Vertical notes
AI for dental practices: a field report
We mapped a typical family practice process by process, then graded AI against an answer key. The dentistry stays human, full stop. The paper layer — 42% of staff hours — is the target.
Open the visual report — same data, built to skim.
The DSOs are buying your competitors, and you can't even hire a hygienist. About 13% of US dentists are affiliated with a dental service organization — roughly double that among dentists under 35 — and around 4 in 10 dentists are 55 or older, which means a wave of practices changes hands this decade. Solo practices typically sell for 60–80% of one year's collections; PE-backed platforms are priced on profit multiples. The gap is transferable process. Meanwhile, most practices recruiting hygienists call it very or extremely difficult. Either way you read it — defend your independence or build toward an exit — the move is the same: take the paperwork off the desks you already have.
Where the hours go
A representative family practice — two dentists, hygiene, front office. In a real engagement, every row is measured, not modeled.
| Activity | Share of practice hours | What AI can carry |
|---|---|---|
| Chairside care — the actual dentistry | 25% | 0% |
| Clinical notes & charting | 15% | ~40% |
| Claims, denials & resubmissions | 14% | ~70% |
| Scheduling, recall & no-shows | 13% | ~65% |
| Insurance verification & benefits | 12% | ~75% |
| Patient billing & statements | 9% | ~60% |
| Front desk phones | 8% | ~35% |
| Treatment-plan follow-up | 4% | ~15% |
Representative family practice. "What AI can carry" = share of that activity's hours where graded performance clears the practice's bar. Estimates until a pilot measures them. Chairside is 0% by design.
42% of all staff hours sit in the paper layer. The dentistry is not the automation target. The paperwork is. Nobody gets cut — the front office stops drowning and the chairs stay full.
Don't trust AI. Grade it.
AI redoes the practice's past paperwork, gets graded against what the team actually did, and touches nothing until it passes.
We ran the first grading pass — 40 cases on made-up dental documents (no patient data, no PHI), scored automatically on the Recursiv platform, July 29, 2026:
| The work | Right | Handed back | Cost each | Verdict |
|---|---|---|---|---|
| Sorting incoming faxes & documents 20 docs: EOBs · referrals · cards · lab slips · traps | 90.0% | 0% | $0.005/doc | On, with review |
| Posting EOB numbers 20 fields, graded one field at a time | 90.0% | 10% | $0.005/field | On, with review |
| Insurance verification | — | — | — | Measured in pilot |
| Recall & no-show outreach drafts | — | — | — | Measured in pilot |
| Denial appeal letters, first draft | — | — | — | Measured in pilot |
| Clinical note first drafts | — | — | — | Measured in pilot |
Right = matches the answer key. Handed back = the AI gave it to a person instead of guessing. Bar: 98% to run alone, 90% to run with review.
Read the failures, because they're the product. One document sort garbled its reply on a HIPAA form. Another filed a CBCT imaging report under LAB instead of REFERRAL — a genuine judgment miss. Both EOB hand-backs were a smudged fax, and handing them back was the correct call. Neither task cleared the 98% run-alone bar on day one. That is exactly why you grade before you trust.
Patient data never leaves the practice
PHI lives under HIPAA's Privacy and Security Rules and HITECH. So Recursiv runs on the practice's own servers — the AI comes to the records, the records don't travel, and every action lands in an audit log. Outside AI services are off by default, behind a BAA and no-storage terms, minimum-necessary only.
The path
- Walk the floor — who does what, hour by hour (~2 weeks).
- Practice runs — AI works alongside the team on real paperwork; patients never see it (~4–6 weeks).
- The scorecard — what passed, what didn't, against the practice's own bar.
- Turn on what passed — behind review gates. Everything else keeps practicing until the numbers earn it.
If you run a practice and want this scorecard measured on your own work, that's what a pilot is. Get in touch.
Sources
- ADA Health Policy Institute: ~13% of US dentists DSO-affiliated (2022); roughly double among dentists under 35.
- ADA HPI polling: most practices recruiting dental hygienists report recruitment very or extremely challenging.
- ADA workforce data: roughly 4 in 10 US dentists are age 55+.
- Practice-broker rules of thumb: solo practices ≈60–80% of annual collections; DSO platform deals quoted on profit (EBITDA) multiples. Illustrative, not an appraisal.
- Scorecard rows 1–2: live graded run on the Recursiv platform, July 29 2026, 40 synthetic cases, $0.21 total inference cost. No patient data.
Quick answers
How accurate is AI at dental front-office paperwork today?
In our July 2026 graded run, AI sorted incoming faxes and documents at 90% and posted EOB fields at 90%, at about half a cent per item — and correctly handed smudged faxes to a person instead of guessing.
Will AI replace dentists or hygienists?
No. Chairside care is 0% of the target by design. The paper layer — insurance verification, claims, recall, notes — is 42% of staff hours, and that is where AI helps.
Is this HIPAA compliant?
The deployment runs on the practice's own servers: PHI never leaves, every action is logged, and no outside AI service is used without a BAA and no-storage terms.
What does a pilot look like?
Six to eight weeks: map the hours, let AI practice silently on real paperwork, then a scorecard decides what turns on. If the grades don't clear the bar, nothing switches on.