In short. Traditional Medicare denies around 5 percent of claims on first pass. Medicaid inpatient denies 44 percent. Your blended denial rate averages populations that behave nothing alike, which is why it tells you something is wrong and nothing about what to change. The money is in the cohorts, and up to 65 percent of it is never chased.
A blended rate averages four populations that behave nothing alike
Across US providers, initial denial rates now run around 11.8 percent, up from 10.2 percent three years ago. That figure is the one that reaches most board packs. Underneath it the spread is enormous.
Traditional Medicare sits near 5 percent. Medicare Advantage has passed 17 percent, more than double. Medicaid inpatient denials run at 44 percent on first pass. These are not variations around a mean. They are different businesses with different failure modes, blended into a single line item because that is how the report was designed.
Same organization, same month, same report.
A blended rate is a management number. It tells you something is wrong and nothing about which process to change on Monday.
Call the thing you actually need cohort visibility: denial rate by payer, by procedure group, by denial reason code, and ideally by the person or step where the claim was assembled. That is where a denial rate stops being a fact about your organization and becomes an instruction.
What this is worth in your organization, roughly
Denials cost US providers around $262 billion a year, and the average hospital absorbs roughly 5 percent of net patient revenue in denial-related losses.
Apply that to the segment this is written for. A single-specialty hospital at $30 million of net patient revenue is carrying something in the region of $1.5 million a year. A mid-size multi-specialty group at $150 million is carrying closer to $7.5 million. For a five-provider clinic the absolute number is smaller and the proportion of operating margin it represents is usually larger.
Then the part that turns a cost into an opportunity. Up to 65 percent of denied claims are never reworked at all, and roughly two thirds of denied claims are recoverable. Reworking one costs somewhere between $25 and $181 depending on complexity, with medical necessity appeals at the top of that range.
So a meaningful share of what you are writing off was collectible, and was abandoned not because anyone decided to abandon it but because the queue was longer than the day.
This is not a reporting problem, which is why more reporting has not fixed it
Most provider organizations of this size already have the data. It sits in the practice management system and the clearing house, and somebody can pull it if you ask.
Three things stop it becoming money.
It arrives late. A monthly denial report describes a month that is already closed. The payer policy change that caused the spike happened six weeks ago and has been quietly repeating ever since.
It stops at the observation. A report that says orthopedic denials from one Medicare Advantage plan rose sharply is interesting. It does not say that the plan changed its documentation requirement on the fifteenth, that eleven claims are failing the same way each week, or which step in your process needs to change.
Nobody has capacity to act on it anyway. Your team is working today's queue. Finding out that last month leaked does not create the hours to go back for it.
What a governed analytics layer does differently
It watches at cohort level continuously rather than compiling monthly. When a specific payer, procedure and reason code combination starts failing, that is a signal, not a number in a table.
It states the cause in operational language and routes it to whoever owns that process, with the evidence attached: these fourteen claims, this reason code, this payer, first occurrence on this date, this is the documentation element that was missing.
And it puts an agent at the point where the work is actually done, so the fix is applied to the next claim rather than proposed at the next meeting. The agent assembles, checks and drafts. Your team decides and signs. Nothing about a denial gets approved by software on its own.
Then the same layer measures whether the leak closed, which is the part that turns an analytics investment into something defensible at budget time.
What to ask for this week
None of this requires an Epic-scale IT program, and none of the organizations this is written for have one. Four questions get you most of the way.
If those four take more than a week to produce, that delay is itself the finding.
Ask for
Denial rate split by payer, not blended
Your top ten denial reason codes by dollar value, not by count
Share of denied claims that were never reworked
Days between a denial occurring and anyone noticing the pattern
What you will probably find
One or two payers carrying most of the problem
The expensive denials are rarely the frequent ones
Higher than anyone in the room expects
This is the number that decides how much you lose
Start with one cohort
Pick the single payer and procedure group carrying the most denied dollars. Baseline it properly: denial rate, rework rate, recovery rate, days to detection. Watch it at cohort level for a month while your team works as usual. Then put an agent on the specific failure the data exposes, and measure the same four numbers again.
One cohort closed, with the before and after in your own figures, will do more for the next budget conversation than a platform selection exercise.
What is your denial rate on your worst payer, and how long did it take you to find out?
Next in this series: why an alert nobody acts on is just a dashboard. And if the question on your mind is what this costs to run per workflow rather than per month, that is covered in You're Measuring AI Spend by Month. Your CFO Wants It by Workflow.
Recent blogs
Secure your agents
We’d love to chat with you about how your team can secure and govern Ai agents everywhere
