ChartExposure

1. One-liner

Reads every note behind your paid crown and SRP claims and ranks which ones a payer will claw back.

2. Trend signal — why now?

Three things moved in the last eight months, and they moved in the same direction.

Delta Dental industrialised its audit function in 2026. Liles Parker, a healthcare defence firm, published in January 2026 that Delta Dental is adopting “a more proactive and technologically advanced approach to fraud prevention and enforcement in 2026,” expanding AI and predictive analytics to spot suspicious billing patterns in real time. The mechanism named is predictive risk scoring — a model that reads current and historical claims to forecast fraud likelihood, shortening the lag between claim submission and audit initiation. The firm names the high-risk codes explicitly: D4341/D4342 (scaling and root planing), D3310–D3330 (root canals), plus crowns, onlays and veneers. Their flagged triggers are comparative: total billings per patient versus peer providers, services per visit, percentage of high-end procedures.

The critical line for a practice owner: “Delta Dental will deny payment for any claims that cannot be supported by patient dental records.” Not “future claims.” Records already submitted, for money already banked.

The evidentiary standard moved at the same time. Carriers are shifting away from written narratives in appeals and increasingly demanding the original clinical notes. A practice that spent a decade getting good at writing persuasive narratives just discovered the narrative isn’t the artifact under review — the chart note written chairside in nine seconds is.

Dentists are already fighting back in court. In May 2026 the ADA reported dentists filing class action lawsuits against Delta Dental in four states. Whatever the merits, this is the market’s tell: the money at stake is large enough that dentists are hiring litigators. People who hire litigators buy software.

Meanwhile the documentation itself is genuinely bad, and the profession admits it. From DrBicuspid’s 2026 piece on clinical notes: “One of the most common mistakes I see is when the reason for a procedure… is documented as ‘Insurance allows replacement every five years.’ That’s not a clinical reason. That’s an insurance policy.” That single sentence is a recoupment demand waiting to be written.

Provenance:

• Signal 1 (Demand): Delta Dental audits expected to increase in 2026 via AI predictive risk scoring; “Delta Dental will deny payment for any claims that cannot be supported by patient dental records”; named high-risk codes D4341/D4342, D3310–D3330, crowns/onlays/veneers — https://www.lilesparker.com/2026/01/13/delta-dental-audits-will-increase-in-2026/ — 13 Jan 2026
• Signal 2 (Feasibility): Open Dental publishes a fully public, self-serve REST API covering Patients, Appointments, clinical Notes and procedure logs, readable without a partnership agreement — https://www.opendental.com/site/apisetup.html — accessed Aug 2026
• Signal 3 (Economic): Dentists filed class action lawsuits against Delta Dental in four states, May 2026; dental billing outsourcing already runs 3–10% of collections or $500–2,000/mo flat, proving established back-office willingness-to-pay — https://adanews.ada.org/ada-news/2026/may/dentists-file-class-action-lawsuits-against-delta-dental-in-four-states — May 2026 Category: Tech-unlock

3. The opportunity

The asymmetry is the whole idea.

A dental chart audit today is a human reading a sample. Dentistry Support’s service — a representative example of the category — audits “10–20 randomly selected charts,” then delivers an action plan. That was a defensible methodology when the payer was also sampling.

The payer stopped sampling. Predictive risk scoring reads the practice’s entire claim history and ranks it. So the practice is defending a population with a sample, and the payer is attacking a population with a population. A 15-chart audit that comes back clean tells you approximately nothing about the 4,000 notes behind three years of paid claims.

Nobody is selling the symmetric product. The dental AI market has money and attention, but it is pointed forward: Bola AI (10,000+ dentist and hygienist users, voice-powered charting), Overjet Voice, Pearl, Denti.AI Scribe — all ambient scribes that help you write a better note today. Excellent products. Completely useless against the exposure that already exists in the chart, because the notes at risk were written in 2023, 2024 and 2025 and no scribe is going back for them.

