A denied claim can tell the same clinical story multiple times.

During physician review, the clinical case gets reconstructed from the medical record: the notes, the labs, the imaging. During an appeal, that evidence becomes a formal argument. In later clinical reviews, or in legal proceedings if a case goes that far, the same information may need to be assembled and interpreted again.

Different stages. Same underlying clinical story. And a question worth asking at every step: how much of that story survives each handoff?

One challenge with specialized workflows is that each stage naturally focuses on its own deliverable. Clinical review emphasizes the medical evidence. Appeals emphasize the argument. Legal review emphasizes the defensibility of the record. None of that is a failure. It is what specialization is for. But even when each function performs well on its own terms, continuity across the full lifecycle takes a different kind of coordination: when the story gets rebuilt at each stage, important context can be lost along the way.

The published financial data reinforces why looking at individual stages is not enough. Kodiak Solutions, analyzing claims and financial data from more than 2,300 hospitals, found that denial rates fell in the first half of 2026 while cash flow did not improve: insurer takebacks climbed to 1.57 percent of accounts receivable, up 13.8 percent from a year earlier, and the share of accounts receivable that providers recovered on appeal fell 12.2 percent. The broader lesson is worth emphasizing: measure final revenue yield, not just the performance of individual stages.

Becker’s, in a piece published with Xsolis, recently profiled a health system where one physician advisor completed 5,118 reviews across 11 hospitals in 2025, with a 78 percent peer-to-peer overturn rate. The interesting part is what came before those results. The system cycled through three operating models, and in the profile’s telling, the recurring problem was the same each time: no unified system for real-time clinical visibility, and no systematic way to prioritize cases. The lesson I take from that example is that better outcomes may require more than individual effort. They may require rethinking how visibility, prioritization, and clinical information connect across the process.

The appeals data points the same direction. A research letter in JAMA Internal Medicine looked at 51,394 closed external appeals in New York from 2019 to 2025 and found that 46.7 percent were overturned once independent reviewers examined them. A separate population tells the other half of the story: in KFF’s analysis of HealthCare.gov marketplace plans, fewer than 1 percent of denied claims were appealed at all. Two different data sets, one important question: if so few denials are appealed, and so many reviewed decisions are overturned, how do we make the process of identifying, preparing, and advancing the right cases more effective?

So the more interesting design question is whether the clinical narrative itself could become the shared asset across the dispute lifecycle. One way to picture it, in general terms: the narrative stays shared and governed as it moves, so each stage builds on what came before instead of starting over, and the story gets sharper at every step instead of being reconstructed. In that model, the measurement changes too. Instead of grading each document on its own, you grade whether the narrative survives from the first review to the final outcome.

This is where the maturity curve bends again. Level one is point tools: a summary tool here, a letter writer there. Level two is workflow automation: faster handoffs, cleaner queues. Both necessary. Level three is where the value compounds: governed continuity, where the clinical narrative is the asset and each stage builds on the last instead of starting from zero. The hard part for the industry is progressing beyond point solutions to that third level.

The question I keep coming back to: did the tool do the thinking, or did it make the thinking better? Answer that first. Everything else is formatting.

This is the second post in The Maturity Curve, a series on enterprise AI adoption. More to come.