AI Insurance Claim Denial Appeal Writer

A denial letter from a health insurer reads like it was designed to be ignored: a form paragraph, a cryptic reason code, a PO box, and a deadline buried in the…

The Problem

A denial letter from a health insurer reads like it was designed to be ignored: a form paragraph, a cryptic reason code, a PO box, and a deadline buried in the fourth paragraph. Most patients open it, feel a wave of dread, and put it in a drawer. That reaction is rational — appealing a denial means decoding a Claim Adjustment Reason Code, tracking down the plan's coverage policy, getting a doctor to restate medical necessity in insurer-speak, and mailing or faxing it all before a deadline that is rarely more than 60 days out. Most people have never seen a CARC code before the day they get one.

The scale of the problem is bigger than any one bad-luck story. In 2025, 41% of healthcare providers say more than 10% of their claims get denied, up from 38% in 2024 and 30% in 2022 — the trend line is getting worse, not better. On ACA Marketplace plans the average in-network denial rate is 19%, out-of-network climbs to 37%, and some insurers deny more than half of what crosses their desk. Even Medicare Advantage initially denies about one in six claims. Lower-income patients face 43% higher odds of a denial than higher-income patients on the same kind of claim — the people least equipped to fight back get hit hardest.

Here is the part that makes this a genuine market failure rather than just bad luck: appeals work. Roughly 55% of properly filed appeals succeed, and in New York the share of denials overturned after a patient or physician appealed rose from 38% in 2019 to nearly 53% in 2025. Yet less than 1% of denied claims are ever appealed. The system isn't broken because appeals fail — it's broken because almost nobody files one. That gap between "would probably win" and "never even tried" is a paperwork-and-confidence problem, exactly what a well-designed AI product exists to close. Small medical and dental practices feel it from the other side too: front-desk staff spend hours per denial hand-writing appeals for patients who can't do it themselves, work that never shows up on an invoice.

The Solution

A focused web app that takes a photo or PDF of a denial letter, EOB, and optional chart notes, and turns them into a submission-ready appeal letter in minutes instead of hours. The user doesn't need to know what a CARC 50 is — the product reads the denial reason code, pulls the plan's own coverage-policy language where available, drafts a structured clinical-and-administrative argument, and cites the specific denial reason back at the insurer in its own vocabulary. Every appeal gets a deadline pulled from the letter and tracked on a dashboard so a reminder fires before the window closes — the biggest reason winnable appeals never get filed.

The product serves two audiences from one core engine: individual patients appealing their own denial, and small medical or dental practices that bill and appeal on behalf of many patients and need a queue, not a one-off letter. A practice gets a shared dashboard of every open appeal sorted by deadline; an individual gets a simple three-step wizard.

How it works:

Upload the denial — Patient or staff uploads or photographs the denial letter and EOB; OCR and an LLM extract the CARC/RARC reason code, denied amount, plan name, and appeal deadline automatically
AI drafts the appeal — The system pulls the relevant coverage-policy language for that denial reason, asks two or three clarifying questions (diagnosis, prior treatments tried, physician notes), and generates a formal, insurer-ready appeal letter citing the exact denial code and policy section
Track and remind — Every appeal gets a deadline on a dashboard; the user gets an email/SMS nudge before it lapses, and can log the insurer's response (upheld, overturned, more info requested) to keep the case moving through a second-level or external review if needed

Market Research

The category exists because two large, well-documented trends collide: denials are rising, and the software built to manage them is still mostly enterprise-facing (hospitals and billing companies), not the patient or the ten-person practice.

Denials are structurally rising, not a blip. 41% of providers report over-10% denial rates in 2025, up from 38% in 2024 and 30% in 2022 (Experian Health, State of Claims 2025).
ACA Marketplace denial rates run 19% in-network / 37% out-of-network, with some insurers denying more than half of submitted claims (MoneyGeek 2026 analysis of CMS transparency data).
Appeals win far more than they're tried. About 55% of properly filed appeals succeed and Medicare Advantage appeals overturn 57% of initial denials, yet under 1% of patients ever file one (Counterforce Health denial statistics; Yahoo Finance analysis of the "1% who fight").
The denials-management software market alone grows from $1.49B in 2026 to $4.46B by 2034, and AI-based denial-management modules are expanding at an 18.21% CAGR — the fastest-growing slice of a market Experian, Waystar, and Infinx already serve at the hospital level (Fortune Business Insights; Towards Healthcare AI-Powered Claim Denial Management sizing).
Lower-income patients face 43% higher odds of denial, and NY regulatory data shows overturn rates climbing from 38% (2019) to nearly 53% (2025) as documentation quality improves (Experian Health; Healthcare Dive coverage of the JAMA-published NY DFS study).

