Draft prior authorization letters for insurance carriers
Medical practices reclaim dozens of staff hours a week by automatically generating payer-specific prior authorization letters from patient records. This gives MSPs a high-impact, HIPAA-compliant wedge offering that immediately reduces their clients' administrative bloat.
The problem today
80 hours
wasted weekly on prior authorizations
$10.97
lost per manual prior authorization
45 mins
lost drafting a single insurance letter
Maria Delgado is the office manager for a 7-provider orthopedic group in suburban Columbus, Ohio. She's been doing prior auths by hand for six years and keeps a color-coded spreadsheet to track denial deadlines because the practice can't afford to miss an appeal window.
01The Problem
A medical assistant's sharpest morning hours disappear into copy-paste busywork that produces no patient care and no billable output.
Staff cost compounds across hundreds of PAs monthly into a five-figure annual drain on work that never improves a single patient outcome.
One wrong CPT code or misquoted guideline delays a patient's surgery by weeks and triggers an appeal cycle costlier than the original letter.
When the one person who knows each payer's quirks leaves, approval rates fall and denial backlogs pile up while a replacement spends months catching up.
Pulling a surgeon into PA drafts trades $300/hour clinical time for insurance paperwork — every interruption is a patient visit that didn't happen.
When headcount is the only way to process more authorizations, the PA backlog becomes a hard ceiling on patient volume.
02The Solution
Solution Brief
Fictional portrayal · illustrative
- Maria's group runs 200–500 PAs/month, mostly by hand
- Each letter: chart pull, payer cross-reference, clinical draft, surgeon sign-off
- Color-coded spreadsheet tracks denial deadlines — no actual system underneath
- Up to $5K/month and 80 staff hours spent on zero-care-value paperwork
- Physician interruptions convert $300/hr clinical time into admin output
- Formatting errors trigger denials that bury billing in appeals for weeks
- One resignation away from missing an unrecoverable appeal window
- Agent pulls structured data from EHR, matches payer-specific requirements
- Complete draft letter returned in under two minutes, correctly formatted
- 45-minute PA drops to five; surgeon scans instead of getting paged mid-clinic
- Denial rate falls as every letter hits payer expectations on first submission
- Patient volume grows without adding administrative headcount
“I used to block out my entire Tuesday for prior auths and still finish the week behind. Now I'm reviewing letters instead of writing them. I actually left on time last Friday, which hasn't happened in years.”
— Maria Delgado is the office manager for a 7-provider orthopedic group in suburban Columbus, Ohio
03What the AI Actually Does
Payer-Specific Letter Drafter
Pulls diagnosis codes, procedure codes, and clinical history directly from the EHR and generates a complete prior authorization letter tailored to each insurance carrier's documentation requirements — in under two minutes per case.
Clinical Justification Builder
Drafts the medical necessity narrative section of the PA letter using the patient's documented history and clinically appropriate language, reducing the need for physician involvement in routine authorization requests.
Human Review Checkpoint
Routes every AI-generated letter to a licensed clinician for review and approval before submission — a mandatory safeguard that catches hallucinated references and keeps the practice legally protected.
Denial Pattern Tracker
Logs outcomes by payer, procedure type, and letter template over time, surfacing patterns in denials so the practice can refine language and improve first-pass approval rates.
04Technology Stack
Doximity GPT (DoxGPT)
$0/month
Phase 1 quick-win tool. HIPAA-compliant AI writing assistant purpose-built for clinicians. Drafts prior authorization letters, appeal letters, and cli…
CoverMyMeds Electronic Prior Authorization
$0/month
Phase 1 pharmacy/medication PA platform. Largest ePA network — processed 43M+ prior authorizations in Q1 2025. Integrated with 350+ EHRs and 500+ heal…
Azure OpenAI Service (GPT-5.4)
$2.50/million input tokens + $10/million output tokens for GPT-5.4; ~$0.02 per PA letter; estimated $10–$50/month for typical practice volume of 200–500 letters
Phase 3 core LLM engine for custom PA letter generation. HIPAA-compliant under Microsoft BAA. Deployed within Azure subscription with private endpoint…
Microsoft Azure Subscription (HIPAA-eligible services)
$200–$500/month for App Service, Cosmos DB, Key Vault, and networking
Phase 3 hosting platform for custom PA letter generation tool. Includes Azure App Service (B2 tier for web app), Azure Cosmos DB (serverless for audit…
Microsoft 365 Business Premium
$22/user/month MSP cost / $30/user/month suggested resale
Foundation identity and productivity platform. Provides Microsoft Entra ID P1 (Conditional Access + MFA), Microsoft Defender for Endpoint P1 (EDR), Bi…
SentinelOne Singularity Complete
$5/endpoint/month MSP cost / $10/endpoint/month suggested resale
AI-powered EDR for all workstations and servers. Provides real-time threat detection, automated response, and forensic investigation capabilities. Req…
Duo Security MFA
$3/user/month MSP cost / $7/user/month suggested resale
Multi-factor authentication for all PHI-accessing accounts. Integrates with Microsoft Entra ID, EHR web portals, Azure Portal, and custom applications…
DNSFilter
$1.15/user/month MSP cost / $3.50/user/month suggested resale
DNS-layer security filtering to block malicious domains, prevent data exfiltration, and enforce acceptable use policies. Lightweight defense-in-depth …
Linear Health (if athenahealth EHR)
$500–$1,500/month estimated per practice
Phase 2 option for practices on athenahealth. Purpose-built referral and PA automation platform with native Athena integration. Most practices go live…
Tandem AI (for specialty practices)
$500–$2,000/month estimated per practice
Phase 2 option for specialty practices (dermatology, rheumatology, oncology). Automates medication prior authorizations and form-based authorization s…
Overjet (dental practices)
$300–$1,000/month estimated per practice
Dental-specific AI platform. Automates insurance verification across 300+ payers — verifies entire day's patient list in under 5 seconds. AI-powered a…
05Alternative Approaches
DoxGPT Only (Zero-Cost Quick Start)
$0/month
Deploy Doximity GPT as the sole PA letter drafting tool. Each clinician uses their verified Doximity account to access DoxGPT, pastes patient information into the prompt, and receives a draft PA letter. No custom infrastructure, no API costs, no application development. MSP role is limited to HIPAA stack deployment (firewall, EDR, MFA, backup) and staff training.
