ScanReferee — remake-fault referee for independent dental labs

1. One-liner

Grades every incoming dental scan at intake, catches defects before you mill, and builds the who-caused-it remake record.

2. Trend signal — why now?

Two things collided in the last 18 months. First, the small dental lab is getting squeezed: the number of formal US dental laboratories has been shrinking (~4,375 businesses, declining ~1.6% CAGR 2021–2026, on a $7.6B industry), most of them owner-operated shops with 1–20 techs, and DSOs keep pushing turnaround and price. Second, the remake — the single biggest silent margin leak in a lab — is still running at a 4% national average with per-dentist rates documented as high as 42%, and the lab usually eats it. A “no-charge” zirconia remake puts the lab roughly $75–$150 in the hole per unit and it has to claw that back out of the next several crowns’ profit.

The kicker is why remakes happen. The literature is blunt: margin-fit discrepancy alone accounts for ~29.5% of remakes, and the dominant root causes are the dentist’s, not the lab’s — inadequate prep, poor impressions (“garbage in, garbage out”), and incomplete prescriptions (industry sources cite that ~80% of dentists don’t complete the required Rx details). The lab knows the scan was bad. It has no clean, objective way to prove it at the moment the case arrives, so it either mills anyway and remakes for free, or awkwardly chases the dentist and risks the account.

What changed on the tech side: AI margin detection, prep-discrepancy flagging (inadequate taper, undercuts), and auto-segmentation on intraoral scans went from research to table stakes in 2025–2026 — but that intelligence lives inside the dentist’s premium scanner or premium CAD. The small lab receiving STL files from a dozen different dentists on a dozen different scanner brands has no unified, lab-side gate that runs at intake and produces a defensible record.

Provenance:

• Signal 1 (demand): Dental lab remake rates average ~4% (range 1–7%, per-dentist to 42%); labs eat the cost — a no-charge zirconia remake is ~$75–$150 out of pocket; margin-fit = 29.5% of remakes; ~80% of dentists send incomplete Rx — Spear Education “The Cost of Laboratory Remakes” & PubMed prosthodontics remake studies — https://www.speareducation.com/resources/spear-digest/the-cost-of-laboratory-remakes/ — accessed 2026-07-26
• Signal 2 (feasibility): AI margin detection, prep-discrepancy flagging (taper/undercut), and STL auto-segmentation are now standard capabilities, exportable via API, reducing remakes — Yucera “AI Integration in Dental Scanning and Design 2026” — https://www.yucera.com/blogs/ai-integration-in-dental-scanning-and-design-2026/ — accessed 2026-07-26
• Signal 3 (economic): US dental laboratory industry $7.6B, ~4,375 formal businesses declining 1.6% CAGR, highly fragmented, small labs squeezed by DSOs; adjacent lab case-tracking tools (TrazaLab) active with pre-milling clarification but no fault-attribution/billing layer — IBISWorld Dental Laboratories 2026 + TrazaLab — https://www.ibisworld.com/united-states/industry/dental-laboratories/4087/ — accessed 2026-07-26 Category: Tech-unlock

3. The opportunity

Every independent lab is running an informal, undocumented version of this already. A tech opens the case, squints at the scan, decides “the margin’s unreadable / the prep has an undercut / there’s no antagonist,” and then makes a judgment call: mill it and hope, or email the dentist and risk annoying a paying account. When the crown comes back, the dentist says “your fit is off,” the lab says “your scan was bad,” and because nobody has a timestamped, objective record from the moment of intake, the lab caves and eats it. Do that 4% of the time across a few thousand units a year and it’s real money — plus the chair-time and goodwill damage on the dentist’s side.

The incumbents don’t cover this seam. Premium scanners and CAD (exocad, 3Shape, SprintRay) grade scans on the dentist’s side or mid-design — great if the dentist owns premium gear and actually looks, useless for the receiving lab that gets whatever the dentist exports. Lab-side case trackers like TrazaLab organize the workflow (Received → In Design → Shipped) and can attach photos, but they don’t automatically grade the incoming scan for defects and they don’t produce the fault-attribution record that lets the lab bill or coach the dentist. ScanReferee owns exactly that gap: a scanner-agnostic intake referee that runs the second a file lands, and turns a subjective “I think your scan is bad” into an objective, dated, defensible artifact.

