The Payers Are Winning. We Know. In A Past Life We Built Their Models.

The PromptWrx leadership team spent years building AI for health insurers. Claims review, prior authorization, the models that decide whether a claim gets paid or gets a letter.

So when Becker's asked 22 revenue cycle leaders to name the most dangerous trend in their world, and Jody Hinton at PDS Health called it "a growing arms race between payers and providers, and right now, the payers are winning," we didn't argue.

Two days later, Mike Marks, CFO of HCA Healthcare, said it from a stage at the Jefferies conference in Nashville: "We're behind the payers, in my view."

HCA runs 190 hospitals. If they're behind, you're behind.

We helped build the arsenal.

What we want to tell you is what the arms race looks like from the other side of the wall, and why most of the countermeasures providers are buying aren't going to work.

What 22 people said out loud

Read the Becker's piece and a pattern shows up fast. Tami McMasters Gomez at UC Davis put the imbalance in one sentence: payers are using automation to review 100% of claims while most health systems still manually review a small fraction of payer activity.

That's the whole war in one line. Every claim you send gets read by a machine. Almost nothing coming back gets read at all.

Aaron Klein at Carle Health named the weapon that worries me most: payers silently reducing payment through downcoding, with no formal denial activity. Shelly Thompson at Sentara called out AI-driven DRG downgrades and post-payment audits. None of it shows up in your denial report, because nothing was denied.

You got paid. Just less.

Then there's the part providers don't like hearing. Jill Gunn at Legent Health described a rush to buy AI without the operational infrastructure to make it work. Christopher Smith at Le Bonheur warned about automating broken processes. Balaji Raman at WMC Health and Dawson Ballard at Rush both flagged moving fast on AI with no governance behind it.

Twenty-two leaders, and a good share of them are more worried about their own AI than the payers'.

What it looked like from our side of the wall

Marks said payers have "been working really hard on their version of AI in their claims shop." Our team worked in that shop, and here's the uncomfortable part: nobody there called it an arms race. We called it tuning.

Every review model got tuned against the appeals that came back. Weak appeals, which was most of them, taught the model it could hold the line. Strong ones, with the right policy citation and the right clinical documentation, taught it where to stop.

We'll admit it. For a long time, our team read a low appeal rate as proof the model was right. It took moving to the provider side to see the other explanation: nobody had the staff to fight.

Your silence trained their models.

That's the piece the arms-race metaphor misses. Payers didn't win by buying better missiles. They won because most of what got fired back at them was a strongly worded letter with no policy behind it, and the machines learned exactly how much that was worth.

Five things we'd do Monday

1. Read 100% of what comes back. Payers read every claim you send. Until you read every remit, every DRG downgrade and every quiet downcode, you're negotiating blind. This is a data problem before it's an AI problem.

2. Triage before you fight. Correct, appeal, or write off. Not every denial deserves a letter, and a letter on a claim the payer got right teaches their model that you're noise.

3. Appeal like the reviewer, not like the biller. The model on the other side is checking policy language and clinical criteria. Cite the payer's own policy, the section, the effective date, and the documentation that meets it. Nothing else moves it.

4. Don't automate the process you already have. Smith is right about this. An AI that writes your current appeal letter ten times faster produces ten times more of the letter that's already losing.

5. Count dollars you should have been paid, not denials. Klein's silent reductions never hit a denial queue. HCA has had Parallon pointed at denials and underpayments since late 2022; notice the second word. If your dashboard starts at the denial, you're measuring a war the payers moved past two years ago.

Bringing it home

Hinton's framing is right, but an arms race implies symmetry. Both sides build missiles.

This one isn't symmetric. The payers built review models. Providers are buying letter-writers.

Marks says the claims environment needs reform between providers and payers. He's right, and reform will arrive on its own schedule. Your denials arrive on yours.

That's why PromptWrx built Reve the way we did: the same people who built the payer models, now working only for providers, and never for a payer. It reads every claim the way the other side does, decides whether you were right or just billed, and fights the ones worth fighting with the payer's own policy in hand.

Do you want to be right, or do you want to get paid?

Read every remit. Triage. Cite policy. Fix the process before you automate it. Count dollars, not denials.

The payers' models have been learning from your silence for years. Give them something else to learn from.