Walk any exhibit hall this year and every booth says the same two words: AI-powered. AI-powered scheduling, phones, imaging, billing, marketing. Some of it is real and useful. A lot of it is a new label on software you already own, with a new price.
The trouble is that from the booth, real and hype look the same: the same buzzwords, the same polished demo, the same confident rep. Here are the five claims we'd be most careful about, and the one question that exposes each. They come from healthcare, especially dental and medical offices, but they apply to any small business buying software. The one-page version is at /hype-filter.
A mental model first
Think of AI as a brilliant, tireless intern. It's great with language, it works at 3 a.m., and it never gets bored. But it has no license, and it needs supervision. AI can sound just as confident when it's wrong as when it's right. Every red flag below is a version of a vendor forgetting that.
1. "Fully autonomous. No human oversight needed."
Sometimes you hear it as "set it and forget it." It sounds like a feature. But anything that reaches a patient, a payer, or a customer should pass a person first. A vendor bragging that nobody checks the work is bragging about the risky part.
Ask: "Where does a person approve it before it goes out?" If the answer is "nowhere, that's the beauty of it," that's your answer.
2. "We're HIPAA certified."
There's no such thing. The government doesn't certify software for HIPAA. A vendor can be HIPAA compliant, and some get outside security audits that are worth asking about. But "certified" is the tell. What matters is whether they'll sign a Business Associate Agreement (BAA) and take on legal responsibility for your patients' data. We wrote up what a BAA is and what to ask.
Ask: "Where's the signed BAA?" If they hesitate, walk.
3. "Don't worry, your data's safe with us."
That's a reassurance, not an answer. A vendor that takes security seriously can tell you where your data is stored, who at the company can see it, how long they keep it, and whether it's used to train their AI. Watch the small things, too. A demo full of real-looking patient records, or a rep who says "just email us your patient list to get started," shows habits that don't stay in the sales process.
Ask: "Where does our data live, and is it used to train your AI?" A good vendor answers in writing.
4. "Our AI diagnoses and recommends treatment."
To be fair, some imaging tools are FDA-cleared to help a clinician spot things on an X-ray, and those can be legitimate and useful. That's helping the dentist or doctor. The red flag is software that claims to make the diagnosis on its own. That's the license the intern doesn't have.
Ask: "Is it FDA-cleared for exactly this use, and does the clinician make the call?" If the clinician isn't the one deciding, keep walking.
5. "Now with AI!" (New price.)
The same software, a new label, a higher bill. It's the most common one, and the easiest to check.
Ask: "What does it do today that last year's version couldn't?" If the answer is last year's feature list with "AI" in front of it, you have your answer.
What a good answer sounds like
A vendor worth your time answers all five without flinching. A person approves anything that goes out. Here's the signed BAA. Your data lives here, these people can see it, and it isn't used for training. The clinician decides. And this year it does this specific thing it couldn't do before, which saves this much time.
Then measure it. Pick one workflow, write down one number before you start (hours on appeals, overdue patients booked, calls to voicemail), and check it again in 30 days. If the number didn't move, the tool didn't earn its place, however good the demo was.
We hold ourselves to the same five questions. If you'd like an honest second opinion on a pitch, or on whether AI helps with a workflow at all, start at /discuss or write to hello@fullspecstudio.com.