Every voice AI demo sounds impressive. That's the problem. Voice agents are still new to dermatology, and few practices have run one long enough to know what to ask. So buyers judge what they can hear, like how natural the voice sounds and how quickly it responds. Those are the easiest things to get right. The differences that matter show up months later: in where the agent gets its information, what it's allowed to touch, and what it does when your schedule doesn't cooperate. These eight questions expose those differences while you can still walk away.
Some agents read the rules already configured in your system. Others rebuild a copy in their own, which means two systems to maintain. The copy you forget to update is the one that books a patient wrong.
Ask whether it uses a documented API or reads the screen like a person. Screen-reading breaks every time your PM vendor ships a redesign, and you'll hear about it from a patient, not an alert.
Forwarding the call uses one, the agent's leg a second, and escalating to a human a third. With a fixed channel count, you can run out of lines exactly when call volume peaks.
At most practices, some providers accept plans others don't. An agent that can't handle insurance won't simply skip a field. It will book new patients with providers who can't see them.
If your aesthetics or surgical side requires a deposit to hold a slot, an agent that only answers the phone has nothing to charge against.
Get the model in writing: per minute, per booking, or flat. Then ask who pays your EHR's API fees, and what happens if the vendor is absorbing them today to win the deal.
Some vendors quote six months to a year. Ask what the agent can do on day one, what it can't, and what you're paying for in between.
Recordings and transcripts show whether it's working and catch the bad call before the patient does. Ask what's captured and who can see it, and take consent and retention questions to counsel.
No vendor answers all eight well today, including the good ones and including my company. A perfect score isn't the point. You're listening for precision. A vendor who answers specifically has thought about a practice like yours. One who steers you back to the demo hasn't, and you'll be the one who finds out what they missed.