LevelBack

1. One-liner

Finds claims your payer quietly paid down a level, then files the batch appeal state law now compels.

2. Trend signal — why now?

Three things happened in the last fourteen months, and together they turn a chronic annoyance into a collectable asset.

Payers industrialised automatic downcoding. Cigna’s Reimbursement Policy R49 took effect 1 October 2025 and, in Cigna’s own words, “may adjust the E/M CPT code 99204-99205, 99214-99215, 99244-99245 to a single level lower when the encounter criteria on the claim does not support the higher-level E/M CPT code reported.” Aetna ran a comparable program from spring 2025. These are algorithmic, applied pre-payment, and applied without anyone opening a chart.

Regulators started punishing it — and documenting the mechanic for me. On 13 March 2026 the Maryland Insurance Administration issued a consent order fining Cigna $80,000, ordering it to stop automatic downcoding, and requiring it to reprocess all impacted claims back to 1 October 2025. On 7 April 2026 the MIA followed with Bulletin 26-9 extending the same prohibition to all health insurers and third-party payors operating in Maryland. That is a regulator publishing, in primary source form, that the practice is unlawful and that retroactive recovery is the remedy.

States wrote the appeal right into statute. Indiana’s HEA 1271 took effect 1 July 2026: downcoding “cannot be based solely on an automated process,” a human reviewer “must examine the relevant medical records,” the insurer must provide “a clear explanation of the decision, including the clinical rationale and any coding changes,” providers get “at least 180 days to appeal,” and — the part that changes the unit economics — providers may submit batch appeals involving substantially similar claims. Arkansas and Virginia enacted downcoding laws in 2025. Georgia SB 444 lands 1 January 2027; Illinois SB 3114 (Transparency in Downcoding Act) lands 1 January 2028.

The economics are already known and already bad. The AMA’s own resource states the quiet part: “certain payers even pay at the adjusted lower rate without changing the billed code, making it extremely difficult to identify in coding audits.” Systematic downcoding costs the average multi-specialty practice $40,000–$180,000 annually. Roughly 68% of downcoded claims are overturned on appeal. And yet nearly 60% of denied claims are never resubmitted at all — because, as one practice calculated, it would have to hire an additional full-time employee just to review and appeal automatic downcoding.

A 68% win rate that nobody collects, against a defendant a regulator just fined, under a statute that now permits batch filing. That is money on the table.

Provenance:

• Signal 1 (demand): AMA E/M downcoding resource — payers “pay at the adjusted lower rate without changing the billed code, making it extremely difficult to identify in coding audits”; lists CO150 / M85 / N610 / CARC 186 as the remittance fingerprints — https://www.ama-assn.org/system/files/payer-em-downcoding-resource.pdf — observed 2026-08-31
• Signal 2 (economic): Maryland Insurance Administration consent order fines Cigna $80,000, orders halt to automatic downcoding and reprocessing of all impacted claims back to 1 Oct 2025; Bulletin 26-9 (7 Apr 2026) extends to all payors — https://adanews.ada.org/ada-news/2026/april/maryland-orders-cigna-to-stop-automatic-downcoding-issues-80000-fine/ — observed 2026-08-31
• Signal 3 (feasibility/regulatory): Indiana HEA 1271 effective 1 July 2026 — human review required, written clinical rationale required, 180-day appeal window, batch appeals for substantially similar claims — https://indental.org/advocacy/beneficial-dental-legislation/ — observed 2026-08-31
• Signal 4 (economic): Systematic downcoding costs the average multi-specialty practice $40,000–$180,000 annually; ~68% of downcoded claims overturned on appeal; structured-workflow practices recover 61–74% of suppressed revenue within 90 days — https://www.medicalbillersandcoders.com/blog/payer-downcoding-denials/ — observed 2026-08-31
• Signal 5 (workaround cost): Cigna requires “appeals and medical records via fax”; Cigna’s bypass program requires proving 80% of a set of charts were billed correctly — https://compasshealthcareconsulting.com/automated-downcoding-how-to-fight-back/ — observed 2026-08-31 Category: Regulatory arbitrage

3. The opportunity

The gap is not detection-in-general. It is that the money is invisible in exactly the place every incumbent looks.