So there are two piles of software: scribes that improve future notes, and human consultants who sample past notes. The gap in the middle — read every historical note behind every high-risk paid claim, score it against what an auditor actually demands, rank the practice’s dollar exposure — is unoccupied.

The wedge sharpens further because the target codes are public. Liles Parker named them. I don’t need to guess what to inspect; I inspect D4341/D4342, D3310–D3330, and the crown/onlay/veneer codes, and I check for the specific artifacts auditors demand: pre- and post-operative radiographs for endodontics, radiograph metadata and audit trails, documented medical necessity, evidence of tooth structure loss. That’s a checklist, not a research problem.

And the output is denominated in dollars, not in advice. “You have $180K of paid claims sitting behind notes that fail on medical necessity, concentrated in 62 SRP cases from your hygienist’s 2024 column” is a sentence that makes an owner reach for a card. “Your documentation could be stronger” is not.

4. Target market

• Primary customer: Owner-dentist of an independent practice or small group, 1–5 locations, US, in-network with Delta Dental (the largest US dental benefits carrier by membership) and billing a meaningful share of D4341/D4342 and crown work. Practices doing $700K–$3M annual collections. The buyer is the owner, sometimes with the office manager as champion. Not DSOs — they have compliance staff and a procurement cycle.

• Why they buy: Recoupment is retroactive, unbudgeted, and lands as a letter demanding money the practice already spent. Insurers may “request refunds, place the practice on a corrective action plan, or in serious cases, terminate the provider agreement.” Termination from a major network is a revenue event, not a compliance event — it’s why this pain outranks ordinary compliance nagging. In California, a dentist appealing a recoupment demand has 30 days from notification. Thirty days is not enough time to reconstruct three years of charting. The product’s pitch is that you find out before the letter, not after.

• Rough TAM reasoning: 178,461 active US dental practices (ADA Health Policy Institute / IBISWorld, 2026); roughly half of dentists still in solo practice. Strip out DSO-owned and non-insurance-participating practices and a conservative serviceable base is 40,000–60,000 independent practices with real in-network exposure. At $349/mo, 500 of them is $2.1M ARR. I need less than 1.5% of the serviceable base for a very good business.

• Why now for them: Their exposure didn’t change — the probability of it being discovered did. Predictive scoring plus a shortened audit lag means the practice that was statistically invisible in 2024 is legible in 2026.

5. Product sketch (MVP)

• Read-only connect to the practice management system (Open Dental first — public self-serve REST API covering clinical notes and procedure logs; Dentrix and Eaglesoft after).
• Exposure report on day one: every paid claim in the last 24–36 months on a high-risk code, with the note behind it scored Green / Amber / Red against auditor-demanded elements.
• Dollar-ranked worklist — not an alphabetical list of problems. Sorted by recoupment value at risk, so the owner fixes the $40K of SRP before the $600 filling.
• Failure reasons in plain language, tied to the specific missing artifact: no perio charting depths, no pre-op radiograph on the endo, medical necessity stated as an insurance policy rather than a clinical finding.
• Radiograph presence and metadata check — flags endodontic claims with no pre-op or post-op image attached to the record, which the audit guidance names directly.
• Provider and hygienist heatmap — exposure concentrated by operator and by year, because it almost always is, and that’s the conversation the owner needs to have.
• Peer-pattern self-check — the practice’s own ratio of high-end procedures and services per visit, i.e. the same comparative signals the payer’s model reportedly scores on.
• Addendum trail — where a note can still be legitimately supplemented, a dated, clearly-marked addendum workflow. Never backdating, never rewriting history.

6. AI angle — what’s load-bearing

Remove the AI and there is no product — you’re back to a consultant reading 15 charts.

The load-bearing work is judgment over unstructured free text at volume. A clinical note is prose written fast by a clinician with a patient in the chair. Deciding whether it establishes medical necessity for a crown means reading it the way an auditor reads it: is there a documented clinical finding, or just a restorative history and a benefit rule? That’s a semantic call across thousands of short, messy, abbreviation-dense documents. No rules engine gets there — dentists don’t write in templates, and the failure mode (“insurance allows replacement every five years”) is grammatically fine and clinically empty. Catching that is exactly what an LLM is good at and what regex is not.

