4 places rehabs lose insurance money they already earned (2026)

Ten rehabs in California are suing one insurance company. They say they billed over $15 million, but they only got paid about two and a half. That's 16 cents on the dollar. Insurance companies have been sued for using algorithms to deny claims. Medicare is testing AI on prior authorization in six states right now. So their side of the table has software. Your side of the table has a biller who is buried. We're going to talk about how to fix this.

If you run an IOP or rehab, you already know this one. Getting paid is one of the top worries in the industry right now. About a third of operators say it's their single biggest challenge. And it's about to get harder. Medicaid work requirements start January 1st, so Medicaid expansion patients have to re-qualify every six months. It used to be a year. People in addiction treatment can be exempt, but they still have to get the paperwork in. A legal aid lawyer in Philadelphia is already warning about that: people who qualify are still going to lose coverage over the form. One treatment CEO said it best: "Our patients are afraid they're going to get billed, so they stopped coming in."

AI is not going to raise your rates, and anybody who says that is definitely selling you something. Where it pays off is collecting money that you've already earned. And for most centers, that's a lot of money. I'm Dave. I help business owners pick and set up AI in their real businesses, and I'm going to show you four leaks to plug first and how to use AI for each one of them.

Think of this as your pipe. Care goes in one end, cash is supposed to come out the other. Every coin is money you already earned. Admitting blind, coverage disappears, notes don't prove the need, nobody fights back.

Leak number one. Somebody calls at 11 PM ready to come in tonight. Your team wants to say yes, and they should, but nobody checks the benefits. Three weeks later, you find out the plan needed an approval first and you never got one. Or you find out it was out of network and it's a fraction of the money collected. Checking benefits used to mean somebody sitting on hold with the insurance companies. AI handles that kind of stuff right now, and it does it really well.

So I put together a simple sandbox version, fake patients, fake data. It's late at night and a caller wants to come in. You've got a plan and a member ID, and that's all you need. We click on benefits and see what happens with this caller when we check benefits. We get some gobbledygook from the insurance company, but we get a clean output here that we're able to read. It pays 80% after the deductible. $1,800 of the $3,000 deductible is still left this year. IOP needs approval from the plan before the first day, so a person needs to check. We click review. It's reviewed by Dana in admissions at 11:06 PM, now calling for approval. Solid.

Let's take a look at one of these other callers. This is a little more of a problematic caller. The coverage is in, but it's out of network. The plan is covered, but it's out of network, so the family is going to be on the hook for the difference, and we need to make sure that's communicated. This is something that can happen from time to time.

Let's run this one. Coverage ended September 30th. The plan shows the renewal paperwork was never filed. Nothing would be paid for if care started tonight. Coverage can often be restored once the paperwork's filed, but this is going to need a person on it immediately. Let's get somebody to help with the re-enrollment. We can mark this as reviewed, clean and clear.

So that basic check comes back in seconds. The details still take a phone call to the insurance company. There are a lot of tools out there that can do the full version of this today. I put two of them on the screen right here. One of them plugs right into a CRM a lot of people already use. It answers your admission calls after hours, and it verifies your insurance. Simple rule: verify before you admit, every single time.

All right, leak two. This is the January problem. Starting next year, Medicaid patients can be covered on day one and then not covered on day 40 because the form didn't get filled out correctly. Nobody catches that, and you may be treating them for free. So the fix is an AI agent that re-checks coverage on a schedule. Once a week is plenty.

It's Monday morning next February, with 12 patients in the program this week. Let's see what happens when we click run. We get some flags. This one's active, but paperwork is due, so this gives us notice that we have to pay attention to it. Same thing here: active, but paperwork is due. And today we have one where the coverage has ended. Call today: three visits since January, and they're unpaid. You get a nice little note over here: two patients still need to re-qualify, one patient's coverage ended January 26th, all three may qualify for addiction exemptions, and nine patients are still active with nothing to do. Mark it reviewed, clean and clear.

So that protects your revenue. More important, it keeps patients from falling out of treatment over paperwork. Ask your eligibility tool or biller if they can run checks on a schedule. If they can't, it's a simple agent to build.

