As denials surge, patients use AI to take on health insurers

AI-fueled appeals are coming for health claim denials - here's how insurers should respond

Insurers are leaning on AI to process claims. Patients are now using AI to push back. That tension is reshaping denial and appeal workflows across the industry.

In 2023, tens of millions of in-network claims in ACA plans were denied - and fewer than 1% were appealed. With consumer-facing AI tools, that gap will shrink fast. Legal experts, including Indiana University law professor Jennifer Oliva, expect more people to challenge denials with better documentation and stronger arguments.

What patients are using AI to do

• Scan EOBs and pinpoint coding issues, missing documentation, and inconsistent reason codes.
• Cross-check denials against plan policies and clinical guidelines (e.g., medical necessity, prior auth, frequency limits).
• Draft appeal letters with citations, timelines, and properly attached evidence.
• Create "paper trails" - summaries of calls, portal submissions, and required response dates.
• Batch similar denials (e.g., modifiers, lab panels) for repeatable, high-quality appeals.

What this means for payers

• Appeal volume will rise, and letters will read like they came from a seasoned RCM team.
• Overturn rates may climb in categories with weak documentation or ambiguous policy language.
• Members and providers will expect clearer rationales, faster timelines, and consistent outcomes.
• Regulatory scrutiny will focus on explainability and fairness in automated determinations.

Operational actions to take now

• Evidence packs with every denial: Return reason codes, policy excerpts, and claim-level facts used by the model. Make it copy-paste friendly for both providers and members.
• Human-in-the-loop thresholds: Route high-dollar, high-impact, or ambiguous cases to senior clinicians before final adverse determinations.
• Denial hygiene: Eliminate vague reason codes; collapse duplicative codes; map each to a clear remediation step.
• Digital appeal pathways: Offer a clean upload flow for clinical notes, pictures of prescriptions, prior auth proofs, and physician letters.
• Proactive outreach: Trigger care management or provider education when you see repeatable, preventable denials.
• Closed-loop learning: Feed overturned appeals back into model training and rule tuning. Prioritize categories with high overturn rates.

Model governance that won't buckle under audits

• Explainability by default: Store model features, policy citations, and decision snapshots. If you can't explain it, don't automate it.
• Bias and drift testing: Monitor denial and overturn rates by product, provider type, geography, and condition group. Investigate outliers.
• Policy retrieval: Use retrieval-augmented generation from a versioned policy library to prevent hallucinated citations.
• Adversarial review: Red-team your prompts and rules with "member-style" and "provider-style" inputs to see what gets through.
• Strong escalation paths: Make it simple for clinicians and compliance to override the model and annotate the reason.

Metrics that actually move outcomes

• Denial rate by reason code and line type, not just header-level.
• Overturn rate (internal and external review) and dollar-weighted impact.
• Time to resolution for appeals and percentage resolved on first appeal.
• Preventable denials (documentation, coding, eligibility) and recurrence after provider education.
• Complaint signals from members and providers, plus regulator inquiries.
• MA Stars/HEDIS impacts tied to grievances and access issues.

Policy and compliance checkpoints

• Ensure adverse determination letters clearly state the specific reason, evidence reviewed, and appeal rights with timelines.
• Honor internal and external review timeframes and maintain thorough audit trails.
• For ERISA plans, align with federal claims and appeals procedures; for Marketplace products, follow federal external review standards.

Useful references: KFF analysis of Marketplace denials and appeals and the DOL ERISA claims/appeals framework.

Where AI can help payers - fairly

• Pre-claim checks: Provider-facing bots that flag missing clinical elements before submission reduce avoidable denials.
• Consistency engines: Compare similar claims and providers to spot outlier decisions and tighten policies.
• Guided remediation: Give providers precise next steps (exact note, code, or attachment) based on the denial reason.
• Member empathy scripting: Assist reps with clear, plain-language explanations and rights without legal jargon.

Team readiness

• Train frontline teams on new denial reasoning, appeal timelines, and how to talk about AI-assisted decisions.
• Publish internal playbooks: model thresholds, escalation rules, documentation checklists, and outreach templates.
• Run monthly calibration sessions using real overturned cases to refine rules and communication.

AI won't replace adjudication policy. It will expose where policy is unclear, where documentation standards are inconsistent, and where communication is weak. Tighten those, and both appeals and costs fall.

If you're upskilling teams to work effectively with AI in claims and operations, explore curated programs by role at Complete AI Training.