Denial-management software keys off denials. Silent downcoding is not a denial. The claim is accepted, the billed code is often left untouched, and the payment simply arrives lower. As the AMA puts it, it “will not show in your denial rate. It shows up as a net collection rate drifting downward quarter over quarter.” Every RCM dashboard in the building reports a healthy clean-claim rate while the practice bleeds.

Look at what the market actually sells:

• Lassie ($3,000 per-location setup, usage-based annual billing) automates EOB posting into Dentrix/Eaglesoft/Open Dental and does catch “payer substitutions, where a covered code is replaced with a lower-reimbursing one.” That is the visible case — code swapped. It does not address the AMA’s documented case where the code is unchanged and only the dollars move.
• Zentist, Medusind, Dentalrobot, Dental Claim Support — EOB retrieval, posting, clean-claim rates, outsourced billing. None address silent downcoding. None generate appeal packets.
• MD Clarity does contract-variance underpayment work — for hospitals. Enterprise motion, enterprise price, wrong segment.
• The AMA itself ships the current best-practice remedy: a sample appeal letter template you fill in by hand, per claim.

So the state of the art for an 8-provider group is: a human runs a submitted-vs-paid variance report nobody has built, notices a pattern, and then hand-writes letters — to Cigna, by fax.

The 10× is not “AI reads your EOBs.” It is this: the per-claim recovery is $30–$60, the per-claim manual appeal cost exceeds it, so rational practices skip it — and payers price that in. Indiana’s batch-appeal provision collapses the denominator. One filing, one clinical-rationale argument, fifty substantially similar claims. The statute converts a pile of individually-uneconomic claims into a single economic one, and it hands you the argument for free: your own automated process is unlawful, and you did not give me the written clinical rationale the statute requires.

That last part is the wedge nobody has picked up. The practice does not have to win a coding argument. It has to point out a procedural defect in the payer’s process — no human review, no written rationale — which is a document-shaped question, not a clinical one. That is exactly the kind of argument software can make at scale and a busy biller cannot.

4. Target market

Primary customer: Practice administrator or owner-physician at an independent medical or dental practice with 2–15 providers, in Indiana, Arkansas, Virginia or Maryland to start. These are groups big enough to bill 400–2,500 E/M claims a month and small enough that there is no dedicated denials analyst — billing is one or two people, or an outsourced biller paid a percentage.

Why they buy, in their words: A practice owner quoted in coverage of these programs called the policy “guilty until proven innocent.” Another practice calculated it would have to hire an additional full-time employee to review and appeal automatic downcoding. Physicians in the Dallas area reported that Aetna had been “automatically downcoding their patient interactions without notice or explanation.” The AMA’s framing is that “appeal is typically the only recourse once a payer has downcoded a claim,” while noting practices must “review payment details, comparing the payment for each billed CPT code to the anticipated amount” — a per-line-item reconciliation nobody has time to do.

Rough TAM reasoning: I am not going to pretend to a precise count. The realistic shape: independent physician practices in the 2–15 provider band number in the tens of thousands nationally, and the four live-statute states plus Maryland’s bulletin cover a meaningful slice. Add dental — the ADA reports 30 new laws in 16 states in the 2026 session alone, with Indiana’s downcoding provision applying to dental plans. At $349/mo average and 1,200 customers that is a $5M business, and I do not need a national footprint to get there. Georgia (Jan 2027) and Illinois (Jan 2028) are pre-scheduled TAM expansions I can see on a calendar.

Why now for them: Before 1 July 2026 an Indiana practice that appealed a downcode was arguing coding opinion against a payer’s algorithm. After 1 July 2026 it is arguing statutory non-compliance, with a 180-day window and a batch mechanism. The argument got easier and the filing got cheaper in the same month. Separately, Maryland just proved the retroactive-reprocessing remedy is real.

