Before buying AI for your dental practice, ask four critical questions: What specific problem are you solving, how will you handle AI errors, should you build or buy, and who owns implementation success? This disciplined approach helps avoid expensive mistakes and ensures AI investments deliver real impact.
- Start with problems, not products: Define the problem first, then evaluate AI against all solutions including non-AI alternatives to avoid FOMO-driven purchases.
- AI makes mistakes with confidence: Unlike traditional software, AI is probabilistic not deterministic. Ensure vendors have monitoring, guardrails, and can explain their error rates and how they catch mistakes.
- Build dashboards, buy chatbots: Internal read-only tools like dashboards are now buildable in-house in weeks, but patient-facing clinical tools need vendor expertise and governance.
- Assign dedicated ownership: Name the person responsible for implementation before signing contracts, including training plans and accountability on both your side and the vendor's.
- Evaluate error visibility and cost: Assess how easily mistakes surface and their impact -- surface-level errors like transcription mistakes are safer than hidden errors in billing and claims.
Ask any dental leadership team what is at the top of their technology shopping list and you likely get the same answer: artificial intelligence. Vendors are lining up to present their proposals, peers are announcing new contracts at conferences, and boards are asking about AI as though AI were a strategy, not a tool.
Plenty of the AI promise is real -- we’re living in one of the most exciting times in dentistry. But the gap between a good AI decision and an expensive one is usually the discipline of asking the right questions before choosing your tools.
I’ve worked as an emergency room doctor, a management consultant to Fortune 500 companies at the Boston Consulting Group, built a digital medical device company, and I now run 01Health, building AI for dentistry. Having sat on all sides of the table, these are the four questions I would ask.
What problem are you actually solving?
Dr. Sonia Szamocki.
First, start with the problem you are trying to solve, then work back to the tools that will achieve that goal -- in other words, work right to left. With the hype and excitement surrounding AI, not to mention the external pressures, it's all too common to work left to right, where you see an impressive product and go hunting for a use case to justify it.
You hear it in your meetings: “This looks amazing, I can imagine this being useful here,” “Well, everyone else is buying it,” “AI is the future, so we can’t miss out.” The question is, if this vendor never called you, would this problem be on your list at all? If the answer is no, FOMO, or the fear of missing out, is driving your procurement.
Then comes the step almost everybody skips: Appraise every possible solution, not just the AI ones. Missed calls after hours might be fixed by an online booking link for a fraction of the cost. Make AI compete on merit against the “boring” options.
AI will be wrong sometimes. Then what?
AI is not deterministic, it's probabilistic. Ask the same question twice and you can get two different answers. It’s what separates this technology from every piece of software you have bought before.
Traditional software is like a coffee machine. Press the button, get the drink. Same input, same output, forever, and if it jams, there is a repair to make. In comparison, AI is a barista. It interprets based on your inputs. Better than the machine on a good day and worse on a bad one, it improves as it learns the preferences of its regular customers, but occasionally, it serves the wrong thing with total confidence.
You do not repair a barista. You taste the coffee. Quality control -- usually called evaluation in AI -- is a new discipline for everyone building and buying.
A serious vendor should have a harness built around the model: monitoring, guardrails, a defined path for when the output looks wrong, and an honest answer when you ask what their error rate is and how they measure it. Ask for that. A vendor who cannot tell you how they catch their own mistakes has not built a way of catching them.
As a buyer, you cannot completely outsource evaluation to the vendor. Ask yourself how easily you could tell if something went wrong. Two things decide that: one, whether a mistake is visible; and, two, what is the cost of that mistake to your patients, your business, and your reputation?
At the easy end, take voice transcription and summarization. If the AI misinterprets something, a human reads the note and fixes it in seconds. The error is on the surface, the cost is trivial, and the loop corrects itself. You can buy that fairly comfortably.
At the hard end, take anything touching claims, coding, or billing. A wrong number sits in a field nobody reads, looks entirely plausible, and compounds quietly for a quarter until the figures will not reconcile and you cannot work out where it started. Nobody is tasting that coffee.
So before you sign, make sure you know what "wrong" looks like with the tool you're considering, who would spot it and how quickly, and what it costs. The harder it is for you to check, the more the vendor has to prove they are checking and the stronger the case for keeping a human in the loop along the way.
Should you buy this solution at all or build it?
My rule for dentists and dental service organizations is to build the dashboard and buy the chatbot.
Anything internal and read only is now genuinely buildable in house, in weeks, by one curious operator in your business. Dashboards can pull practice management, accounting, and marketing into one view. Three years ago, these required an engineering team, but AI has changed that.
Anything patient-facing, clinical, or regulated is a different animal. It needs an evaluation pipeline, clinical governance, and a team that owns it permanently. That is a company, not a feature, and you should buy it from someone who has built exactly that.
The awkward part is that the tools you could most easily build yourself are the ones being pitched hardest because they are the easiest things to build and sell. A license fee for something your ops lead could have shipped in two weeks is a very common way to spend money.
Even if you buy, build one small thing yourself first. It teaches you what is actually hard, and it changes every vendor conversation afterward.
Who owns making it work?
This isn’t about who bought the tool. It’s about who owns making it work once the contract is signed.
Software doesn’t usually fail to drive impact because it breaks. On the contrary, it fails because people quietly go back to the manual process they already trust, or they decide not to make full use of the capability for which the software was purchased.
That is true of every new tool you buy, and AI adds three complications that make dedicated ownership essential.
- Trust in the tool is hard to win, easy to lose. A clinician who has seen the tool be confidently wrong once will discount it, and that trust is hard to win back. That has to be actively managed: Be clear about where it is reliable, honest about where it is not, and find an easy way to flag it when the output looks wrong.
- There is no best practice to copy. With a new practice management system, you can ring three peers and ask how they set it up. With AI, the case studies don’t yet exist because the technology is new and often these tools adapt to your context, so somebody else's rollout does not necessarily transfer cleanly.
- The product will change underneath you. Conventional software moves on a release schedule you can plan around. AI tools shift capability day by day, month on month. Things that were impossible when you started using the product may be routine six months in, and things will change and break. Somebody on your team has to be paying attention.
Put those together and you are not really implementing a tool. You are closer to managing a new hire: onboarding, supervision, feedback, correction, and trust extended as it gets earned.
So name the person who is responsible for this before you sign, and give them the time to do this. Then get the vendor's half in writing: Who is accountable on their side, what the training plan is for the people who will actually use it, and what happens when clinicians push back, because they will. A vendor who cannot name a person has a sales team and no implementation function.
Where this leaves you
Write down the problem and the measure before your next vendor call, then compare the AI option against the "boring" alternatives, or build it yourself. Name the person who owns making it work in your organization, because that is as much your job as the vendor’s.
Everything on the market today will be superseded faster than any of us expect. What compounds is being good at buying things that have impact, not the shiniest things.
The organizations that win the next few years will not be the ones that bought the most AI, they will be the ones that had the discipline to buy the right tools, at the right time for them.
Dr. Sonia Szamocki is the founder and CEO of 01Health, an artificial intelligence-driven platform that aims to launch and scale new specialist revenue streams within dental offices and dental service organizations. Szamocki began her career as an emergency room doctor, then joined Boston Consulting Group and BCG Digital Ventures to build and scale healthcare and medtech businesses. Szamocki has proven 01Health’s model in two specialties, through clear aligner platform 32Co and dental sleep medicine platform Aerox Health.
The comments and observations expressed herein do not necessarily reflect the opinions of DrBicuspid.com, nor should they be construed as an endorsement or admonishment of any particular idea, vendor, or organization. Some content may be AI-generated.
