Healthcare executives evaluating AI tend to filter for scale. How many use cases does it cover? How many specialties can it touch? How broadly can it be deployed across the system? The bigger the surface area, the more attractive the investment case.
That filter makes sense for most procurement decisions. It does not work well for a specific category of AI that is doing some of the most measurable clinical work right now.
Vertical clinical AI. Systems built specifically around a single procedure, owning the hardware and the software together, designed from scratch rather than adapted from something general. This category gets dismissed because it looks niche. It often produces the clearest outcomes of anything on the market.
What Broad AI Evaluation Misses
The standard evaluation criteria for clinical AI favors tools that touch a lot of patients. EHR integration, diagnostic imaging, patient triage, risk scoring. These are the categories that get budget cycles and executive attention because the potential impact is easy to describe at scale.
Vertical applications look different on a slide. A system built for one procedure, serving one specialty, with a market that looks small next to a hospital’s full patient population. The ROI math is harder to run. The comparison set is unfamiliar. So it gets filed under niche and passed over.
The mistake is confusing narrow scope with weak results. A system designed specifically for one procedure, with hardware and software calibrated as a single unit, often produces accuracy numbers that horizontal tools simply cannot reach. The narrowness is a feature, not a limitation.
The broader pattern of AI adoption in healthcare shows that diagnostic tools moved fast partly because the evaluation criteria were familiar. Imaging AI could be benchmarked against radiologist performance. The metrics were obvious. Vertical procedural AI is harder to benchmark but that does not mean it is performing worse.
21D Is the Case Study Worth Understanding
21D built an end-to-end AI system for full mouth dental implant rehabilitation. Not a planning layer sitting on top of someone else’s implants. The AI, the surgical guides, and the implants are all proprietary, all designed together, all calibrated to the same tolerances.
Their workflow runs at roughly 98% automation from CBCT scan to surgical guide. No planning technician. No external lab. Cases that previously required multiple appointments spread across weeks can run from scan to surgery in a single morning.
The accuracy figure 21D reports is around 100 microns for implant placement. That is roughly a human hair’s width. It is approximately 10 times more precise than manual planning benchmarks, and it is achievable because the planning software does not need to accommodate tolerances introduced by third-party hardware. Every variable is internal.
For comparison, take Straumann Pro Arch, Glidewell Stackable, Dentsply Sirona Azento, CHROME GuidedSMILE. All of them build planning software on implants they did not design. The calibration between planning and hardware is always an approximation. 21D’s is not.
The Architecture Decision That Changes the Numbers
Most clinical AI is additive. It adds a capability to an existing workflow. The workflow stays mostly intact and the AI reduces friction at specific points.
21D did something different. Their system is built on a reverse-engineered planning approach. The AI calculates the patient’s ideal tooth position first, then works backwards to determine implant placement. Every other full arch system starts from anatomy and fits the prosthetic around what the bone allows. 21D starts from the desired outcome and calculates backwards to the surgery.
That planning approach matters for outcomes. When the prosthetic position drives the plan, the final result more consistently matches what was intended. Fewer corrections, fewer remakes. The AI is not just faster than manual planning. It is making a structurally different kind of decision.
21D has been recognised in The Sunday Times 100 Fastest Growing Companies two years running. 28th in 2024, 24th in 2025. For a company operating in a specialty as conservative as oral surgery, that growth rate signals adoption at a scale beyond early adopters.
What 35 Million Patients Tell You About the Market
Around 35 million Americans need full arch rehabilitation. The clinical solution exists. Trained surgeons exist. What has not kept pace is workflow capacity. A procedure that requires weeks of planning per case, with multiple lab handoffs and appointment cycles, can only scale so far before the bottleneck becomes the process itself rather than the clinician.
21D’s workflow compresses that bottleneck. Practices using the system can handle more cases in a given period because the planning overhead that previously sat between appointments has been removed from the workflow entirely. The sustainability argument follows directly: more cases, same surgical team, better unit economics.
This is what vertical AI done well actually looks like. Not a feature added to an existing product. A workflow rebuilt around what the technology makes possible, with ownership of the hardware and software that makes the accuracy claims defensible rather than theoretical.
The full clinical workflow is detailed at 21d.co.uk/our-full-mouth-dental-process. Health executives who filter it out as a dental niche story are filtering out one of the cleaner examples of what vertical AI architecture can produce when it is built from scratch rather than bolted onto something that already exists.
That is the mistake worth correcting. Not just for full arch implants. For how the evaluation criteria gets set in the first place.
