per minute fee?
• Does the greeting count? Does hold time count? Does the part where they read your own address back to the caller count?

Then look at what that structure rewards. Nobody at that desk gets a bonus for taking twenty seconds off your average call, because the twenty seconds is your money, not theirs. Call it structure rather than malice. But across four hundred calls in a quarter, "nobody is paid to hurry" shows up on your statement the way a slab leak shows up on a water bill. Slowly, then all at once. Law firms bill in six minute increments for the same structural reason, and everybody has opinions about how those invoices read.

Now listen to the actual call

Made up scenario, but make it concrete. Tuesday, 7:40 in the morning. Somebody's garage door has jammed halfway with a Corolla underneath it, and she's calling the first two truck garage door company that came up on the map.

Version one, human answering service. The greeting is scripted and it has your company name in it. Then she asks what a broken spring costs. The person on the line doesn't know, because they're covering thirty businesses this morning and what sits in front of them is an intake form, not your price list. "I'll have someone get right back to you." Can anyone come out today? Same answer. Ninety seconds, call over, message drops into an inbox. Billed as two minutes after rounding, plus the per call charge if your plan carries one.

Version two, an AI receptionist that was set up with the actual business behind it. Same spring question. It gives her the range the owner loaded in: a double door spring job typically runs 260 to 340 depending on the hardware, with a tech confirming on site before anything gets touched. She asks about today. It offers 11:15, books it into the owner's Google Calendar, takes her name, address, callback number and the age of the door, then repeats the appointment back once. Two minutes and change. All of it on the company's own phone number, so she never finds out the voice was software, and wouldn't care if she did.

Nobody was rude in either version. Only one of the two was allowed to answer the question she actually asked.

That's what the per minute comparison hides. Minutes are the visible part of the bill. What you're paying for underneath is the range of things the voice is permitted to say, and a message taker's range is close to nothing: no prices, no availability, no service area, no confirmation that you even do the job she's describing.

The callback is where the money dies

So the message arrives. The owner is up a ladder until noon. He calls back at 1:10, gets her voicemail, leaves one of his own. She booked the outfit three listings down at 9:30, because her car was trapped and she had a shift at eleven.

A 411 Locals study of 85 businesses found that 62 percent of calls to small businesses go unanswered, and a message service does fix that part. Somebody picks up. The Numa Small Business Phone Report found that 85 percent of callers who reach voicemail never call back, and what that number really measures is how people respond to being told to wait. A promised callback asks the customer for the same favor voicemail does. Nicer voice, same favor.

Booking on the first call deletes the whole loop. No message. No callback. No second chance to lose her.

Where a person still wins, and I'll say it plainly

A customer disputing a March invoice with real heat in her voice. Someone crying because the water is over the baseboards. An 80 year old who needs a human being to be patient with him for four straight minutes. Those calls want judgment, and judgment is the thing people still have and software mostly doesn't.

Which is why "AI or a human" is a fake choice. The AI takes everything, instantly, at 2am and during the Saturday morning pileup, and hands the call to a person when the call earns it. An answering service turns every call into a message, including the ones that were three questions and a booking. A badly configured AI tries to handle a call it should have passed along. One of those you fix with a transfer rule and an afternoon.

Now do your own arithmetic

Pull last month's call volume. Say it's 220 calls averaging two and a half minutes, rounding to three. That's 660 minutes. Multiply by your rate, add the per call fees, add whatever the overage cost you. Write the total on a sticky note.

Rocketship runs a free tier and paid plans from 12 dollars a month, with no meter anywhere in it, so a chatty caller at 9pm on a Sunday costs you nothing extra and neither does a wrong number. Put that next to the sticky note.

Then add the figure that never appears on any invoice: the spring and opener job, call it 1,900 dollars, that went to whoever answered her question at 7:41 instead. Two or three of those in a year and the per minute rate stops being the interesting number.