Nobody answered the phone. The AI texted them back.
An office manager on maternity leave, a newer hire at the front, and a voicemail box nobody could empty. The fix was not an agent that picks up. It was one that texts back the call nobody took and books the patient, once the data underneath it was clean enough to trust.
Disclosure: the practice in this story is one our own team is involved in, and the practice pays for Peerlogic as its customer. DentalTechHub pays Peerlogic nothing and earns nothing from Peerlogic. Its classification in our catalog was set by DTH’s editors, by the problem it solves, the same as every other vendor in it.
The problem was staffing
Six months into new ownership, the office manager went out on maternity leave. She was the one who knew how the phones actually worked, and the other front desk team member was newer.
Calls started going unanswered during the day, not only after hours. Then the voicemail box filled up.
A full voicemail box does something worse than lose a message. It tells the patient the practice is closed. A staffing shortage had become a patient pipeline problem.
Four requirements, written before the first demo
- After-hours coverage. The calls that arrive when nobody is in the building.
- Works with Denticon, and with the phone platform already in place. Replacing either was out.
- Holds a conversation. Not an agent that answers, sends a booking link and hangs up.
- Texting, if it came along with it.
Then they asked Mola, our AI advisor, working over the 547 vendors DTH’s editors classify by the problem each one solves. All four requirements are things the catalog records, which is why the question returns a shortlist rather than a category page of companies describing themselves the same way. They demoed several agents and chose Peerlogic’s Aimee.
The fourth requirement turned out to be the whole mechanism
Texting was ranked last, as a nice-to-have. It is the channel the entire solution runs on.
Aimee does not pick up the phone. When a call goes unanswered, she texts that patient back and opens a conversation. From inside the thread she answers the question, picks a time, books it on the schedule and sends the confirmation. The patient never calls back, and never leaves a message nobody was going to retrieve.
That turns the second requirement from a checkbox into the two ends of the mechanism. The phone platform has to tell her a call was missed. Denticon has to let her hold a real slot.
One trade-off, because a vendor page will not name it. Aimee texts from her own number, so the patient hears from a number they have never seen about a call they made a minute ago, and the practice now texts from two numbers instead of one. Not a reason to skip it. A reason to tell your team first.
A correction to something we published
We wrote about this practice once before, in The phone system that came with the practice. That piece ended by saying the AI answer arrived from the phone vendor, without a second contract.
That is not what happened. The practice kept its phone platform for phones and texting, and pays Peerlogic $399 a month for the layer that sits behind it and catches what it drops. Two vendors, two contracts, on purpose.
The AI feature your phone vendor ships arrives because they built it, not because it met your requirements. The agent is its own decision, even when the vendor you already pay is offering one.
Then the real work started
An AI agent is only as good as the instructions you give it and the data it reads. Two things in this office were not clean.
The provider list. The roster carried forward from the previous ownership, so dentists who no longer worked there were still active in the system. A person at the front desk knows to ignore those names. An agent does not, and will offer a patient an appointment with a dentist who left before the sale closed.
The schedule. Make the days resemble each other, because if Monday’s template looks nothing like Tuesday’s there is no pattern for the agent to follow. Then decide how many slots are actually bookable. Open everything and the agent books a new patient exam into the slot beside a root canal.
Neither is an AI problem. Both determine whether the AI works.
What 60 days looked like
| Conversations Aimee started after an unanswered call | 354 |
| Patients who replied | 154 |
| Handed to a person | 39 |
| Appointments booked | 13 |
| Appointments rescheduled | 6 |
| Appointments cancelled | 3 |
| Booked revenue | $4,251 |
354 missed calls got a text back, 44% of those patients replied, and about one in seven conversations ended in a change to the schedule. Against $399 a month, that is roughly $798 across the window and 5.3 times that back: $61 per booked appointment, $2.25 per missed call.
Three honest notes. It is booked revenue, not collected. It undercounts, counting only what Aimee booked herself, not the 6 she saved by rescheduling or whatever came of the 39 conversations she handed to a person. And $399 covers more than bookings, because recording and tagging every answered call comes out of the same line item.
Of those 39 hand-offs, 24 happened because the patient asked for a person and 15 because Aimee decided the conversation needed one. An agent that knows when to stop is worth more than one that insists on finishing.
What the owner would tell you
The vendor choice felt like the whole project and turned out to be the smaller half. Write your requirements down before you take a demo, and expect the one you rank last to matter most. Then clean the two things the agent reads from before you turn it on.
Point an agent at a stale roster and an inconsistent schedule and it books the wrong patient with the wrong provider at the wrong time, and you will conclude the technology does not work.
It was the data that did not work.
Ask Mola what fits your practice.
Give it your practice management system, your phone platform and the hours you need covered. When you find a vendor worth seeing, request the demo through the profile and DTH coordinates it with your office.
Khamir