A 350-bed not-for-profit community hospital serving a mixed urban-suburban population — with active surgical, cardiac, orthopedic, and emergency departments. The hospital processes approximately 18,000 inpatient and outpatient claims per month across commercial payers, Medicare, and Medicaid, with appeals handled by a four-person team within the Patient Financial Services department.
The hospital’s denial rate had been climbing steadily — from roughly 9% two years earlier to just over 13%. The rise was driven largely by two categories: medical necessity denials on inpatient admissions, where commercial payers were increasingly challenging observation-to-inpatient conversions, and prior authorization denials on surgical procedures, where payers were tightening requirements faster than the front-end team could keep up.
The four-person appeals team was experienced. On the cases they worked, their overturn rate was strong — around 58%, in line with industry averages. The problem wasn’t their skill. It was their capacity.
Each appeal required 45 minutes to an hour: reviewing the denial reason, pulling clinical documentation from the EHR, cross-referencing the relevant payer’s medical policy, and drafting a letter that made the case for medical necessity or authorization validity. The team could file roughly 8–10 appeals per person per day. At that throughput, they had to triage — and they did. High-dollar inpatient denials got worked. Complex surgical cases got worked. Everything else waited.
A backlog analysis commissioned by the CFO found that the hospital had approximately 1,400 appeal-eligible denials from the prior 12 months that had never been worked — claims where the denial type, payer history, and available documentation suggested a reasonable probability of overturn, but the team simply hadn’t had the bandwidth to file. The estimated recoverable value: over $3M.
The hospital had explored outsourcing appeals. Two vendors quoted $85–$120 per appeal, with estimated overturn rates of 40–50% — lower than the internal team’s rate, because the outsourced teams didn’t have the same familiarity with the hospital’s clinical documentation patterns or its payer-specific history. At those economics, only the highest-dollar denials justified the outsourcing cost. The mid-range denials — $1,500 to $5,000 — remained uneconomical to appeal manually.
NDS was brought in to solve a specific capacity problem: generate submission-ready appeal letters for the volume of denials the four-person team couldn’t get to — without compromising the quality that made the team’s overturn rate strong in the first place.
The system was configured to read the hospital’s 835 remittance files and denial correspondence, identify denied claims, and classify each by denial type, payer, dollar amount, and filing deadline. The hospital’s appeals leadership defined the scope: the AI would handle coding denials, bundling denials, and routine authorization denials across all payers. Complex medical necessity denials on high-dollar inpatient cases — where clinical nuance and physician involvement were critical — remained with the human team.
For each denial in scope, the AI gathered the supporting information relevant to the denial type — clinical documentation from the hospital’s EHR, coding logic, authorization records, and submission timelines. It then referenced the applicable payer’s guidelines to identify the basis for appeal and the supporting language needed. From there, it drafted a complete, original appeal letter — not from a template, but structured around the specific denial, the specific payer, and the specific supporting evidence for that claim.
Every AI-generated letter was presented to the appeals team in a review interface before submission. Staff could read the full draft, edit any section, add or remove supporting documentation, and approve or reject. In practice, the team found that the vast majority of letters required only minor adjustments — a clarification on a date of service, an additional clinical note reference. The AI had done the research, assembled the evidence, and written the argument. The human reviewer confirmed it was right.
The impact on throughput was immediate. Reviewing and approving an AI-generated appeal took 8–12 minutes — compared to the 45–60 minutes it took to research and write one from scratch. Each team member could now process 25–30 appeals per day instead of 8–10. The team’s capacity tripled without adding a single FTE.
Filing deadline tracking gave the team visibility they hadn’t had before. Previously, deadline management was a manual spreadsheet process, and the team estimated they missed 5–10 filing deadlines per month on lower-priority denials. The system eliminated that entirely — every denial in the pipeline was tracked with its payer-specific deadline, and approaching deadlines triggered automatic prioritization in the review queue.
Pilot on coding and bundling denials from the hospital’s three highest-volume commercial payers. One early finding: the AI’s initial letters for bundling denials on a specific payer cited a CCI edit rationale that the payer didn’t recognize — the payer used its own proprietary bundling logic. The appeals team flagged it, NDS updated the payer intelligence layer, and subsequent letters for that payer referenced the correct policy basis. The correction improved approval rates on bundling appeals for that payer from the first month.
Expanded to all commercial payers and added routine authorization denials to the AI’s scope. The team also began using the AI to draft initial appeal letters for a subset of medical necessity cases — ones where the clinical documentation was clear-cut — with physicians reviewing the clinical argument before submission. This freed physician time on the cases that genuinely required their direct involvement.
Full production across all payers and in-scope denial types. The hospital used the AI’s denial classification data to identify that 22% of its authorization denials traced to three specific surgical procedure codes where the front-end team was consistently submitting authorizations with insufficient clinical documentation. A targeted training intervention on those three codes reduced authorization denials on those procedures by 31% within two months.
In recovered revenue from denials that had previously gone unworked — representing roughly 70% of the $3M the backlog analysis had identified as recoverable.
increase in appeals filed per FTE per day. Review-and-approve replaced research-and-write — from 8–10 appeals per person per day to 25–30.
overturn rate on AI-generated appeals — exceeding the internal team’s historical 58% rate on manually written appeals. The AI’s consistent citation of payer-specific guidelines contributed to the higher success rate.
missed filing deadlines since go-live. Automated deadline tracking and queue prioritization eliminated the 5–10 monthly deadline misses the team had been experiencing.
Unexpected outcome: With the AI handling the volume of routine appeals, the four-person team was able to redirect their time to the complex medical necessity cases that required physician collaboration and clinical nuance. The overturn rate on those high-dollar inpatient cases — which had always been the team’s priority but often got rushed under volume pressure — improved from 52% to 64% in the six months after deployment. The CFO estimated the combined impact of AI-generated appeals on routine denials and improved outcomes on complex cases at over $2.8M annually.
What the Client Said
"We had 1,400 denials sitting there that we knew were recoverable. We just couldn’t get to them. The AI didn’t replace my appeals team — it gave them back the time they were spending on research and letter writing so they could focus on the cases that actually need a human. The surprise was that the AI’s overturn rate was higher than ours. Turns out, consistently citing the right payer policy in the right format matters more than we thought."
Discover the way forward through more case studies.
Share your sample remittance data and your organization’s business rules, and we’ll build a proof of concept showing how NDS reads your denials and generates submission-ready appeal letters — tailored to your payers and denial types. No setup fees. No commitment. Just proof on your actual denials.
© 2026 NDS InfoServ. All rights reserved.