A dental bitewing radiograph drawn in ledger lines, read by one AI overlay from the chair side and a mirrored overlay from the payer side

One vendor, both sides of the claim.

The last Bench Test looked at Pearl Second Opinion. This week we run the same five-line scorecard on Overjet AI Assist, the product most people mean when they say "the other one." Overjet is genuinely well engineered — and it is also the vendor whose subscription funds the machine that, on the other side of the table, reads your radiographs a third time and denies your claim. That is not a rhetorical flourish. It is what the company sells. So we have to grade both halves.

The rules are unchanged. I do not take money from vendors. Every fact below has a link. Anything a vendor will not publish gets "NOT PUBLISHED" instead of a guess. And I bench-test each product against the same unit — one chair-hour of clinical time, roughly $154 in fixed burn on the 1,600-hour year.

The five-line scorecard

Line

Overjet AI Assist

Evidence

Seven FDA 510(k) clearances 2021–2025 including K210187 (Dental Assist), K222746 (Caries Assist), Charting Assist, and CBCT Assist (Dec 2025); positioned by independent review as the only dental AI cleared to detect, outline, and quantify disease. Vendor cites a 32%-more-carious-surfaces figure in a 7,000-surface Caries Assist analysis with no published study, author, journal, or DOI. Overjet-affiliated authors (including CEO Wardah Inam) appear on real PubMed papers about AI in dentistry (PubMed search), but no head-to-head accuracy study against Pearl. Grade: B+.

Workflow fit

PMS integrations reported for Dentrix, Eaglesoft, and Open Dental (review.dental); Overjet's own pages say sensor-agnostic and publish no PMS list (overjet.com). Full platform is AI Assist + IRIS (AI-native imaging viewer that can replace legacy suites) + Overjet Voice (ambient documentation, GA early 2026). Per-image speed, setup time, training time: NOT PUBLISHED. Adopting the full platform is not a feature flip — SmbAI's comparison calls it a multi-week software migration for anything larger than a solo.

Unit economics

No published price. No public pricing page. Quote-only. Third-party reports of AI Assist alone: $250–$500/month. Full platform (AI Assist + IRIS + Voice): $1,000–$3,000/month, per SmbAI and ai.dentist ranges cited by review.dental. On the chair-hour frame: $250/mo → $1.88/chair-hour; $500/mo → $3.75/chair-hour; $1,500/mo (mid full platform) → $11.25/chair-hour on a single-chair 1,600-hour year. Setup fee, bundle price, three-year total, cancellation clause: NOT PUBLISHED.

Data terms

BAA offered — HIPAA policy says Overjet "fully complies with all business associate obligations under HIPAA/HITECH." Storage is U.S.-based, in Overjet facilities and third-party data centers (Privacy Policy). Two things every owner should read twice: retention is functionally forever — "Overjet stores all Traffic Data and review postings indefinitely, even after 'deletion,' and may archive such information elsewhere" (source); and training use is explicit — "We use your data to provide the Services and to continually improve and develop the Services" (source). Image-ownership language is Data-Processor/Data-Controller framing, not the plain "you own your images" statement solos want to see.

Verdict by buyer size

Solo: conditional yes on AI Assist only ($250–$500/mo), if you can negotiate a month-to-month or 12-month clean break — otherwise Pearl. Group (2–10): yes on AI Assist + optional Voice; skip IRIS unless you were already replacing the imaging viewer. DSO: yes to the full platform — this is where Overjet lives, wins, and where the enterprise rollout playbook is worth the price.

What the vendor page will not tell you

Two things you will not find on overjet.com that a buyer should know before signing.

One — Overjet also runs the payer side. The same class of technology that reads your radiographs at the chair reads them again at the insurer. Per review.dental, Overjet's payer platform reviews claims against medical-necessity guidelines across 350+ procedure codes for most of the ten largest U.S. dental insurers, and advertises a 90% reduction in manual claim review. Providers report AI-driven denials, bone-level disagreements with the treating clinician, and appeal logic that is hard to fight. The published position of the vendor is that this is good for the system. The lived experience in the operatory is that your subscription is helping co-fund the denial engine on the other side of your claim. You do not have to conclude anything from that. You do have to know it before you sign.

Two — the pricing spread is the widest in the category. Reported figures for the same product range from $250/month to $3,000/month depending on which combination of AI Assist, IRIS, and Voice you get, whether pricing is per-location or per-provider, and how well your practice manager negotiates. There is no published rate card. That is a policy choice, not an oversight, and it means the number you pay is a function of how much friction you are willing to put on the sales cycle. Ask for the three-year TCO in writing, not just the monthly.

How Overjet compares to Pearl on the one line most owners get wrong

The data-terms line is where these two products diverge most, and it is the line most owners skim.

Pearl Second Opinion

Overjet AI Assist

BAA offered

Yes

Yes

Storage location

U.S.

U.S.

Training use of your images

Reviewed in Bench Test #1 — check the current Pearl Privacy Policy yourself before you sign

Explicit: your data is used to "continually improve and develop the Services" (Overjet Privacy Policy)

Retention

Reviewed in Bench Test #1

Indefinite — traffic data and review postings stored "even after 'deletion'" (Overjet Privacy Policy)

Image ownership language

Reviewed in Bench Test #1

Data Controller / Data Processor framing rather than an explicit "you own your images" statement

You may be fine with any of this. Many owners are. The point is that the plain-English answer to "do they train on my patients' images?" is yes, and the plain-English answer to "how long do they keep them?" is indefinitely. That is not a scandal. It is a business model. Buy it with your eyes open, or negotiate the terms out of the contract before you sign.

What I would actually do this quarter

If I were shopping today for a single-doctor practice: I would trial Pearl and Overjet AI Assist on the same twenty anonymized radiographs and compare false-positive rate on early interproximal lesions and readability of the overlay, not on the marketing deck. Whichever one my hygienist and I both stop overriding gets the seat.

If I were a two-to-ten-provider group: AI Assist plus a short Overjet Voice pilot on real appointments, hold IRIS unless my imaging viewer was already up for replacement.

If I were a DSO: full Overjet platform, negotiated with an annual term and a 12-month clean-break clause per SmbAI's guidance, and a separate internal policy on what my clinicians do when the payer-side Overjet review disagrees with the treating dentist. That last one is a policy, not a product decision. Write it before you need it.

Next Bench Test target is up to you — reply with the vendor name.

Practice Ledger takes no vendor money — no industry sponsors, no affiliate links. Every claim above links to a source. If a fact is wrong, tell me; corrections append here, never silently rewrite.

The clinical side of this desk: OVN Nexus, my research platform, now carries a free two-minute brief for dentists and their teams on what the oral-systemic evidence supports and where it stops, with a one-page discussion guide for the next team meeting. That page points here; this one points there.

Disclosure: I’m building a small group of practices myself (that’s the Diaries), so I sit on the buy side of this table, and I have a commercial interest in periodontal care. Nothing in this issue names it or is sold by it. The OVN Nexus brief linked above is free clinical education from my research platform; it discloses its own support on the page.

— Thad

Bench Test scorecard: evidence grade, workflow fit, unit economics, data terms, verdict by buyer size. One product per issue. Neutral data publication.