The Override Nobody Uses — clinicians.build

The setup

19.6%

Median in-network denial rate

0.15%

Median share of denials appealed

41.9%

Median appeal overturn rate

Every dot below is one insurer in one state. Its denial rate runs left–right; how often its denials get overturned on appeal runs bottom–top; the size is how many claims it processed. The question the newsletter asks about the bedside — is the override real, or decorative? — has a payer-side twin you can measure.

The explorer

Denial rate vs. how often the denial doesn't hold up

Minimum appeals filed: 1111 of 111 insurers shown

↑ Drag right to hide insurers with thin appeal counts. Watch the vertical scatter tighten toward the middle band — most of the 80%+ and sub-15% overturn "outliers" are small-denominator noise, not real generosity or cruelty.

one insurer-state = more claims processed

111

Insurers shown

—

Overturn-rate spread (IQR)

41.9%

Median overturn (shown)

The critical lens

Two things this data does — and does not — prove

A high overturn rate is easy to misread as "this insurer is fair, it fixes its mistakes." But look at the spread as you raise the filter: an insurer with 18 appeals and a 90% overturn rate is telling you almost nothing. The signal that survives is the appeal rate itself — and it survives at a fraction of a percent.

Override by omission. If ~42% of appealed denials get reversed, but only ~0.15% of denials are ever appealed, then the reversal isn't protecting most patients — it's protecting the tiny, motivated slice who file. The denial stands for everyone else not because it was right, but because the loop was never entered. A safeguard that 99.8% of people never trigger is, functionally, not a safeguard. It's a receipt.

Funnel reflects the 111 insurers currently passing the filter. "Appealed" and "overturned" are actual counts from the CMS PUF; the appeal rate is appeals ÷ denied claims.

Why it matters for builders

The newsletter's thesis — "a human is only a safeguard if they can actually override, and mean it" — is usually argued at the bedside. The payer side already ran the experiment at national scale. The override exists: reviewers overturn ~2 in 5 appealed denials. It's just priced so high in friction that almost no one uses it. If you're building automated adjudication, the number that decides whether your "appeal path" is real isn't the overturn rate. It's the appeal rate — the fraction of affected people who can actually find, afford, and survive the loop.

Read the KFF denials analysis → Or read today's newsletter — "The Override Problem" →