Volume is the second half. A three-year lookback for a mid-size practice is thousands of notes. Cheap inference is what turns a $4,000 consulting engagement covering 15 charts into a $349/mo subscription covering all of them. That’s the tech-unlock: the same job, three orders of magnitude more coverage, at a tenth the price.

One discipline I’d hold hard: the model flags and explains, it never rewrites clinical history. The product’s value is telling the truth about exposure early. A tool that helps a practice launder documentation is both unethical and, in an audit trail-aware PMS, actively dangerous to the customer.

7. Localization angle

N/A — this is a US-only play. The entire opportunity is manufactured by a specific US payer market structure: commercial dental benefit carriers with retroactive recoupment rights, CDT procedure codes, state-level appeal windows, and a concentrated carrier (Delta) industrialising audits. Single-payer and public dental systems elsewhere don’t produce this pain. Attempting a global version dilutes the only thing that makes it sharp.

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

• Pricing: $349/mo per location, annual billing preferred. A one-time “exposure baseline” onboarding at $750–1,500 for the historical lookback — this is the part with real inference cost and real perceived value, and it converts far better than a free trial because the report itself is the product.
• ACV: ~$4,200/location/yr subscription; ~$5,000 first-year with the baseline. Multi-location groups at 2–5 locations lift ACV to $8K–20K.
• Why this price holds: the wallet is already open and much wider. Dental billing outsourcing runs $1,000–5,000/mo flat, or 4–9% of collections; eAssist charges roughly 4.99–7.99% of collections plus minimums. A practice paying $2,500/mo to a billing service will not blink at $349/mo for the thing that stops the money from being taken back. I am pricing at roughly one-seventh of an adjacent, well-established line item.
• Math to $1M ARR: 240 locations × $349/mo × 12 = $1.005M. That is 240 owner-dentists out of a serviceable 40,000+.
• Math to $5M ARR: ~1,100 locations, which requires the small-group segment (2–5 locations) to carry maybe a third of revenue and at least one state dental association partnership feeding steady inbound. Achievable, not automatic.
• Expansion path: per-location seats as groups grow → continuous monitoring tier (score notes weekly as they’re written, not just the historical sweep) → appeal-support module that assembles the documentation packet when a demand letter does arrive → and the genuinely interesting one, anonymised benchmarking of exposure patterns by code and region, which nobody else will be positioned to compute.

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

• The free exposure scan, offered to a named list. Open Dental’s public API means onboarding is genuinely low-friction. Offer a no-cost 90-day scan on the top three risk codes. The output is a number in dollars. Practices that see a scary number convert to paid annual at the full lookback. I’d target a 20–25% scan-to-paid rate because the scan does the selling — the practice’s own chart is the sales collateral, which is the strongest asset I could ask for.
• Buy attention where the fear already is. The dentists filing class actions against Delta in four states, and the ones reading Liles Parker and dental defence-firm content, are pre-qualified and self-identified. Sponsor the newsletters and podcasts serving that anxiety (DrBicuspid, dental practice management shows, state dental society bulletins). This is not “content marketing” — it’s buying placement in the exact channel where the audit story is already being told to the exact person who signs the cheque.
• Partner with dental CPAs, practice-transition brokers, and dental attorneys. Two high-intent moments: a practice sale (buyer’s diligence badly wants a documentation exposure number, and today nobody can produce one) and a recoupment letter (the attorney needs the scope of the problem fast). Both are referral relationships, both are 20–40 introductions per partner per year, and the transition-broker channel in particular pays for itself because the report becomes a diligence artifact.
• Billing companies as a channel, not a competitor. eAssist-style billing firms serve thousands of practices and are measured on collections. Recoupment reverses their scorecard. White-label or revenue-share the scan so their account managers can run it across their book — one signed billing partner can put the product in front of several hundred practices without me making a single cold call.
• State dental association CE. Deliver a legitimate continuing-education session on documentation and audit defence. Dentists must complete CE anyway, associations perpetually need speakers, and the session’s natural closing artifact is “here’s how to see your own number.”

