Prompt Chain: Denial Appeal Automation System

What This Builds

A structured, repeatable system for processing denial appeals — using a 3-stage prompt chain that (1) analyzes the denial, (2) selects the right argument strategy from your library of successful appeals, and (3) drafts the final letter. What used to take 90 minutes per appeal now takes 20 minutes of review. Over a year, for a department handling 15–20 appeals per month, this system saves 200+ hours.

Prerequisites

• Claude Pro ({{tool:Claude.price}}/month) — for longer context and Projects
• 10–15 successful past appeal letters (organized by denial type — remove patient PHI before uploading)
• Access to your EOB/denial reports from your EHR or billing system
• Comfort with the basic appeal drafting workflow from Level 3

The Concept

Most denial appeal letters follow the same logic: establish the clinical basis, cite the regulatory authority, connect the two, and demand reconsideration. The differences are in the specific clinical facts and specific regulatory citations. A prompt chain breaks this into distinct steps — so each step produces its best possible output before passing it to the next step.

Think of it like an assembly line: Step 1 takes the raw denial information and converts it into a structured case summary. Step 2 matches that case to your best historical appeal arguments. Step 3 combines them into a polished letter. Each step is simple; the chain produces a result that would be hard to get in one prompt.

Build It Step by Step

Part 1: Organize Your Appeal Letter Library

Before building the chain, you need a library of successful appeals organized by denial type. Create a folder (or a Claude Project's Knowledge section) with at least one winning appeal letter per major denial category:

• Medical necessity denials (inpatient admission)
• DRG coding disputes (payer recoded your DRG)
• CC/MCC documentation disputes (payer rejected a complication code)
• Untimely filing denials
• Duplicate claim denials
• Authorization denials

For each, remove all patient PHI (replace names with [PATIENT], SSNs with [REDACTED], account numbers with [ACCT#]).

Part 2: Set Up the Claude Project

Create a Claude Project called "Denial Appeal System." Upload:

• Your organized appeal letter library (all denial types, de-identified)
• AHIMA compliant query guidelines (if CDI-related denials are common)
• A brief "department cheat sheet" noting: hospital name, key service lines, payer mix, most common denial types

Write Project Instructions:

You are a denial appeal specialist for [Hospital Name] HIM department. You help process claim denial appeals using a structured 3-stage prompt chain. You have access to our library of successful appeal letters organized by denial type. Always reference the most relevant successful letter when drafting new appeals. Never include actual patient PHI in any output.

Part 3: Build the 3-Stage Prompt Chain

Stage 1: Denial Analysis Prompt

Use this every time a new denial comes in. Paste it into your Claude Project and fill in the denial details:

STAGE 1: DENIAL ANALYSIS

Please analyze this denial and produce a structured case summary for appeal drafting:

Payer: [payer name]
Date of service: [date]
Billed DRG/codes: [DRG or code list]
Billed amount: [amount]
Denial reason (exact language from EOB): [paste verbatim]
Denial category: [medical necessity / coding dispute / auth / other]
Clinical summary (1-2 sentences): [brief clinical facts]
Supporting documentation available: [what records you have]

Please produce:
1. One-sentence characterization of the denial argument (what the payer is claiming)
2. One-sentence characterization of our counter-argument (what we will argue)
3. The regulatory framework most likely applicable (ICD-10 guideline section, LCD, CMS rule)
4. Whether this matches any denial type in our appeal library (check uploaded letters)
5. Recommended appeal strategy: [aggressive / standard / documentation request first]

Stage 2: Argument Selection Prompt

After Stage 1, use the output to run Stage 2:

STAGE 2: ARGUMENT SELECTION

Based on the Stage 1 analysis above, identify the most relevant successful appeal from our library and extract the key arguments we should reuse.

From our uploaded appeal letters:
- Which letter most closely matches this denial type and payer?
- What specific regulatory citations did we use successfully?
- What clinical framing was most effective?
- What was the outcome (overturn amount or percentage)?

If no close match exists in our library, identify the regulatory citations and argument structure I should build from scratch.

Stage 3: Letter Drafting Prompt

After Stage 2, draft the final letter:

STAGE 3: DRAFT THE APPEAL LETTER

Using the Stage 1 analysis and Stage 2 argument selection, draft a complete denial appeal letter.

Letter format:
- Date: [today's date]
- Payer address: [I'll add after]
- Re: Appeal of Claim Denial — Account [ACCT#] — DRG/Code [number]
- Body: Opening paragraph (state the denial and our appeal position) → Clinical Basis paragraph → Regulatory Basis paragraph → Conclusion paragraph (specific demand for overturn + reimbursement amount)
- Closing: [My name, credentials], Health Information Manager, [Hospital Name]
- Enclosures: [I'll list after]

Use the argument structure and citations identified in Stage 2. Maintain a professional, assertive tone. Do not include any patient PHI beyond the account number placeholder.

Part 4: Test the Chain with a Real Denial

Take a pending denial and run it through all three stages. Compare the final output to how you'd normally write the appeal. Adjust Stage 1 to capture the most important information for your common denial types.

Real Example: Medical Necessity Denial

Input to Stage 1:

Payer: Blue Cross [State]
Date of service: [date]
Billed DRG: 470 (Major joint replacement — lower extremity)
Denial reason: "Inpatient admission not medically necessary; procedure could have been performed in outpatient setting."
Clinical summary: 72-year-old patient, bilateral knee osteoarthritis, previous failed conservative therapy, BMI 38, moderate sleep apnea.
Documentation available: Pre-op evaluation, surgical notes, anesthesia evaluation, discharge summary.

Stage 2 output (if you have a successful hip replacement appeal in your library): "Best match: DRG 470 medical necessity appeal against [same payer], June 2025. Used Milliman Care Guidelines and CMS Two-Midnight Rule documentation. Argued that comorbidities (BMI, sleep apnea) required inpatient monitoring. Outcome: overturned at 100%."

Stage 3 output: Complete appeal letter, 4 paragraphs, citing Milliman Care Guidelines, the Two-Midnight Rule, and the anesthesia evaluation documenting sleep apnea risk requiring post-op monitoring — framed in the same language that won the prior appeal.

Time: 20 minutes review vs. 90 minutes writing from scratch.

What to Do When It Breaks

• Stage 1 produces an incomplete analysis → Add more specific denial details; the chain produces better output when the Stage 1 input is complete
• Stage 2 finds no relevant appeal in the library → Run Stage 3 without the library reference: "Build the argument from scratch using the regulatory framework identified in Stage 1"
• Stage 3 letter is too generic → Add to Stage 3 prompt: "Be specific about the clinical comorbidities and how each one justifies inpatient-level care per the payer's own medical policy"
• Payer-specific argument isn't working → Add the payer's own medical policy criteria document to your library and reference it in Stage 2

Variations

• Simpler version: Skip Stage 2 (the library matching) and just run Stage 1 → Stage 3. You lose the library leverage but still save 40–50% of writing time.
• Extended version: Add a Stage 4 — after the letter is drafted, ask Claude to "identify the 3 weakest points in this letter and suggest how to strengthen them." This produces a self-reviewed first draft.

What to Do Next

• This week: Run 3 pending appeals through the full chain and compare quality and time to your normal process
• This month: Expand your appeal library to cover your top 5 denial types; track win rates for AI-drafted vs. manually drafted appeals
• Advanced: Connect this system with your Claude Project so the library and department context are always available — eliminating the need to paste context at the start of each chain

Advanced guide for Health Information Manager professionals. Claude Pro required for Projects and document uploads.