The 10× isn’t the AI grading — that’s increasingly commoditized. The 10× is putting it on the lab’s side of the fence, brand-agnostic, and wiring it to the money: prevent the remake, or make the remake billable.

4. Target market

• Primary customer: Owner or production manager of an independent US dental lab, 2–25 employees, processing roughly 300–3,000 restorative units/month, receiving digital cases from 15–150 dentist accounts across mixed scanner brands (iTero, 3Shape TRIOS, Medit, Primescan).
• Why they buy: In their words — “I’m eating remakes that aren’t my fault and I can’t prove it,” “half my crew’s time goes to chasing dentists for a decent rescan,” and “if I push back too hard the doctor sends the case to the lab down the street.” They need leverage without losing the account.
• Rough TAM reasoning: ~4,375 formal US labs (14,000+ if you count single-tech studios). If ~2,000 are digitally-mature enough to have a file-based intake worth gating, at $300–$600/mo that’s a $7M–$14M ARR ceiling in the US alone before adding Canada, UK, and the DSO-owned mid-market. Small TAM — exactly the bootstrapper’s lane, too small for a VC to bother crowding.
• Why now for them: Digital case volume crossed the threshold where scans are the default input, not the exception, so a software gate finally applies to most of the work. And the DSO squeeze means margin defense (not growth) is the lab owner’s 2026 priority.

5. Product sketch (MVP)

• Drop-zone intake: lab forwards or auto-routes incoming STL/PLY case files (any scanner brand); ScanReferee grades each within minutes.
• Defect scorecard per unit: margin visibility, prep taper/undercut flags, missing antagonist/bite, insufficient inter-occlusal clearance, incomplete Rx fields — each rated pass / marginal / fail with the specific location annotated on the 3D view.
• “Rescan or proceed” recommendation: a clear go/no-go so the tech doesn’t burn a puck on a case that will remake.
• Dentist-facing clarification note (one click): auto-drafts a polite, specific message with the annotated evidence image — “margin on #19 distal is obscured by tissue, please rescan” — instead of a vague phone call.
• Fault-attribution record: a timestamped, immutable intake report per case (who sent what, what was flagged, what the dentist was told) that becomes the evidence if a remake dispute happens later.
• Remake billing helper: when a case flagged “proceed at dentist’s risk” comes back, one click assembles the record and drafts the remake charge or credit-denial to the dentist.
• Per-dentist quality dashboard: ranks accounts by defect rate so the lab can coach the worst offenders (or fire the account) with data, not vibes.

6. AI angle — what’s load-bearing

Remove the AI and there is no product. The core is a vision/geometry model reading a raw 3D scan and reliably calling margin visibility, prep defects (taper, undercut), and bite/antagonist problems — the exact judgments that today require an experienced tech to eyeball every case. That grading is what makes intake automatic and scalable; a human doing it case-by-case is just the status quo the lab already can’t afford. The second AI job is turning a flagged defect into a specific, professional, dentist-appropriate clarification note (right tooth, right issue, right tone) — a language task that makes techs actually use the gate instead of avoiding the awkward call. No AI, and you’re left with a case-tracker checklist, which already exists and doesn’t solve the who-pays problem.

7. Localization angle (if any)

N/A — this is a US-first play. The wedge is the US remake-cost structure and the fragmented independent-lab market. It ports cleanly to Canada/UK/Australia (same scanner brands, same remake economics, same English Rx), which is the natural expansion, not a localization rework. No payment-rail or language advantage in India/SEA where lab economics and digital penetration differ enough to be a separate business.

