Every dental-AI sales pitch ends the same way: install the software, catch more cavities, save money. It’s a tidy story — so a team of health-economics researchers went looking for the receipts. They searched six databases and found that the entire case for the cost-effectiveness of AI in dentistry rests on just four studies.
The 30-Second Version
- A 2026 systematic review asked whether AI actually saves money in dentistry — and out of 334 records, only four studies qualified, all from Germany, all about cavities.
- In cost models, AI-assisted caries detection came out cheaper (€298–€378 vs €322–€419) and preserved slightly more teeth (up to 64 vs 62 “tooth-years”).
- But the one randomized trial found no advantage: identical cost (€330 vs €330) and identical tooth retention (49 vs 49 years) — because sharper detection led to more drilling.
- The honest caveat: four German cavity studies can’t tell us whether AI pays off for oral cancer, gum disease, or any clinic outside that one reimbursement system.
It sounds revolutionary — smarter eyes, fewer misses, lower bills. But in a systematic review published in June 2026 in Frontiers in Medical Technology, Rakesh Kumar Sahoo and colleagues asked the question the marketing usually skips: does AI in dentistry actually pay for itself? Following PRISMA guidelines and a pre-registered PROSPERO protocol, they combed six databases for economic evaluations of AI oral-disease screening. Of 334 records, exactly four made it through.
The study, in one glance
This was a systematic review of health-economic evaluations — not a lab experiment — appraised for quality with the JBI economic-evaluation checklist and the CHEERS 2022 reporting standard. All four surviving studies looked at AI detecting dental or proximal caries on bitewing radiographs, and all were set in the German dental system. Three were long-term cost models (mostly Markov simulations); one was a randomized clustered cross-over trial using a commercial AI, dentalXrai Pro. Every study measured the same things: cost, and “tooth-years” retained, with AI versus conventional assessment.
The promise: catch more, drill less, spend less
On paper, the case is genuinely appealing. AI is good at spotting the faint shadows of early decay that a tired human eye skims past — the same pattern we saw when researchers tried to turn a smartphone photo into a cavity screener. In one cost model, an AI system read cavities more accurately than dentists (0.80 vs 0.71); in a proximal-caries model it caught far more early lesions (sensitivity 0.75 vs 0.36), at the price of a few more false alarms (specificity 0.83 vs 0.91). Three of the four studies favored AI on both cost and tooth retention, and in the strongest model more than 77% of simulations came out ahead for AI. Crucially, that advantage appeared only when early lesions were then managed conservatively — sealed or monitored, not immediately filled.
So is AI in dentistry cost-effective?
Here the tidy story meets its hardest test. The single randomized trial in the review — the only study that watched real dentists treat real patients rather than a simulation — is the reality check. AI-supported assessment was, as promised, more sensitive. But that extra sensitivity nudged clinicians toward more invasive treatment, and the expected savings evaporated. The AI and standard groups landed on nearly identical costs (€330 vs €330) and identical tooth retention (49 vs 49 years). Only 41% of simulations favored AI; 43% favored no AI at all.
⚠ More detection, more drilling
AI’s extra sensitivity only pays off if clinicians respond with restraint. If every newly flagged shadow becomes a filling, you haven’t bought better health — you’ve bought more treatment, at more cost, for the same teeth.
A more sensitive detector is like a smoke alarm with the volume turned up: it catches real fires earlier — but only saves you money if you don’t call the fire brigade every time it chirps at the toast.
And the evidence base is thin enough to see through. Four studies. All from one country. All about cavities on bitewing X-rays. Three of them were models built on assumptions, not observations, and there were too few to pool into a meta-analysis. The review’s authors rate the studies as low risk of bias and well-reported — but low-bias evidence that barely exists is still barely evidence.
Who decides what happens after the alarm?
