Here's an odd thing to admit as someone who sells patient engagement software: I think you should be more skeptical of AI vendors. Including us.

Over the past six months, I've talked with eye care administrators at every stage of the AI journey — live and thrilled, live and disappointed, mid-pilot, and deliberately waiting. The pattern I keep hearing isn't that AI doesn't work. It's that every demo sounds identical, every pitch promises the same things, and practices are left choosing on vibes and pricing. Some are now paying for tools they've quietly stopped using — and discovering that getting out is harder than getting in.

Ophthalmology makes this harder than most specialties. Your scheduling isn't a doctor's calendar — it's techs, lanes, dilation, subspecialists, and a patient who says "I need an eye exam" and means one of eleven different things. Your patient base skews older, which changes what "patients love it" actually means. And your no-show math is brutal enough that a wrong pick costs real money.

So here are the ten questions I'd ask any AI vendor — including us — before signing anything. They're designed to be asked in the demo, live, where the answers are hardest to fake.

1. "Book me a comprehensive exam — with the tech, the dilation, and the lane."

Not a provider slot — the whole visit. Ophthalmology scheduling is multi-resource, but most AI tools only understand doctor calendars. If the demo can't coordinate equipment, testing rooms, and technicians, your staff will be cleaning up behind it daily.

2. "A patient says 'I need an eye exam.' What happens next?"

Routine vision, medical, or post-op? Patients don't know the difference. Many can't tell LASIK from cataracts no matter how carefully the phone tree is worded. Ask how the system tells the difference between visit reasons, and what happens when it guesses wrong. A routine eye exam booking over a surgical slot is not a rounding error.

3. "Show me the write-back to my EHR — live, on my system."

Nextech, ModMed, athenahealth, NextGen, AdvancedMD — name yours and ask to see a booking land in it, in real time, and correctly the first time. "We integrate" can mean anything from deep two-way sync to your staff re-keying everything the AI touched.

4. "What happens when my patient says 'representative'?"

They will, especially when the answer is one they don't like. Ask what triggers escalation and whether the context transfers or the patient starts over from scratch with a human. An AI that hands off gracefully saves a call; one that doesn't generate a complaint and a call.

5. "What's your adoption data for patients over 70?"

A cataract-heavy panel is not a fintech user base. Ask about performance with hearing aids, regional accents, landlines, and background noise — and whether the system keeps its patience, because early adopters report AI agents that get audibly snippy under repetition. Your patients notice tone.

6. "Which appointment types can I wall off from the AI entirely?"

Retina injections, post-ops, surgical consults — some visits require human judgment for clinical or compliance reasons, full stop. If the vendor can't give you appointment-type-level exclusions, you don't have control; you have unwanted exposure.

7. "Which three numbers will you commit to moving — and where's the dashboard?"

Answer rate, hold time, calls resolved end-to-end, no-show rate, schedule utilization. Practices that measure honestly report the win isn't headcount reduction — its phones answered in seconds at roughly the same total cost, with staff freed for complex work. If a vendor promises you revenue lift instead of operational metrics, they're selling you the brochure.

8. "What are my exit terms if this isn't working at 90 days?"

Ask specifically: notice period, cancellation approval process, annual lock-in. Practices across ophthalmology are currently paying for AI tools they've stopped using because getting out requires a review board and 60 days' notice — or waiting for a contract anniversary. In a market moving this fast, a hard exit is a bigger risk than a wrong pick.

9. "What share of my actual call mix can you complete — not answer, complete?"

Ask for a call-type breakdown from a comparable eye care practice. If your patients already self-schedule and reschedule online, the calls that remain are the complex ones — combined billing-scheduling-clinical questions an agent can't finish. The better your existing digital tools, the less an AI phone agent has left to do; make the vendor prove the remainder is worth the price.

10. "If my PM system ships this feature next year, why do I keep you?"

Putting a language model on a phone line is quickly becoming the easy part; the market is filling with near-identical offerings, and PM vendors are building their own. The durable value is everything underneath: the scheduling rules, appointment logic, and integrations the voice sits on. If the answer is about their AI rather than their workflow depth, you're buying the commodity, not the moat.

A closing thought on where this market is going: Putting a language model on a phone line is quickly becoming the easy part. In a year, I expect a voice feature in nearly every product you're offered — at which point "do you have AI?" stops being a useful question at all. The ten above will still work because they were never really about AI. They're about whether a vendor understands how an eye care practice actually runs.

If you want to pressure test these against what we're building — including where we'd give you an imperfect answer — I'm easy to find.