The enterprise players (Experian Health, Waystar, Infinx) sell denial-workflow software to hospital revenue-cycle teams for five- and six-figure annual contracts — nobody in that tier builds a $15-a-month product for a patient at their kitchen table or a two-dentist practice with no billing department. That gap, between "hospital-grade RCM suite" and "empty text box in a Word doc," is where this product sits.

Competitive Landscape

A handful of nonprofit and early-stage tools have proven patients will use AI for this — none has built the deadline-tracking, multi-appeal, practice-facing layer that turns a single free letter into a retained subscription.

Claimable — Consumer app focused mostly on medication and prior-authorization denials; drafts and helps submit the appeal for the patient. Reports roughly an 80% success rate. Priced around $50 per appeal, one-off, no ongoing deadline tracking across multiple denials.
Counterforce Health — Free nonprofit tool, grant- and donation-funded, generates a customized appeal letter from the denial and the plan's own coverage policy. Reports around a 70% reversal rate. Strong proof of the AI concept, but no dedicated support, no practice dashboard, and funding tied to continued grants rather than revenue.
Fight Health Insurance — Free, built solo by ex-tech-worker Holden Karau; generates several letter variants from an uploaded denial and translates policy language into plain English. No paid tier yet, and no multi-patient or deadline-tracking layer for practices.
EZAppeal — Paid, per-letter tool: first appeal free, then $3 per generated appeal or prior-authorization letter, marketed as HIPAA-compliant. Closest in spirit to this idea, but packaged as a one-off generator rather than a subscription with a deadline-tracking pipeline.
Traditional medical billing advocates — Human advocates (e.g., Patient Advocate Foundation's free nonprofit case management; private firms like Haven Healthcare Advocates) charge hourly private-pay rates or a contingency cut of what they recover. Effective but slow, and priced out of reach for a routine $200 denied claim.

Your Opportunity Every existing tool optimizes for a single letter, once. None of them are built to be the system of record for a person or a small practice that will see denials repeatedly — a chronic-illness patient with quarterly infusion denials, or a dental office with a dozen open appeals at any given time. Win by owning the parts none of them do well: automatic deadline tracking with reminders so appeals never lapse, a status pipeline through second-level and external review (not just "letter sent"), and a practice tier priced for the two-to-ten-provider office that today does this by hand in a shared inbox.

Business Model

Freemium subscription with a practice tier that carries the real margin. The free tier proves the core loop (upload denial, get a draft) and converts on the second denial, when the deadline-tracking dashboard becomes the reason to stay rather than start over with a free nonprofit tool.

Free ($0) — 1 appeal letter draft per month, manual deadline entry, watermark on exported PDF
Individual ($15/mo or $120/yr) — Unlimited appeal drafts, automatic deadline extraction and SMS/email reminders, appeal-status tracking through second-level review, policy-language citation lookup
Practice ($99/mo, up to 5 staff seats) — Everything in Individual across a shared patient roster, bulk denial upload, CARC/RARC code library with practice-specific templates, exportable appeal log for compliance records

Unit Economics

~$0.40–$0.90 — AI + OCR cost per appeal (document parsing, drafting, and one revision pass)
~82% — Blended gross margin at Individual/Practice pricing
$25–$45 — Target CAC (SEO on denial-code long-tail queries, patient-advocacy communities, dental/medical practice-management forums)
~$140 — Estimated 12-month LTV per paying Individual subscriber, driven by chronic-condition patients who file three to five appeals a year

MRR path: 300 Individual + 20 Practice subscribers clears roughly $6.5K/mo; 1,200 Individual + 100 Practice clears about $27.9K/mo. Practice accounts anchor retention because they're billing infrastructure, not a one-time crisis tool — once a front desk builds the habit of uploading every denial, switching cost rises fast.

Recommended Tech Stack

The hard engineering problem here is not the letter generation — it's reliably extracting structured data (CARC code, denied amount, deadline, plan name) from messy scanned PDFs and handwritten fax cover sheets, and doing all of it in a stack that can sign a BAA and treat every uploaded document as PHI from the first byte.

Next.js 14 (App Router) + Vercel — Patient/practice dashboard and upload flow; Vercel's HIPAA-eligible plan and BAA support cover hosting.
Supabase (Postgres + Auth + Storage) — Tables for users, practices, denials (carc_code, rarc_code, denied_amount, deadline_at, status), appeals (letter_body, submitted_at, outcome), documents (encrypted storage references). Row-level security scoped per user/practice; Supabase supports a signed BAA on paid tiers.
AWS Textract or Google Document AI — OCR layer for scanned denial letters and EOBs; both offer HIPAA-eligible/BAA-covered configurations with structured key-value extraction built for form-heavy insurance documents.
Claude (Sonnet) via the Anthropic API — Two-stage prompting: extract structured fields (code, plan, deadline, denied service) as strict JSON, then draft the formal appeal letter citing that code and any retrieved policy language.
Resend + Twilio — Deadline reminder emails and SMS nudges (T-14, T-3, T-1 days) — the feature that actually moves the "under 1% of people appeal" number.
Stripe Billing — Free/Individual/Practice tiers plus annual prepay; Customer Portal handles practice seat management for the multi-staff tier.

