Here's the trap. Your receptionist left, the phone rings forty times a day, and two people have given you numbers that are three orders of magnitude apart.
A vendor quoted $399 a month. Someone on a forum said you could do the same thing on Vapi for five cents a minute. Neither is lying, and neither number is useful until somebody tells you how many minutes you actually use.
So here's that arithmetic, in the unit a practice manager thinks in.
What a minute actually costs
A voice agent is four things bolted together: something that hears, something that decides, something that speaks, and a phone line. Priced separately at published rates, that comes to $0.0465 a minute. The workings are on our cost page, with a date against every component.
Divide any monthly price by that figure and you get the volume at which the subscription stops being a markup and starts being a bargain.
| Tool | Monthly | Break-even | In 4-minute calls |
|---|---|---|---|
| Rosie | $49 | 1,054 min | ~264 calls |
| Goodcall | $79 | 1,699 min | ~425 calls |
| Slang.ai | $399 | 8,581 min | ~2,145 calls |
| Sameday AI | $449 | 9,656 min | ~2,414 calls |
| Structurely | $598.80 | 12,877 min | ~3,219 calls |
Now the number that matters. A single-site practice taking 300 calls a month at four minutes each uses 1,200 minutes. At component cost that's about $56.
Against $399, you are paying roughly seven times the cost of the minutes. Against Structurely's $598.80, closer to eleven. Those multiples are the honest answer to the question, and they are not the end of it.
So what is the other 86% buying?
Four things, and they are real. The mistake is assuming the markup is the product.
The integration. Somebody built a write path into your calendar or practice system and maintains it against an API that changes without telling them. This is the expensive one and the one you cannot see.
The prompt work. An agent that knows what a hygiene appointment is, what to do when someone says they're in pain, and which requests need a human. That's weeks of iteration against real calls, done once, spread across every customer.
A phone number and a person to call. When the agent books someone into a slot that doesn't exist, there is a support queue. On the component route, the support queue is you.
A bet on your volume. Flat pricing means a busy December costs the same as a quiet August. The vendor absorbs that, which is worth something and worth less than it sounds if your months don't vary much.
When does the component route actually win?
Three cases, and they are narrower than the forum makes them sound.
High volume. Above the break-even the arithmetic flips and keeps flipping. A clinic group running 20,000 minutes a month is paying a vertical vendor many times component cost, and at that scale an engineer is cheap by comparison.
Many identical sites. Build once, deploy forty times. That is a different business from buying one phone line, and it is what the white-label market exists for.
A requirement nobody sells. If what you need isn't what any vertical vendor built, configuration won't get you there and the platform is the only route.
Notice what isn't on that list: being technical enough to try. The build is a week of somebody competent and then a standing maintenance job forever, and at a loaded internal rate one week costs more than a year of Slang.ai.
The cheaper question nobody asks first
Before choosing between $399 and five cents, check whether you need either.
Rosie at $49 breaks even at 1,054 minutes and Goodcall at $79 at 1,699. A practice running 1,200 minutes is close to break-even on both, which means at that volume the cheap vertical tools are priced near the cost of their own components and the markup argument mostly evaporates.
The $399 and $598.80 tiers are not the same product sold dearer. They are sold to businesses with more integration surface, and the way to find out which you are is to ask what the setup actually involves rather than comparing the monthly figures.
So is $399 a rip-off?
At 300 calls a month, you're paying about seven times what the minutes cost. Whether that's a rip-off depends entirely on one question, and it isn't a question about the software.
Would anyone at your business otherwise build this and keep it running? If the honest answer is no, then the component price is a number you were never going to be able to pay, and the premium is buying the only version of the product that exists for you. If the answer is yes and you have the volume, the arithmetic says build.
What you should not do is pay $399 because it sounded reasonable, or reach for the components because five cents sounded cheap. Both numbers are real. Only one of them is yours, and the break-even table tells you which.