10. Build complexity — justification

Medium. Nothing here is research-grade. The read path into Open Dental is a public, self-serve, documented REST API — the single biggest de-risking fact in this proposal, because PMS integration is normally where dental startups die. Scoring is LLM work over short documents against a published checklist of auditor-demanded elements. The UI is a ranked table with drill-down; no real-time anything, no complex state.

The genuine work is threefold: HIPAA posture and BAAs from day one (this is PHI, non-negotiable, and it gates your first customer); building the rubric with a real dental billing/audit expert so the Red flags are actually right; and each additional PMS integration after Open Dental — Dentrix and Eaglesoft are meaningfully less friendly and each is weeks, not days.

Call it 12–16 weeks to a v1 a paying practice can use, for a technical founder plus a part-time domain advisor. Add 4–6 weeks per additional PMS.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Read-only analysis of the practice’s own records under a BAA. Flags and explains; never backdates or rewrites.
Ethical — no harm / dark patterns	✅	Helps clinicians document what they actually did. Hard line against retroactive fabrication — that’s the difference between audit defence and fraud.
Market exists (evidence above)	✅	178,461 practices; audits expanding in 2026; adjacent back-office spend of $1K–5K/mo already proven.
1–5 person team can build this	✅	Technical founder + domain advisor. Public API, off-the-shelf inference.
Launchable with <$50K / ₹40L	✅	Inference, hosting, HIPAA tooling, a compliance review, and advisor fees. Comfortably under.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Retroactive clawback plus possible network termination is genuinely expensive and unbudgeted. Docked because it’s episodic, not daily — many practices have never been audited and will discount the risk right up until the letter lands. That’s a real selling obstacle, not a rounding error.
Demand evidence	15	12/15	Strong indirect evidence: named 2026 audit expansion, class actions in four states, a thriving human chart-audit consulting market, heavy adjacent spend. Docked because I found no direct evidence of practices searching for this specific product — the demand is inferred from the pain, not observed in a purchase.
Build feasibility	15	11/15	Public Open Dental API and off-the-shelf inference carry most of it. Docked for HIPAA/BAA overhead and the fact that Dentrix/Eaglesoft coverage — needed for most of the market — is real integration work.
Distribution clarity	15	11/15	The free-scan-to-dollar-number motion is concrete and the report sells itself. Billing-company and CPA/broker channels are named and reachable. Docked because owner-dentists are famously hard to reach and slow to respond, and no channel here is proven yet.
Revenue mechanics	15	12/15	$349/mo sits far below well-documented adjacent spend; 240 locations to $1M is credible. Docked because retention past the first cleanup is the open question — see risk flags.
Time to first revenue	10	7/10	Paid baseline reports can be sold during the pilot phase, so revenue in roughly 6–10 weeks post-launch. Not faster because BAAs and PMS access add a real onboarding step.
Defensibility	10	6/10	Soft moat that compounds: an accumulating corpus of scored notes mapped to actual audit outcomes, plus PMS integrations and channel relationships. But an ambient-scribe incumbent with existing PMS hooks could add a retrospective mode. Month 3 is thin; month 12 is defensible if the outcome data is being captured.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

You need someone who can ship a HIPAA-compliant integration and someone who knows what a Delta auditor actually rejects. If that isn’t the same person, the second one is a founding advisor with equity, not a contractor. Getting the rubric wrong in either direction kills it: too many false Reds and the tool is noise; too few and it’s negligent.