The review’s sharpest insight is that AI creates no value by detecting alone. A model flags a spot; a human still decides whether to watch it, seal it, or drill it — and that decision is where the euros and the tooth-years are won or lost. The cost-effective results all hinged on non-restorative management of early lesions, which is a clinical philosophy, not a software setting. The authors point to American Dental Association guidance urging independent validation of any dental image-analysis system before clinics lean on it. In other words: the algorithm is accountable for the beep, but never for the treatment.
Cheaper — but for whom?
Every euro and every tooth-year in this review came from the German system: its fees, its wages, its reimbursement rules. Those numbers don’t travel. What looks cost-saving in Berlin may look very different in Boston, Bangalore, or a rural clinic with no radiograph reader to begin with. The review found zero economic evidence for AI in oral cancer or gum disease, and none from low- or middle-income countries — precisely where oral-disease burden is heaviest and budgets thinnest. AI could widen access by extending expert-level screening into underserved places, or it could deepen the divide if only well-capitalized practices can afford the license. On today’s evidence, we simply don’t know which.
What this means for you
If you’re a patient
AI in your dentist’s office should mean earlier, gentler catches — an early spot that gets sealed or watched, not an automatic filling. If a scan flags something, it’s fair to ask: does this need treating now, or watching? “The computer found it” is not, by itself, a reason to drill.
If you’re a clinician
The economic case for AI isn’t the accuracy printed on the box — it’s your own treatment threshold. Paired with minimally invasive protocols, AI can save cost and teeth; bolted onto a drill-first workflow, it mostly adds expense. And insist on validation in a population like yours, not a German cavity model.
The bottom line
AI can sharpen the dentist’s eye, but it can’t decide what a finding is worth — that remains a clinical, and human, judgment. Until the economics are tested well beyond a handful of German cavity models, “AI saves money in dentistry” is a hypothesis in a lab coat, not a proven fact. Treat it as a promising assistant to be audited, not an oracle to be trusted.
Frequently asked questions
Does AI in dentistry actually save money?
Sometimes, in cost models — but the only randomized trial in this 2026 review found no savings, with identical costs (€330 vs €330) and identical tooth retention (49 vs 49 years). The overall evidence rests on just four studies, so any claim of guaranteed savings is premature.
Why did only four studies qualify?
The reviewers searched six databases and screened 334 records, but only four met the bar for a genuine economic evaluation of AI in oral-disease screening. All four were conducted in Germany and all focused on detecting cavities on bitewing X-rays.
Is AI better than a dentist at spotting cavities?
In one cost model, AI read cavities more accurately than dentists (0.80 vs 0.71) and caught many more early lesions (sensitivity 0.75 vs 0.36). But higher sensitivity can trigger more invasive treatment, which in the randomized trial cancelled out the expected cost benefit.
Should my clinic buy dental AI to cut costs?
The current evidence doesn’t support a cost guarantee. Savings depended on conservative, non-restorative management of early lesions and on the local reimbursement system — plus performance validated in a population like yours. AI is a tool that can help, not a switch that lowers costs on its own.
What is a “tooth-year”?
It’s a measure of how long a tooth is kept in the mouth. The models estimated up to about 64 vs 62 tooth-years with versus without AI; the randomized trial found 49 vs 49 — no difference.
Source & author credit
This article interprets, and does not reproduce, the following peer-reviewed study. All figures are the authors’ original findings.
Sahoo RK, Sinha A, Satpathi S, Kumar G, Sahoo KC, Bhattacharya D, Panda B, Pati S. Economic evaluation of AI-based oral disease screening: a systematic review. Frontiers in Medical Technology. 2026;8:1792401. DOI: 10.3389/fmedt.2026.1792401
ORCID — no ORCID iDs were registered for the authors in the article’s Crossref metadata at the time of writing.
Published open access under a Creative Commons Attribution 4.0 (CC BY 4.0) license. Decadentry is an independent educational publication and is not affiliated with the study’s authors.
Leave a Reply