5. Product sketch (MVP)

• Silent-downcode finder. Ingest 835 remittance files and compare, line by line, submitted CPT against paid allowed-amount versus the practice’s contracted fee schedule — catching both the code-swapped case and the AMA’s harder “payment adjusted, code unchanged” case. Flag CO150, M85, N610 and CARC 186 as corroborating fingerprints, but do not depend on them.
• Pattern view by payer and provider. Twelve-month rollup showing which payer downcodes which provider at what rate — the evidence that distinguishes sporadic edits from a systematic program, and the thing you need to argue you have been placed in a global prepayment review.
• Batch builder. Cluster downcoded claims into “substantially similar” groups (same payer, same code pair, same clinical pattern) so one filing carries fifty claims under Indiana’s batch-appeal provision.
• Statutory appeal packet. Generates the filing citing the applicable state provision — automated-process prohibition, absence of human review, missing written clinical rationale — with the AMA MDM criteria mapping, the remittance evidence, and the highlighted chart excerpts attached.
• Rationale-gap detector. Tracks which downcodes arrived without the statutorily required written explanation. This is the strongest and most mechanical argument in the packet, and it needs no clinical judgment.
• 180-day clock. Per-claim countdown to the appeal deadline, with the batch ready to file before it expires.
• Recovery ledger. Filed, overturned, paid, still open — dollars recovered against subscription cost, which is the renewal conversation.

6. AI angle — what’s load-bearing

Remove the AI and two things break.

Clustering “substantially similar.” The statute permits batch appeals for substantially similar claims but does not define the boundary. Grouping fifty downcoded 99214→99213 claims into batches that share a defensible clinical pattern — same problem complexity, same data-reviewed profile, same risk tier — is a judgment call over unstructured chart text, made hundreds of times a month. That is a language model job. Get the grouping wrong and the payer splits the batch and the economics collapse back to per-claim.

Mapping the record to MDM criteria. A winning appeal, per the research, contains “explicit mapping to AMA 2021 MDM criteria.” That means reading the encounter note and asserting which of the three MDM elements — problem complexity, data reviewed, risk — were met at the billed level, then pointing at the exact sentences that prove it. Doing that by hand is the FTE the practice refuses to hire. Doing it with a model over the note text is the whole product.

The variance detection itself is arithmetic, not AI, and I want to be honest about that. The AI is what converts a list of variances into a filing someone will actually send.

7. Localization angle

N/A — this is a US-only play, and deliberately so. The product’s value comes entirely from specific state statutes (Indiana HEA 1271, Arkansas, Virginia, Georgia SB 444, Illinois SB 3114) and US claims infrastructure (X12 835 remittance, CPT/CDT code sets, CARC/RARC remark codes). There is no version of this that ports. The right expansion is not geographic but statutory — follow the states as their laws commence.

One structural caveat worth naming here: ERISA preempts state insurance law for self-funded employer plans, which is why the ADA is pushing the federal IDA Act. A meaningful share of a practice’s book will sit outside the statutory argument. Those claims still get the variance detection and the MDM-mapped appeal; they just do not get the procedural-defect argument. The product has to be honest about which claims are which.

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

Pricing: Tiered by provider count, flat monthly, no contingency.

• Solo / 2 providers — $199/mo
• 3–7 providers — $399/mo
• 8–15 providers — $799/mo

Flat subscription rather than a percentage of recovery is a deliberate choice. Contingency pricing invites comparison to outsourced RCM firms and drags me into a services sale; it also caps me at the recovery rate. Flat pricing at $399 against a documented $40K–$180K annual bleed is an easy arithmetic conversation, and Claimable already anchors this category at $79–$239 per user per month, so $399 per practice reads as cheap.

ACV: ~$4,200 blended.