8. Business model — path to $1M–$5M ARR

• Pricing: Tiered by volume. Starter $299/mo (up to ~400 units graded), Pro $499/mo (up to ~1,200), Shop $899/mo (unlimited + per-dentist analytics + billing helper). Optional per-unit overage.
• ACV: ~$5,500 blended (most paying labs land on Pro/Shop).
• Rough math to $1M ARR: ~180 labs × ~$460/mo × 12 ≈ $1.0M. That’s under 10% of the ~2,000 digitally-mature US labs.
• Rough math to $5M ARR: 750 labs at a slightly higher blended ACV ($550/mo) as billing-helper and analytics upsells land, plus Canada/UK expansion. Requires becoming the default intake tool in the independent-lab segment — plausible but needs a distribution engine, not just word of mouth.
• Expansion path: land on grading, expand to (1) remake-billing recovery (charge a small % of recovered remake charges), (2) per-dentist analytics as a retention hook, (3) a dentist-side “score before you send” companion that the lab can offer its accounts — turning the lab into a distribution channel for you.

9. Go-to-market wedge — first 100 customers

• Lab directories + cold outreach with a free audit: pull the NADL member list and state dental-lab association rosters (thousands of named labs). Offer a free “remake-leak audit”: send us last month’s 20 cases, we’ll show you which ones we’d have flagged and the dollars you likely ate. Personalized, evidence-first — expect a strong reply rate because you’re quantifying a pain they already feel.
• Dental Lab Network + LMT (LabManagement Today) channels: the independent-lab world lives in a small number of forums, the LMT magazine/conference, and NADL events. A booth at LMT LAB DAY and a founder writing “here’s what your remakes actually cost you” content hits nearly the entire target market in one venue.
• Scanner-agnostic angle as the hook: target labs that get files from mixed scanner fleets (the majority) — they’re the ones with no unified gate. Message: “grade every dentist’s scan the same way, whatever they sent.”
• Referral loop through the pain: a lab that stops eating remakes tells the other lab owners it drinks with at the association meeting. This is a tight, gossipy, referral-driven market — nail 15 flagship labs and the segment hears about it.

10. Build complexity — justification

Medium. The web app, file intake, doc/note generation, dashboards, and billing helper are all off-the-shelf standard-stack work. The real work is the scan-grading model: STL/PLY geometry parsing plus a vision/geometry model that reliably flags margin visibility, taper/undercut, and bite defects across scanner brands. The underlying capability exists (it’s now standard inside scanners and CAD), and there are open datasets and API building blocks, but tuning it to be trustworthy on messy real-world lab inputs — and validating it against real tech judgments — is 3–5 months for a small team with a dental-domain advisor. Not research-grade, but not a weekend either.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Internal lab workflow/QC tool; not a diagnostic or treatment-planning medical device. Positioned as decision-support for the technician, not automated clinical judgment.
Ethical — no harm / dark patterns	✅	Improves restoration quality and honest fault attribution; no dark patterns. Keep it advisory, never auto-rejecting a case without a human.
Market exists (evidence above)	✅	$7.6B industry, documented remake costs, active adjacent tools.
1–5 person team can build this	✅	Standard stack + one hard model; small team in ~4–5 months.
Launchable with <$50K / ₹40L	✅	Solo/pair build + off-the-shelf compute; well under $50K to a paying pilot.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real, recurring, dollar-quantified pain — but it’s a margin leak the lab has tolerated for years, not a hair-on-fire “shut down tomorrow” crisis. Felt weekly, worked around today.
Demand evidence	15	11/15	Strong sourced stats (remake rates, cost, root causes) and active adjacent tools, but I could not pull 5–10 verbatim customer quotes — the core lab forums block scraping. Downgraded honestly for thin first-person voice.
Build feasibility	15	10/15	Everything but the model is standard. The scan-grading model is real, buildable work — trustworthy cross-brand accuracy is the risk, ~4–5 months.
Distribution clarity	15	11/15	Tight, reachable, gossipy market with named directories and one dominant conference (LMT LAB DAY). Free-audit wedge is concrete. Conversion math still unproven.
Revenue mechanics	15	12/15	Clear per-lab SaaS pricing benchmarked against existing lab software; ACV and customer counts to $1M are conservative. $5M needs real expansion.
Time to first revenue	10	7/10	Free-audit → paid pilot funnel can close in 6–8 weeks, but the model has to be good enough to demo credibly first, which gates launch.
Defensibility	10	5/10	Execution + accumulating per-dentist quality data + workflow lock-in. But the grading capability is commoditizing fast; the moat is lab-side integration and the fault-attribution/billing layer, not the AI itself.
Total	100	71/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required — needs someone who can ship a geometry/vision model and a dental-lab advisor who knows what a tech actually looks for and won’t tolerate false flags.

