Loading
Skip to content
Miracle Minds AI
  • Try “HIPAA”, “model risk”, “FinOps” or “agentic”.

Revenue cycle and claims

Payers reverse most of the denials that providers appeal, yet few denials are ever appealed. The yield is in clean documentation before submission and in appealing what you can win, with every automated code traceable to the chart.

Part of AI for Life Sciences & Healthcare. 10 sources, checked .

spent by US hospitals in 2025 trying to collect payment from insurers for care already delivered (AHA estimate)
$43bn
spent by US hospitals in 2025 trying to collect payment from insurers for care already delivered (AHA estimate)
American Hospital Association, 2026(opens in a new tab)
of appealed Medicare Advantage prior authorization denials were overturned in full or in part in 2024, yet only 11.5% of denials were appealed
80.7%
of appealed Medicare Advantage prior authorization denials were overturned in full or in part in 2024, yet only 11.5% of denials were appealed
KFF, 2026(opens in a new tab)
average share of in-network claims denied by HealthCare.gov insurers in 2024
19%
average share of in-network claims denied by HealthCare.gov insurers in 2024
KFF, 2026(opens in a new tab)

The problem

Hospitals spent an estimated $43 billion in 2025 trying to collect payment for care already delivered, including nearly $18 billion on overturning claim denials.1 In 2024 the average hospital employed about 64 administrative and billing staff on this work.1 Physician practices carry the same load: they complete an average of 40 prior authorizations per physician each week, and physicians and staff spend 13 hours a week on them.2

Denials are frequent and often reversed. HealthCare.gov insurers denied an average of 19% of in-network claims in 2024; administrative reasons accounted for 25% of denial reasons and medical necessity for only 5%.3 Medicare Advantage insurers denied 7.7% of 52.8 million prior authorization requests in 2024, in full or in part. Only 11.5% of those denials were appealed, and 80.7% of appeals were partially or fully overturned.4

Regulation is forcing new plumbing. Under CMS-0057-F, most impacted payers must decide prior authorizations within 72 hours for urgent and seven calendar days for standard requests from January 2026, give a specific reason for denials, and run a Prior Authorization API by January 2027.5 Readiness lags: 63% of medical plans were still developing electronic prior authorization FHIR APIs, and only 20% of providers had begun implementation.6

AI can draft codes, assemble prior authorization packets and write appeals, but accuracy cannot be assumed. In a 2024 benchmark, GPT-4 returned the exact ICD-10-CM code for a code description only 33.9% of the time.7 Payers face limits too: Medicare Advantage plans must base medical necessity decisions on the enrollee's history, physician recommendations and clinical notes,8 and a qualified clinician must review any adverse decision.9 Physicians are wary: 60% worry AI will increase prior authorization denials.2

What we build

  1. 01

    Pre-submission claim checks

    Models trained on your own remittance history flag claims likely to deny for eligibility, authorization or documentation gaps before they go out, with the specific fix for each.

  2. 02

    Prior authorization packet assembly

    An agent checks the payer's requirements, pulls the clinical evidence from the chart and drafts the request for staff review, ready for payer Prior Authorization APIs as they come online.

  3. 03

    Evidence-linked coding assistance

    Suggested codes shown beside the chart text that supports each one, with a coder accepting or changing every code. No autonomous final coding in high-risk categories.

  4. 04

    Denial triage and appeal drafting

    Denials ranked by expected recovery, with first-draft appeal letters that answer the payer's stated reason from the clinical record.

  5. 05

    Payer behavior analytics

    Denial rates, reasons and turnaround by payer and service line, combined with the prior authorization metrics payers must now publish, to support escalation and contract talks.

Where it goes wrong

The failure modes we design against from the first week, and the control for each.

  • Upcoding at machine speed

    A model that nudges codes upward repeats the error on every claim, and the False Claims Act reaches claims made in reckless disregard of their accuracy, with treble damages.10 Control: every suggested code links to its documentation, a coder signs off, and a monthly sample is audited by an independent reviewer.

  • Trusting general-purpose models with code lookup

    Tested language models often returned imprecise or fabricated billing codes.7 Control: constrain models to retrieval from current code sets and payer policies, validate every code against official tables, and measure accuracy on your own claims before use.