Rough math to $1M ARR: 240 practices × $349/mo × 12 = $1.0M. That is roughly 60 practices per state across four live-statute states — a number I can reach with outbound alone.

Rough math to $5M ARR: ~1,200 practices at the same blended ACV. Requires the Georgia (2027) and Illinois (2028) commencements landing on schedule, plus dental as a parallel segment, plus a modest move upmarket into 16–40 provider groups at $1,500/mo. If the state pipeline stalls, this is a $1.5–2M business, not a $5M one — and I would still take it.

Expansion path: Provider-count tiers grow with the practice. Adjacent modules that reuse the same 835 ingestion: contract-rate variance (paid below contracted rate for reasons other than downcoding), and recoupment-window policing — Indiana cut recoupment to 180 days and Connecticut to 12 months, so a payer clawing back outside the window is another mechanical, document-shaped argument off the same data.

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

• Free downcode audit as the entire top of funnel. The practice sends one month of 835 files; I return a one-page number: “Cigna paid you $4,180 below billed level on 63 claims last month; 41 of them are batchable and 38 are still inside the 180-day window.” That report is the demo, the pitch, and the close. It costs me an API call and it is falsifiable on the spot — the administrator can check three claims by hand. Nothing about downcoding is persuasive in the abstract and everything about it is persuasive as a dollar figure with the practice’s own name on it.
• State dental and medical association channels, starting where the law just changed. The Indiana Dental Association publicised HEA 1271 to its own members; ISMA tracks legislative wins for Indiana physicians. These bodies have just told their members they have a new right and have given them no tool to exercise it. Component-society newsletters, district meetings, and a co-branded “how to use your new HEA 1271 appeal right” webinar. This is the highest-intent audience that will ever exist for this product, and the associations want the content.
• Cold outbound to the exact affected specialties. Downcoding concentrates in high-level E/M — 99204/99205/99214/99215 and modifier-25 claims — which means internal medicine, family medicine, cardiology, rheumatology and oncology practices treating complex panels. State medical board licensee files and NPPES give me name, specialty, address and group size. Filter to 2–15 provider groups in IN/AR/VA/MD, send the administrator a short personalised note offering the free audit. Target 2,000 practices, 8% take the free audit, 20% of those convert = 32 customers from one sweep.
• Ride the enforcement news. Every new consent order, bulletin or state commencement is a dated, citable event with a named payer. Maryland Bulletin 26-9 and the Cigna order are the template. Write the practical explainer within 48 hours of each — “what Bulletin 26-9 means for your Cigna remittances, and how to pull the claims it covers” — and put it in front of the affected state’s practices. Regulatory-arbitrage products get their distribution from the regulator’s own calendar.
• Outsourced billers as a channel, not a competitor. Billers paid a percentage of collections have their incentives aligned with recovering this money and no tooling to do it. Offer a multi-practice view at a per-practice rate; one biller signing brings 10–30 practices.

10. Build complexity — justification

Medium. The 835 remittance format is a documented, stable X12 standard and parsing it is well-trodden — that part is off-the-shelf. The real work is threefold: getting remittance and fee-schedule data out of practice management systems (Dentrix, Eaglesoft, Open Dental on the dental side; Athena, eClinicalWorks, Epic-lite on the medical side) where integration quality varies from decent API to clearinghouse file drop; encoding the state-by-state statutory argument logic, which is genuine domain work and needs a healthcare-attorney review before a single packet goes out; and the chart-to-MDM mapping. HIPAA obligations mean BAAs, encryption and access controls from day one — not optional, and it is the main reason this is not a 6-week build.

