A claim denies on a Tuesday morning. The payer’s system read it, applied a rule, and rejected it in milliseconds, before any human looked at it.
The Claim
By the time it reaches the work queue, the dashboard has already done its job: flagged it, ranked it by recoverable dollars, sorted it to the top. Everyone can see it is worth chasing.
Then it sits. Behind a few thousand other claims that are also flagged, also ranked, also worth chasing. Ninety days later it crosses the timely-filing limit and becomes a write-off. Nobody decided to abandon it. There just weren’t enough hands to reach it in time.
“Nobody decided to abandon it. There just weren’t enough hands to reach it in time.”
I run Anka, a revenue cycle company. My teams work denied and underpaid claims for US physician groups every day, and we build the AI meant to do parts of that work. So I see both sides of a question the industry keeps fumbling: what a machine can actually recover, and what still needs a person. You learn that line fast when you are paying salaries for the part that doesn’t automate.
Here is the line.
Denial Is Automated. Recovery Is Not.
A payer rejects a claim in milliseconds. A person recovers it in twenty minutes to two hours, reading the remittance, finding the reason, pulling documentation, working the portal, following up. That asymmetry is the whole problem, and almost nothing in the last decade of revenue cycle technology has touched it.
And the pressure is rising. Recent market analysis shows industry-wide initial denial rates have climbed to 11.8%. For physician groups and healthcare operators already managing thin margins, that means more work entering the queue faster than teams can realistically clear it.
What That Decade Sold Instead
What that decade sold instead was visibility. Dashboards. Analytics. “AI-powered” denial insights. They are very good at one thing: showing you exactly where your money is stuck. And then they stop. The work, the appealing, the submitting, the following up, drops right back into a human queue, exactly where it sat before.
And seeing doesn’t solve, because the arithmetic doesn’t close. One specialist clears about 40 claims a day at full tilt. Denials, rejections, no-activity claims, and underpayments are in the hundreds to low thousands every month, and last month’s backlog is still in the queue.
The Execution Gap
When the first number grows faster than the second can scale, you don’t have a backlog you clear. You have a structural deficit that widens every month. Hiring against it means scaling the most expensive, least durable part of the operation faster than the denials rise, which you can’t. A better dashboard doesn’t help either. A sharper map of a fire you can’t reach won’t put it out.
What Closes the Gap
What closes the gap is a different kind of AI, not the kind that watches the work, the kind that does it.
Call it an execution layer.
That is the part the industry skipped. Not seeing the work. Doing it.
Two Things It Is Not
Because I build this and the hype cuts both ways.
It is not the fully autonomous, no-humans revenue cycle on the vendor slide, that is the asymmetry sold back to you in reverse. A person still owns the judgment calls and the real exceptions. The machine carries the volume.
And it is not yesterday’s RPA, which broke the moment a payer moved a button on a portal. Execution AI reasons through the variation that makes recovery hard, which is exactly why it can finally operate at the speed denials arrive.
Because that is what this resolves: the asymmetry we started with. Payers answer at machine speed. Until now, recovery couldn’t. An execution layer is how it finally does.
The Real Question
So, the question for a revenue cycle leader, CFO, or healthcare investor isn’t, “Do you know where your revenue is leaking?”
You know, your dashboard told you months ago.
“The real question is what, in your operation, closes the distance between knowing and doing. Because today, for most, that distance is measured in claims that died exactly like the one above.”