AI Prompts to Build This

Copy and paste these into Claude, Cursor, or your favorite AI tool.

1. Project Setup

Create a Next.js 14 (App Router, TypeScript, Tailwind) app called "AppealDraft" for generating health insurance claim appeal letters. Provision Supabase with these tables: users (id, email, role TEXT CHECK role IN ('patient','practice_staff'), practice_id NULLABLE, plan TEXT default 'free'), practices (id, name, stripe_customer_id, seat_count), denials (id, user_id, practice_id NULLABLE, carc_code, rarc_code, denied_amount_cents INT, plan_name, service_description, deadline_at DATE, status TEXT default 'new', raw_document_url), appeals (id, denial_id, letter_body, generated_at TIMESTAMPTZ, submitted_at TIMESTAMPTZ, outcome TEXT CHECK outcome IN ('pending','overturned','upheld','partial')), reminders (id, denial_id, send_at TIMESTAMPTZ, channel, sent BOOLEAN default false). Enable row-level security so patients only see their own denials and practice staff only see denials scoped to their practice_id. Wire Stripe with Free, Individual ($15/mo), and Practice ($99/mo, 5 seats) products. Add env vars for ANTHROPIC_API_KEY, AWS Textract (or Google Document AI) credentials, RESEND_API_KEY, and TWILIO credentials.

2. Denial Parsing + Appeal Letter Generation

Build the core pipeline. Step 1: POST /api/denials/upload accepts a PDF or image of a denial letter/EOB, sends it to AWS Textract (or Google Document AI) for raw text and key-value extraction, then passes that text to Claude with a strict JSON schema prompt: "Extract from this insurance denial letter: carc_code, rarc_code, plan_name, denied_amount, service_description, appeal_deadline (ISO date), and appeal_level ('first'|'second'|'external'). If a field is not present, return null. Never guess a deadline — only extract dates explicitly stated as an appeal or reconsideration deadline." Store the result on the denials row and schedule reminders at deadline_at minus 14, 3, and 1 days.
 
Step 2: build POST /api/appeals/generate. Given a denial_id, gather the denial's extracted fields plus any user-provided clinical notes and prior-treatment history from a short intake form. Prompt Claude: "You are drafting a formal health insurance claim appeal letter. Cite the specific CARC/RARC code and its stated reason. Reference the patient's clinical necessity using only the notes provided. Use a formal, insurer-ready business-letter format with patient identification, claim number, date of service, and a clear, explicit request for reconsideration by the deadline. Do not fabricate clinical facts not present in the provided notes — flag any gaps as a comment to the user instead." Return the letter as plain text plus a rendered PDF, and let the user edit before export.

3. Deadline Dashboard + Status Tracking

Build a dashboard at /app/denials showing every open denial sorted by deadline_at ascending, with a color-coded urgency badge (red under 7 days, amber under 21 days, green otherwise). Each row expands to show the extracted CARC/RARC code and plain-English meaning, denied amount, current status (new, drafted, submitted, pending, overturned, upheld), and a one-click "Log insurer response" action that updates the appeals.outcome field and, if upheld, offers to generate a second-level or external-review appeal referencing the first appeal's arguments. Build a Vercel Cron job that runs daily, finds reminders where send_at is today and sent = false, and dispatches via Resend (email) or Twilio (SMS) with a short deadline nudge naming the denied service and days remaining. Mark reminders sent = true after dispatch. For practice accounts, add a roster view filtering denials by patient name across the whole practice.

4. Landing Page

Design a single-page marketing site for AppealDraft. Hero headline: "Insurance denies you. We help you win the appeal." Sub-headline: "Upload your denial letter, get a formal appeal citing the exact reason code, and never miss a deadline again." Sections: a before/after showing a raw denial letter next to a generated formal appeal; a stats strip citing the 55% appeal success rate against the under-1% of people who actually appeal; how-it-works (3 steps: upload, draft, track); pricing (Free / Individual $15 / Practice $99) with a callout that billing advocates charge hourly or a contingency cut of recovered amounts; an FAQ covering data privacy, whether this replaces a lawyer for large denials, and how deadline reminders work. Use a calm, clinical-but-warm palette (deep navy, single lime-green accent), Geist font, and a primary CTA: "Upload your denial letter — free."

Sources

Verify current competitor pricing on live product pages before citing in investor materials — appeal-tool packaging shifts as nonprofit players (Counterforce Health, Fight Health Insurance) test paid add-ons.

Want me to build this for you?

Book a consult and let's turn this idea into your MVP.

Book a Consult

See all startup ideas