Guest experience: your dental team has better things to do

There’s a hospitality business inside every dental practice. A great dentist can still run an office people dread calling.

Steve Wynn talks about this: recognize employees for exceptional service and make each guest feel personally looked after. Give people a reason to come back.

If you run a DSO or dental group, I recommend taking a few minutes to watch this. You might find it inspiring.

Play here

Why am I bringing this all up? Most of you aren’t going to hire somebody whose only job is hospitality.

Your best front desk person, the one with the gift of gab, is already doing that work. They’re fantastic with patients. They’re also six tabs deep in an insurance portal while somebody waits for their attention.

Billing and admin eat a big chunk of their day. I get it. The work has to get done. But they can be good at both jobs and still have to choose which one gets done first.

The patient may have been putting this appointment off for six months. They’re worried about the bill and unsure what their insurance covers. Your front desk could remember their name, explain the bill, and answer questions without making them feel like an idiot for asking—if they weren’t six tabs deep in a portal.

But AI is changing how much of the admin people need to do themselves. A lot of that work should move to software.

Let’s make the humans better at making other humans feel better. Give them time to connect. Leave the stuff computers can do better to the computers.

We’ve trained a model for this exact job. It’s called Altadena.

Altadena is a Mulholland model designed to read dental insurance information and turn it into a benefits breakdown: what the patient’s plan covers, what limits apply, and what still needs to be checked before treatment.

Altadena outperformed GPT-6 Luna on our September development test. We ran both models on the same ten cases. Across the breakdown sheets, including fields left blank in the answer key, Altadena matched the key 93% of the time versus 70.9%. On fields with a reference answer, it also returned more correct answers. The cases were held out of Altadena’s training, but the insurers were represented in it.

In separate internal reviews of dental insurance verification services, our team estimates DAPA’s accuracy at roughly 60% and CEC’s at roughly 80%. These are service-review estimates, not scores from the shared Altadena/GPT-6 Luna test above. We haven’t published the sample size or scoring protocol for those reviews.

Already using a verification service? Put us side by side on the same cases. Compare the answers. Count the corrections. Ask your front desk which one actually took work off their plate.

We’re developing it as part of a larger service. Our production software already gathers payer information and applies rules for organizing it into a breakdown. The model is one piece. Getting the right information, checking it, and handling the exceptions are part of the job too.

Humana doesn’t list a procedure? We don’t copy the benefit from the one next door just to fill the box. We leave it unresolved unless another source answers it.

Missing answers and conflicting information get flagged for review. When we find a mistake, we can build a check for it. Wrong plan year? A benefit that only applies to kids? That check runs on the next breakdown, and the next hundred. Your front desk has enough things to remember.

But if you still have to hunt down every answer and fix the whole sheet yourself, congratulations. You’ve bought more admin.

The real test is what we took off your plate. How long did the finished breakdown take, including review? How much did your team have to fix? That’s what we need to measure.

The point is to give your people time to look up. Remember the patient’s name. Explain the bill. Help someone who’s nervous. Make the office nice as hell.

Your best people should be the reason patients come back. Give them the time.

Published October 6, 2026. Encore at Wynn Macau photograph by WiNG, CC BY 3.0. Resized and converted to WebP.