  • Payer algorithms that replace individual review

    Medicare Advantage rules require medical necessity decisions to rest on the enrollee's own record,8 and a physician or other appropriate professional must review expected adverse decisions.9 Control: AI may triage and summarize, but a qualified clinician makes and documents every adverse call.

  • Automating appeals without fixing causes

    Writing more appeals treats the symptom. Control: feed denial root causes back to registration, documentation and authorization teams, and track whether each denial reason falls month over month.

Rules that apply

  • CMS Interoperability and Prior Authorization final rule (CMS-0057-F)

    Primary text on cms.gov(opens in a new tab)

    Medicare Advantage, Medicaid and CHIP payers must meet 72-hour and seven-day prior authorization decision times from 2026; impacted payers must state a specific denial reason, publish prior authorization metrics, and build a Prior Authorization API by January 2027.

  • False Claims Act, 31 U.S.C. 3729

    Primary text on govinfo.gov(opens in a new tab)

    Knowingly submitting false claims, including with reckless disregard for accuracy, brings per-claim penalties and three times the government's damages, so automated coding needs documented controls.

  • Medicare Advantage medical necessity, 42 CFR 422.101(c)

    Primary text on ecfr.gov(opens in a new tab)

    Plans must base medical necessity decisions on the enrollee's medical history, physician recommendations and clinical notes, and may not deny basic benefits on criteria the rule does not allow.

  • Medicare Advantage clinician review, 42 CFR 422.566(d)

    Primary text on ecfr.gov(opens in a new tab)

    A physician or other appropriate health care professional must review any expected adverse medical necessity decision before it is issued.

How to measure it

  • Initial denial rate by payer and denial reason
  • Clean claim rate at first submission
  • Appeal rate and overturn rate on denied dollars
  • Days in accounts receivable and cost to collect
  • Coding accuracy on audited samples, counting both over-coding and under-coding

Questions buyers ask

Is it legal to use AI for medical coding?

Yes, but you remain responsible for every claim. The False Claims Act covers claims submitted with reckless disregard for accuracy, so an unchecked model is a liability. Use AI to suggest codes linked to the supporting documentation, keep a certified coder accountable for the final code, and audit samples regularly against an independent reviewer.

What does CMS-0057-F require, and when?

It applies to Medicare Advantage organizations, state Medicaid and CHIP programs and their managed care plans, and qualified health plans on the federal exchanges. From January 2026, most must decide prior authorizations within 72 hours for urgent and seven days for standard requests, and give a specific reason for denials. By January 2027 they must run a Prior Authorization API.

Which denials should we automate appeals for first?

Start with high-volume denials where the record already holds the evidence: missing or mismatched authorizations, eligibility errors and documentation requests. These are rules-based and quick to verify. Leave medical necessity disputes for clinician review. Rank the queue by expected recovery, meaning denied dollars times your historical win rate for that payer and reason.

Sources

Every figure on this page comes from the sources below. We checked each link on .

  1. 1.Costs of Caring: Challenges Facing America's Hospitals as They Care for Patients in 2026(opens in a new tab)American Hospital Association, March 2026 · aha.org
  2. 2.2025 AMA prior authorization physician survey(opens in a new tab)American Medical Association, 2025 · ama-assn.org
  3. 3.Claims Denials and Appeals in ACA Marketplace Plans in 2024(opens in a new tab)KFF, March 24, 2026 · kff.org
  4. 4.Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024(opens in a new tab)KFF, January 28, 2026 · kff.org
  5. 5.CMS Interoperability and Prior Authorization Final Rule CMS-0057-F (Fact Sheet)(opens in a new tab)Centers for Medicare & Medicaid Services, January 17, 2024 · cms.gov
  6. 6.2025 CAQH Index Report: Recovery by the Numbers (Executive Report)(opens in a new tab)CAQH, 2025 · index.caqh.org
  7. 7.Large Language Models Are Poor Medical Coders: Benchmarking of Medical Code Querying(opens in a new tab)NEJM AI, April 19, 2024 · ai.nejm.org
  8. 8.42 CFR 422.101: Requirements relating to basic benefits(opens in a new tab)Electronic Code of Federal Regulations (CMS) · ecfr.gov
  9. 9.42 CFR 422.566: Organization determinations(opens in a new tab)Electronic Code of Federal Regulations (CMS) · ecfr.gov
  10. 10.31 U.S.C. 3729: False claims(opens in a new tab)U.S. Government Publishing Office (United States Code), 2023 · govinfo.gov