Call it 14–18 weeks to a v1 that works on file-drop ingestion for two states, with PMS integrations added per-customer afterwards. A technical founder plus a part-time RCM domain expert. Deliberately start with the clearinghouse/file-drop path rather than blocking on integrations.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Helping providers exercise a statutory appeal right. No licence required to prepare appeals on a practice’s behalf; the practice remains the filer. HIPAA BAA required.
Ethical — no harm / dark patterns	✅	Recovers payment for care actually delivered and documented. The product argues the level billed was correct, not that a higher level should be billed — no upcoding nudge, and that line must be enforced in the product.
Market exists (evidence above)	✅	Regulator consent order, four state statutes, AMA resource, quantified $40K–$180K annual bleed, five named vendors adjacent but not on it.
1–5 person team can build this	✅	Two people plus domain advisor and a one-off legal review.
Launchable with <$50K / ₹40L	✅	~$25–35K: build, HIPAA-grade hosting, legal review of packet templates, first outbound sweep.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	$40K–$180K annually per multi-specialty practice, felt every remittance cycle, and currently invisible — practices are losing money they cannot see. Docked 3 because it is a slow bleed, not a hard stop: nobody’s licence lapses and no penalty meter runs. Deferred-ish pain, but the dollar figure is large and recurring enough to carry it.
Demand evidence	15	13/15	Multiple independent hard signals: a regulator’s $80K consent order with retroactive reprocessing, four enacted state statutes, the AMA publishing a countermeasure resource, physicians on record calling it “guilty until proven innocent,” a practice costing the fix at one FTE. Docked 2 because I have no direct evidence of practices paying for this specific product — the 68% overturn and $40K–$180K figures come from RCM-vendor content, which has an interest in the number being large.
Build feasibility	15	11/15	835 parsing is standard; PMS integration variance and HIPAA controls push this past a quick build. 14–18 weeks, two people. Not a 6-week solo ship.
Distribution clarity	15	12/15	The free downcode audit is a strong, falsifiable, self-demonstrating wedge, and state associations have just told members about a right they cannot exercise. Docked because the named-list arithmetic (2,000 → 8% → 20%) is my estimate, not a tested conversion.
Revenue mechanics	15	12/15	$399/mo against a $40K+ bleed is easy math, category anchored by Claimable at $79–$239/user/mo, 240 practices to $1M. Docked 3 because the $5M case leans on Georgia 2027 and Illinois 2028 commencing on schedule — a dependency I do not control, and my own catalogue notes that vendor-quoted commencement dates are often wrong.
Time to first revenue	10	7/10	The free audit shortens the cycle materially — the prospect sees their own dollar figure in week one. But healthcare buyers move slowly, BAAs need signing, and getting the first remittance file out of a practice is itself a small sale. 8–10 weeks to first paying customer is realistic; 4 is not.
Defensibility	10	5/10	Honest score. The variance arithmetic is copyable in a weekend. Lassie already does the easy half and could extend. The real moats accrue slowly: a library of which arguments beat which payer, per state, and a corpus of overturn outcomes that tells me which batches are worth filing. That is a month-12 asset, not a month-3 one. Execution-only moat for the first year.
Total	100	77/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

You need someone who can parse X12 and someone who has actually worked denials. Without the second person this ships a variance report nobody files.

Key assumptions to validate (3–5)

1. Assumption: Silent downcoding — payment reduced with the billed code unchanged — is detectable at meaningful volume in real 835 files from small practices, not just in AMA prose. How to test: Get 3 practices in IN/VA to share 3 months of 835s under NDA. Compute submitted-vs-paid variance against their fee schedules. Need ≥$1,500/mo of recoverable downcoding at a 5-provider practice for the price to work.
2. Assumption: Payers actually honour Indiana’s batch-appeal provision rather than splitting batches into individual claims. How to test: File 3 real batch appeals through a friendly practice in Q4 2026 and see what comes back. This is the single most load-bearing assumption in the model — if batches get split, per-claim economics return and the product is much weaker.
3. Assumption: The procedural argument (no human review, no written rationale) wins more often and more cheaply than the clinical-coding argument. How to test: Run both argument types across matched batches; compare overturn rate and days-to-resolution.
4. Assumption: Practice administrators will hand over remittance files to a new vendor before paying anything. How to test: Offer the free audit to 100 cold-outbound practices; measure the file-share rate. Below 5% and the wedge is broken regardless of everything else.
5. Assumption: Enough of a typical practice’s book is non-ERISA for the statutory argument to matter. How to test: Segment the pilot practices’ claims by plan funding type and measure what share the state law can actually reach.