Strengths
- Zero software cost
- Immediate deployment (1 day)
- HIPAA-compliant out of the box
- No ongoing maintenance for the AI component
Tradeoffs
- No automation — staff must manually copy/paste patient data into DoxGPT for each letter
- No EHR integration
- No centralized audit trail (each clinician's usage is siloed)
- No payer-specific templates
- Not available for dental providers (dentists are not eligible for Doximity)
- No workflow enforcement — relies on individual clinician discipline for review quality
Best for: Solo practices or 2-provider offices with low PA volume (<50/month) where the cost of a custom solution cannot be justified.
Turnkey SaaS Platform Only (Tandem AI / Linear Health / Ethermed)
$500–$2,000/month estimated per practice
Skip the custom Azure OpenAI build entirely and rely on a purpose-built PA automation SaaS platform. The vendor handles all AI model hosting, EHR integration, payer connectivity, and compliance. MSP role is vendor selection, onboarding coordination, HIPAA stack management, and ongoing vendor relationship management.
Strengths
- Faster deployment (4–8 weeks vs. 8–16 for custom build)
- Vendor handles AI model updates, compliance, and payer integration
- Purpose-built for PA workflows with payer-specific logic built in
- Vendor assumes shared liability for AI output quality
Tradeoffs
- Higher monthly cost ($500–$2,000/month vs. $200–$500 for custom build)
- Less control over prompt engineering and output customization
- Vendor lock-in risk
- Limited white-label opportunity for MSP — cannot resell across client base
- May not cover all PA types (some vendors focus only on medication PAs)
Best for: Practices wanting fast results without custom development investment, particularly those on athenahealth (Linear Health) or specialty medication-heavy practices (Tandem AI).
Microsoft 365 Copilot with HIPAA Controls
$30/user/month on top of M365 E3/E5
If the practice is already on Microsoft 365 E3/E5 with Copilot licenses, use Microsoft 365 Copilot as the AI letter drafting engine. Staff draft PA letters in Word using Copilot, with custom prompt templates stored in SharePoint. Copilot is covered under the Microsoft HIPAA BAA when used within the M365 enterprise environment.
Strengths
- Leverages existing M365 investment
- Familiar Word interface for staff
- Covered under existing Microsoft BAA
- No separate application to maintain
Tradeoffs
- Copilot licensing is expensive ($30/user/month on top of M365 E3/E5)
- No structured workflow enforcement — relies on staff following manual processes
- No automated EHR data pull
- No centralized audit trail specific to PA letters
- Copilot output quality for medical letters is generally inferior to GPT-5.4 with specialized prompts
- No payer-specific template automation
Best for: Practices already invested in M365 E5 + Copilot ecosystem that want minimal new tool adoption and have low PA volume.
Open-Source LLM Self-Hosted (Llama 3 / Mistral on Azure)
$2,000–$5,000/month for GPU VMs
Instead of using Azure OpenAI (GPT-5.4), deploy an open-source LLM (Meta Llama 3 70B or Mistral Large) on Azure Virtual Machines with GPU (NC-series). This eliminates per-token API costs and keeps all data processing within the MSP-controlled Azure environment with no third-party AI vendor dependency.
Strengths
- No per-token API costs after infrastructure setup
- Complete data sovereignty — PHI never leaves the Azure subscription
- No dependency on OpenAI's availability or pricing changes
- Potentially simpler BAA story (only Microsoft BAA needed)
Tradeoffs
- Significantly higher infrastructure cost ($2,000–$5,000/month for GPU VMs vs. $10–$50/month API costs)
- Open-source models produce lower-quality medical text than GPT-5.4 — more clinician edits required
- Requires MLOps expertise to deploy, monitor, and update models
- Longer implementation timeline
- No model improvement without manual updates
Best for: Large practice groups or DSOs processing thousands of PAs monthly where API costs would exceed $1,000/month, or organizations with strict policies against sending PHI to third-party AI APIs.
Hybrid Approach: Voice AI (Simbie AI / Asepha AI) + Letter Generation
Voice AI platform cost + letter generation cost (varies by vendor)
Deploy a voice AI platform like Simbie AI or Asepha AI that automates the phone-based PA workflow (calling payers for benefit verification, status checks, peer-to-peer scheduling) alongside the letter generation system. This addresses both the written and verbal components of the PA process.
Strengths
- Addresses the full PA burden — not just letter writing but also hold times, phone tag with payers, and status follow-up
- Simbie AI claims up to 60% reduction in administrative costs
- Can handle appointment scheduling, prescription refills, and patient intake beyond PA
Tradeoffs
- Higher total cost (voice AI platform + letter generation)
- More complex implementation with multiple vendor integrations
- Voice AI for PA phone calls is still early-stage technology with mixed reliability
- Additional BAA and compliance overhead for voice recording/transcription
Best for: Practices where the majority of PA time is spent on phone calls with payers rather than letter drafting. Common in practices with high volumes of urgent PAs requiring real-time payer communication.
Ready to build this?