Key assumptions to validate (3–5)

1. Assumption: An owner-dentist shown a credible dollar exposure figure on their own charts will pay $349/mo. How to test: run 15 free scans, present the number, ask for the card on the spot. Track how many pay before any remediation work is delivered.
2. Assumption: Real-world notes fail auditor criteria often enough to produce an alarming number — if 95% of notes are fine, there’s no product. How to test: score 500 real notes across 5 practices with the domain advisor grading the model’s output. Need a defensible failure rate on high-risk codes, and model-vs-expert agreement above ~85%.
3. Assumption: Practices will grant read access to clinical records for a scan. How to test: count how many of the 15 pilot practices complete the BAA and API connection versus how many stall. This is the most likely silent killer.
4. Assumption: The product retains after the initial cleanup rather than churning. How to test: offer pilots a monitoring tier at month 3 and measure uptake — this is the difference between a $1M and a $5M business.
5. Assumption: A billing company will channel-partner rather than build it themselves. How to test: pitch three mid-size dental billing firms and see whether the conversation goes to partnership or to “interesting, we might do that.”

Risk flags

1. Retention / one-and-done risk. The scariest structural risk. Fix the backlog, cancel the subscription. Mitigation is designing for continuous scoring of new notes from day one and pricing the historical sweep separately — but if practices treat this as a one-time project, ARR becomes lumpy project revenue and the $5M path closes.
2. Incumbent encroachment. Bola has 10,000+ users and existing PMS integrations. Adding “score the notes you already have” is a feature, not a moonshot, for them. The defence is speed, the outcome dataset, and owning the audit-defence positioning before a scribe vendor thinks of it.
3. Liability and evidentiary exposure. A tool that formally scores a practice’s notes as “Red” creates a discoverable record that the practice knew about deficiencies. That cuts both ways in litigation and a plaintiff’s attorney would enjoy it. Needs real legal review of report retention, wording, and customer data ownership before the first customer — not after.
4. Payer-behaviour dependency. The urgency rests substantially on Delta continuing to expand audits. If the class actions chill enforcement, or a settlement constrains audit practice, the fear that drives conversion softens considerably.
5. Model error in both directions. False Reds waste clinician time and destroy trust fast; false Greens give false comfort on real exposure. Confidence bands and a human-review path on the highest-dollar items are mandatory, not polish.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (HIPAA-comfortable) + dental billing/audit
                        expert as founding advisor with equity
Time to revenue:        6–10 weeks post-launch; paid baseline reports during pilot
Capital to launch:      $15–25K (inference, HIPAA hosting, legal review, advisor equity)
Top 3 assumptions to validate first:
  1. Real notes fail auditor criteria at an alarming rate — score 500 real notes
     across 5 practices, expert-graded, need >85% model-expert agreement
  2. Owners pay on seeing their number — 15 free scans, ask for the card at the
     reveal, target 20%+ close before any remediation is delivered
  3. Practices will actually grant record access — measure BAA + API completion
     rate across the pilot cohort
Kill criteria:
  - Abandon if fewer than 3 of 15 pilot practices complete the BAA and connect
    their PMS — access friction beats the value prop and nothing downstream matters
  - Abandon if expert review finds under 10% of high-risk-code notes are genuinely
    deficient — the exposure number won't be scary enough to sell
  - Abandon if pilot practices consistently cancel within 90 days of cleanup and
    reject the monitoring tier — that's a consulting project, not a SaaS business

15. Next step — 1-week validation sprint

• Day 1–2: Recruit a dental billing/audit expert for a paid week of their time. Build the scoring rubric from published auditor demands — medical necessity, perio charting depths, pre/post-op radiographs on endo, radiograph metadata — for the named high-risk codes only: D4341/D4342, D3310–D3330, crowns/onlays/veneers.
• Day 3–4: Get 3–5 practices to share de-identified historical notes on those codes (offer the analysis free, in writing, in exchange). Score 500 notes. The expert independently grades a 100-note sample blind. Measure the failure rate and the model-expert agreement rate.
• Day 5: Sit with 10 owner-dentists, show each their own exposure number in dollars, and ask for $349/mo starting today. Not “would you pay” — ask for the card.

Go if: >10% of high-risk notes are expert-confirmed deficient, model-expert agreement exceeds 85%, and ≥2 of 10 dentists pay on the spot.

No-go if: the deficiency rate is low enough that the dollar figure doesn’t alarm anyone, or every dentist wants to “think about it” — because a problem that survives a week of thinking is a problem they’ll keep living with until the letter arrives, and by then they’re buying a lawyer, not software.