Risk flags

1. Regulatory dependency, both directions. The product’s edge comes from statutes. Georgia (2027) and Illinois (2028) could slip, be amended, or — as I have seen elsewhere — turn out to have borrowed a commencement date from a neighbouring bill. Verify each against the enrolled act, never a vendor blog. Conversely, if payers respond to enforcement by adding a perfunctory “human review” rubber stamp, the strongest procedural argument weakens.
2. ERISA preemption caps the addressable claim pool. Self-funded employer plans sit outside state insurance law. This is a known structural ceiling, not a surprise, and the ADA is lobbying federally precisely because of it. It shrinks the statutory argument, not the variance detection.
3. Incumbent extension. Lassie already detects code substitution and sits inside the PMS with EOB data in hand. Extending to payment-only variance and appeal packets is a quarter of work for them, not a year. Speed and state-law depth are the only defence in year one.
4. Payer retaliation as a sales objection. Some administrators will fear that aggressive appealing invites prepayment review. The AMA notes payers do place physicians in global prepayment review programs. Whether or not the fear is well-founded, it will come up in every third sales call and needs an honest answer.
5. Data access friction. If practices cannot easily export 835s and the PMS vendors are unhelpful, onboarding cost per customer rises and the model degrades toward services.

14. Structured verdict

Score:                  77/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder who can parse X12 835, paired with an
                        RCM/denials domain expert; healthcare attorney on retainer
                        for packet templates
Time to revenue:        8–10 weeks from launch
Capital to launch:      $25–35K
Top 3 assumptions to validate first:
  1. Silent downcoding is detectable at ≥$1,500/mo per 5-provider practice —
     analyse 3 months of real 835s from 3 pilot practices
  2. Payers honour batch appeals rather than splitting them — file 3 real
     batch appeals under Indiana HEA 1271 in Q4 2026
  3. Cold practices will share remittance files for a free audit — measure
     file-share rate across 100 outbound targets
Kill criteria:
  - Abandon if median recoverable downcoding at a 5-provider practice is
    under $800/mo (the $399 price cannot be justified)
  - Abandon if payers split batch appeals into individual claims in 2 of 3
    Indiana test filings (per-claim economics return, wedge gone)
  - Abandon if fewer than 5% of 100 cold-outbound practices will share
    remittance files for a free audit
  - Abandon if Lassie or an equivalent PMS-resident vendor ships
    payment-variance detection plus appeal-packet generation before v1

15. Next step — 1-week validation sprint

• Day 1–2: Pull the actual statutory text — Indiana HEA 1271 as enrolled, Arkansas and Virginia’s 2025 downcoding acts, Georgia SB 444, Illinois SB 3114 — and confirm commencement dates against the enrolled acts rather than summaries. Specifically confirm the batch-appeal provision’s exact wording and any definition of “substantially similar.” If the batch mechanism is narrower than the ADA summary implies, the economics change and I need to know on day two.
• Day 3–4: Recruit 3 practices (2 medical, 1 dental) in Indiana or Virginia via a state association contact. Get 3 months of 835 files under NDA. Build the submitted-vs-paid variance analysis by hand in a spreadsheet — no product, no code beyond a parser. Produce the one-page dollar figure for each.
• Day 5: Sit with each administrator, show the number, and ask one question: “If I filed these as a batch for $399 a month, would you sign today?” Then ask their biller whether they knew the money was missing.

Falsifiable outcome: Go only if (a) median recoverable downcoding across the three practices exceeds $1,500/mo per 5 providers, (b) at least 2 of 3 administrators say yes to the price, and (c) the enrolled Indiana text genuinely permits batch appeals without a claim-count cap that kills the arithmetic. Any one of those failing sends this back to VALIDATE.