Half the denials nobody works

the number

Between 50 and 80 percent of denied claims are never reworked

Start with the three figures that describe a denial file. The denial rate across US claims reached 11.8 percent in 2024. Around 27 percent of denials are eligibility errors, which means they were catchable before the patient walked in. And between 50 and 80 percent of denials are never reworked at all.

The third figure is the one worth an evening. A denial that is reworked is paid about two thirds of the time. A denial that is not reworked is paid none of the time. So the question is not what your denial rate is. It is what share of your denials nobody ever touched again, and what that share is worth.

Here is the arithmetic, with the assumptions in the open so you can put your own numbers in. Take a six-clinician practice billing 1,200 claims a month at an average allowed amount of 180 dollars. At 11.8 percent, 142 of those claims are denied each month. If 65 percent of them are never reworked, that is 92 claims, or roughly 16,600 dollars of billed value, going out of the door every month without anyone deciding it should.

Not all of it is recoverable. Some of those denials are correct. But the eligibility share alone, the 27 percent that were avoidable before the appointment, is about 25 claims a month in this practice, and those are not a billing problem. They are a front-desk problem that shows up as a billing problem.

How to check it on your own file this week: export ninety days of remittance advice, count the distinct claims with a denial, then count how many of those have a later resubmission. The ratio is the only number in this section you cannot get from a benchmark.

the frontier

An evaluation you did not run yourself tells you nothing

Every vendor in clinical documentation, ours included, will show you an accuracy figure. Almost none of them will tell you which notes it was measured on, who graded them, or whether the graders knew which system produced what. Without those three facts the figure means nothing.

The reason is not that the numbers are dishonest. It is that a note is not a transcript, and the thing that goes wrong in practice is almost never a word. It is a decision: what got carried into the assessment, what got dropped from the history, whether the plan matches what was said. Word error rate cannot see any of that, which is why a system can score beautifully and still produce notes a clinician rewrites.

The version that does tell you something takes an afternoon. Fifty of your own de-identified notes, produced twice, shuffled so nobody can tell which system wrote which, and three reviewers scoring against a rubric you wrote before you saw any output. It is not glamorous, but it is the only evaluation you can trust.

We publish the rubric in a later issue. The rubric is the part you can build now; the shuffling is the part people skip, and it is the part that matters.

from the floor

We sent a reminder two hours early and eleven people missed appointments

A reminder job that had been running fine for months started firing against the clinic's time zone rather than the patient's. For patients in the same city as the clinic nothing changed. For eleven patients who had moved, or who were travelling, the reminder landed two hours before it should have, and the appointment had already been and gone by the time they read it.

The fix was an hour. The interesting part is why it took four days to notice: nothing failed. Every message sent, every message delivered, every dashboard green. The only signal was eleven no-shows in a week that should have had four, and no-show counts are noisy enough that a bad week does not look like a bug.

What we changed: reminders now carry the patient's own offset, and the no-show rate is alarmed on a seven-day rolling comparison rather than eyeballed. What we should have done: alarmed it before shipping the job, because a scheduled message can send correctly and still be wrong.

three links

1. Allocation of Physician Time in Ambulatory Practice

   Sinsky et al., Annals of Internal Medicine, 2016

   The two hours of desk work per hour of care, measured directly rather than surveyed. Almost every claim about clinician burden traces back here, and it is worth reading rather than citing.

2. The Short Life of Online Sales Leads

   Oldroyd, McElheran and Elkington, Harvard Business Review, March 2011

   The audit of 2,241 companies behind every response-time statistic you have been quoted. Not about medicine, which is why it is useful: nobody wrote it to sell you a phone system.

3. Why we publish our prices

   wai

   Ours, and the only one of the three that is. Four of the six largest products in this category will not tell you what they cost without a demo first, and this is the argument for why that is a choice rather than a norm.