All right, leak three, your notes. Colorado ran an audit on this. They pulled Medicaid denials for residential treatment, and 86% came back with the same reason: the chart didn't prove the patient needed to be there. And the auditor agreed with the insurance company almost every single time. The paperwork just wasn't there.

The fix starts in the session. There is an AI scribe built for behavioral health. It listens with the patient's consent and it drafts notes. It even handles group notes, and your clinicians still review and sign them. Then there's the weekly review call. You know it as concurrent review. This is where the insurance company decides if they want to pay for another week.

Same sandbox, one patient, one week of notes, all fake. Let's run this and see what happens. We have a bunch of notes over here on the left done by Jordan. Let's run those notes through the system. These are the things we're going to fix before the call today. The treatment plan was last reviewed in August, and it wasn't updated after the patient was caught drinking on August 15th. No withdrawal checks were written down after August 15th. The August 19th group note is one line. It shows attendance, but nothing else. We mark it reviewed by Sam K. at 3:40, chasing the signature before the call, and it shows which notes to go fix. It organizes what's already in the chart. It never fills in what's missing. And the clinician still decides the level of care.

All right, leak number four, and back to that 16 cents. This next part is what the lawsuit says. The claims went to an outside company. Its job was to lower the payment, and the lawsuit says it kept a cut of whatever it saved. When the story ran, Cigna had not responded.

So think about your own center. In my experience, most centers never catch an underpayment, because nobody compares what got paid to what was promised, claim by claim, every month. This is a perfect job for AI. Forty claims, two months, and again, everything's fake. The one question is: did we get paid what we were promised? This is a big one, and this is where a ton of money leaks.

Let's run this. $7,127 short. Seven of the 37 paid claims, all from Cascade Ridge. We can look through these on the left. These are each of the short-paid claims, and this is the difference in the amount. Some of these were straight denied. So we had 30 claims paid in full, seven short payments and three flat-out denials. Those three denied claims were worth another 2,500 bucks. One of them is Jordan M.'s weekly treatment, denied as not medically necessary. So let's draft a quick appeal for this one. Just like that, we have an appeal drafted with the claim number. We read through it, approve it and sign it. Bam. Now the appeal can be sent to the insurance company. There are a lot of tools that already do this for hospitals, so ask your biller if they have something like that.

Is it worth it? There's a big lawsuit right now against UnitedHealth. The patients suing say that nine out of 10 people who appealed a denial won. That's their claim, and that case is about Medicare, which is a different kind of care. But the point holds: you can't win appeals you never file.

And almost nobody is using these tools yet. A survey asked 250 billing people about AI. Two out of three said they know it well, but one in seven actually used it. Most of the ones who did say their denials went down. Big operators are already getting it. One of the largest behavioral health companies in the country told investors it's rolling out a new billing technology, and its CFO said AI let them reduce headcount.

Before any of this touches your records, two rules. Rule one is privacy. There's a federal rule for addiction records called Part 2, and it's been enforceable since February. AI vendors that touch patient data have to sign a BAA. That's a HIPAA contract. And they have to handle Part 2 consent. If they look confused when you say Part 2 consent, walk away. Rule two: AI drafts, a human decides. No AI picks the level of care, and no appeal goes out without a person signing off on it.

So if you run a center, here's what I'd do on Monday. First, I'd ask your biller four questions. How many claims get denied the first time? What percentage of those denials do we appeal? How many of those appeals do we win? And how many days does it take to get paid? If they can't answer that, that tells you something. Then I'd turn AI benefits checks on for every admission. That one pays back the fastest. Before January, I'd set up coverage re-checks for Medicaid patients. After that, the scribe and the appeals.

The insurance companies already have software on their side, and you should too. AI is a force multiplier, but it's not magic. So use it to collect every dollar you've already earned. Comment PAID and I'll send you those four questions for your biller. And if you want me to map this out for your center, there's a free roadmap link below. Go ask those four questions this week. Those answers will tell you exactly where your money's leaking.