TECH

Your AI cheat sheet: The basics DSO leaders need to know right now

Everyone’s selling AI now, but you need more than a sales pitch to separate hype from the true capabilities of the technology. Here's your no-fluff field guide to what AI can actually do, where it still falls flat, and where it’s headed. 

What's happening: AI has graduated from "cool demo" to "line item on the P&L" faster than most business leaders (at least outside top AI labs) expected. Seventy-eight percent of organizations reported using AI in some capacity in 2024, according to Stanford's AI Index. 

• In dentistry specifically, the technology has moved well past the chatbot phase. FDA-cleared imaging tools, voice-enabled perio charting, automated insurance verification, and ambient clinical documentation are all live and productized today.

Yes, but: Just because companies are spending money on AI doesn’t necessarily mean they understand it. The terminology is moving just as fast as the tech, and that's creating a knowledge gap. When a vendor drops "agentic AI” or “RAG” into a pitch, you need to know what's real and what's resume padding. So here's a quick decoder ring:

• Foundation models / LLMs: The engines behind tools like ChatGPT and Claude. They're very good at drafting, summarizing, classifying, and coding. They are not good at being right 100% of the time, a feature the industry politely calls "hallucination" and that you might call “BSing.”

• Reasoning models: Newer systems that spend more compute time "thinking" before answering. Better for complex analysis, worse for speed and cost. Great for hard clinical questions, overkill for appointment reminders.

• RAG (retrieval-augmented generation): Instead of generating a prediction as to what the answer will be, the model pulls from a curated set of trusted documents first, then generates a response. This is how the best knowledge assistants and patient communication tools stay on-script. If your vendor isn't using this pattern, ask why.

• Agentic AI: Models that don't just answer questions but take actions, like booking appointments, pulling eligibility, and drafting claims. The promise is huge, but independent testing finds that even the best systems still succeed at real-world computer tasks less often than humans. Agents are improving, but proceed with guardrails for now.

• Human-in-the-loop (HITL): A system-design principle, meaning a human must verify AI output. This matters because it's becoming the regulatory and clinical expectation for every AI tool that touches patient care, from imaging overlays to auto-generated chart notes. Being able to enforce clinician review of sensitive work is an important thing to look for in AI tools deployed into any healthcare setting.

What it means for you: When it comes to dental, AI capabilities can be grouped into roughly two baskets:

• Clinical imaging AI: There are now dozens of FDA-cleared products for caries detection, bone loss measurement, periapical findings, CBCT segmentation, and orthodontic remote monitoring. The ADA has also published its first ANSI-approved standard on validation datasets for dental imaging AI, which gives you a credible lens for evaluating vendors.

• Operational and RCM automation: Insurance verification, claims attachment generation, patient FAQ bots, and voice perio charting are all scaling quickly. They don't make clinical claims, which means lower regulatory risk and faster deployment.

Why it matters: The gap is widening between DSOs that are piloting AI and learning what works, what doesn’t, and what guardrails they need to use the technology safely, and those still debating whether to "wait and see." AI is no longer something that’s coming on the horizon. It’s here, and one more thing (as if you needed another) that you’ll have to stay on top of to keep up with the competition.