Key assumptions to validate (3–5)

1. Assumption: Labs will actually pay $300–$900/mo to defend remake margin (vs. continuing to eat it). How to test: 30 discovery calls with independent lab owners; offer 5 the free audit and ask for a signed pilot LOI at the price.
2. Assumption: The grading model can hit tech-trusted accuracy (low false-flag rate) across mixed scanner brands. How to test: blind-test the model against 3 senior techs on 100 real anonymized cases; require ≥85% agreement and low false-positive rate.
3. Assumption: Labs will use the dentist-facing note rather than avoid the confrontation. How to test: in pilots, measure what % of flagged cases actually trigger a sent clarification note within 48h.
4. Assumption: Fault attribution translates into recovered dollars (billed remakes / avoided free remakes), not just a nice report. How to test: track pilot labs’ remake-cost delta over 60 days vs. their baseline.

Risk flags

1. Trust / false-positive risk: if the model flags good scans, techs stop trusting it and the tool dies. Accuracy is existential, not a nice-to-have.
2. Channel-conflict / relationship risk: labs fear that pushing back on scans loses accounts. If the tool feels like it creates dentist friction rather than defusing it, adoption stalls. The polite-note framing is the mitigation and must be great.
3. Commoditization risk: scanner and CAD vendors keep absorbing scan-grading. The defensible layer must be the lab-side, cross-brand, fault-attribution/billing workflow — not the grading, which will keep getting cheaper.
4. Regulatory framing risk: must stay clearly on the QC/decision-support side of the line and never drift into automated clinical judgment that could pull it into medical-device territory.

14. Structured verdict

Score:                  71/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (geometry/vision) + dental-lab domain advisor
Time to revenue:        6–10 weeks after a demo-credible model exists (~4–5 months to that point)
Capital to launch:      $15K–$40K (compute + advisor + pilot)
Top 3 assumptions to validate first:
  1. Willingness to pay $300–$900/mo — 30 owner calls + 5 signed pilot LOIs at price
  2. Model accuracy ≥85% tech agreement, low false-positive rate — blind test vs. 3 senior techs on 100 real cases
  3. Fault attribution recovers real dollars — 60-day remake-cost delta in pilots
Kill criteria:
  - Abandon if fewer than 3 of 30 labs will sign a paid pilot LOI at target price
  - Abandon if the model can't clear ~85% tech agreement with a low false-flag rate on real messy inputs
  - Abandon if a scanner/CAD vendor ships a free, brand-agnostic, lab-side intake gate before your v1

15. Next step — 1-week validation sprint

• Day 1–2: Pull the NADL + state-association lab lists. Line up and run 12–15 calls with independent lab owners. Ask: what’s your remake rate, who eats it, how do you handle a bad scan today, would you pay to gate intake and bill the dentist? Listen for the who-pays pain in their own words.
• Day 3–4: Collect ~50 real anonymized case files from 2 friendly labs. Hand-grade them with a senior tech to build ground truth, then run existing off-the-shelf scan-analysis APIs against them to see how close commoditized tooling already gets — this tells you how much model work you actually own.
• Day 5: Go / no-go. Go only if (a) ≥5 of the ~15 owners say they’d pay $300+/mo and ≥2 will sign a pilot LOI, and (b) off-the-shelf grading gets close enough that trustworthy accuracy looks reachable in ~4 months. Falsifiable: fewer than 5 interested or the tech gap looks research-grade → no-go, revisit when grading APIs mature further.