Dental AI Procurement: 12 Questions to Ask Any Vendor

Buying dental AI well is mostly a matter of asking the right questions and listening for specifics: what exact job the tool does, what its documented limits are, what happens to patient data, and whether the vendor is honest about all three. A tool that answers crisply is worth a trial; one that dodges is selling you an acronym. Here are the twelve questions to put to any dental AI vendor before you trust a tool with your patients.

Why does procurement matter more than the demo?

Because a demo is designed to impress and your patients are not. Vendors show curated examples that flatter the tool; your job is to find out how it behaves on ordinary cases, at the edges, and with real patient data over time. Good procurement replaces the seductive demo with hard questions and a trial on your own cases, which is the only test that predicts daily reality. The twelve questions below are structured to surface what a demo hides — the limits, the data handling, and the honesty — so you buy a capability, not a showreel. Ask them in writing, and weigh the willingness to answer as much as the answers.

What should you ask about the job it does?

Start with capability, because a vague answer here predicts a vague product.

1. What specific task does this perform? A clear, narrow answer is a good sign.
2. How do I verify it on my own cases? Insist on a trial with your patients, not the demo set.
3. What are its documented limits? Honest vendors name what it cannot do.
4. What happens when it's wrong? There should be a human check by design.

If a vendor cannot name the specific job and its limits, the rest of the conversation is moot.

What should you ask about patient data?

This is where the real risk lives, and where vague answers are most dangerous.

5. What is stored, and for how long? “Nothing” should be backed by specifics.
6. Are inputs used to train models? For patient photos, the right answer is no.
7. Where is data processed? Region matters under many data-protection regimes.
8. Will you sign a Business Associate Agreement if my workflow needs one? And does one exist?

Insist on written answers; a vendor serious about data will be specific and boring about it, which is exactly what you want. The deeper privacy version is in the privacy guide.

What should you ask about honesty and fit?

The last four separate a partner from a pitch.

9. Does it overclaim? Watch for guarantees, “replaces the dentist,” or unearned compliance badges.
10. How is output framed to patients? A preview should be labelled a goal, not a promise.
11. What does support, training, and onboarding look like? Adoption fails without it.
12. What are the pricing, contract, and lock-in terms? And can you leave cleanly?

A vendor honest about its limits, its data, and its price is telling you how it will behave once the contract is signed.

How do you read the answers?

Weigh specificity and candour over polish. The tell is not whether a vendor sounds confident but whether they will be concrete: named retention windows, an explicit no-training commitment, documented limits, a clear exit. Green flags are specific and slightly dull; red flags are glossy and vague — guarantees, badges without detail, deflection on data, discomfort with a real trial. A vendor who says “we do not do that yet” about a capability is often more trustworthy than one who claims to do everything, because honesty about limits predicts honesty everywhere else. Reward the concrete answer, even when it is less flattering than the pitch.

What does an honest vendor sound like?

Plain, specific, and willing to say what it does not do. When we are asked these questions, the answers are deliberately unglamorous: the tool generates a believable preview labelled a visualization aid, not a guarantee; the photo is processed transiently and never stored by Smileproof (our AI provider may retain inputs for a limited period — up to 55 days — solely for abuse monitoring under its data-processing terms); inputs are never used to train models; and for workflows that need it, BAA and in-region processing are on our roadmap, not yet available. That last answer — naming what is not yet available — is the kind of thing a marketing-first vendor buries, and exactly the kind of candour good procurement rewards. Hold every vendor, including us, to the standard of a straight answer you could hand a patient. The honesty-versus-badge issue is in why most dental AI tools quietly ignore HIPAA.

Does this checklist travel across markets?

Yes — the twelve questions are jurisdiction-agnostic, even though the data-law answers are local. Whether you practise in the US, the Gulf, the Levant, or elsewhere, you still need to know the job, the limits, the data handling, and the honesty; only the specific compliance regime behind the data answers changes. Adapt the data-processing question to your local law, keep the rest as they are, and the checklist works anywhere. A clinic in the Gulf swaps in its own data-protection regime for the compliance question; a clinic elsewhere swaps in theirs; the other eleven questions are unchanged, because the job, the limits, the framing, and the honesty are universal concerns regardless of where the tool is deployed. The broader evaluation framing is in the state of AI in cosmetic dentistry.

Want straight answers to all twelve before you trial anything? book a demo — qualified clinics get a trial set up personally after a short demo.