What AI Finds When It Reads Your Benefits Contract

An AI model can read your benefits contract in minutes and tell you where you’re getting worked over. Mark Cuban does this himself, in front of CEOs, and he’s convinced it’s the single highest-return use of AI available to any company in the country right now. 

Mark Cuban guest-hosted Pivot last week, and somewhere in the middle of an hour on healthcare, he made a claim I haven't stopped thinking about. He said that for every company in America, the single highest-return use of AI available right now, the one that pays for itself immediately, isn't in recruiting, support, or code. It’s working with AI and an AI partner to review your benefits.  

He does it himself via one of the leading-edge large language models. Uploads the contracts, asks one deliberately blunt question (how am I getting F__ked), and the answer comes back showing at a high level where the company is being worked over. He's walked into some very large companies and done this in front of their CEOs. His read is that if executives just did this and acted on what they found, the whole industry would change overnight.

I've read a lot of "AI will transform HR" takes. Most of them are vendors explaining how AI will transform your budget. This one is different because benefits are the second-biggest line item most companies carry, and almost nobody has ever looked at it properly. 

Why hasn’t anyone read these contracts? 

Here's the line that made me laugh out loud. Cuban on why these documents go unexamined: they run 50, 100, sometimes 200 pages, and "there's no human that has enough coffee available to them" to get through it.

He's right, and it isn't an accident. The confusion is the business model. More information asymmetry means more leverage, which means higher prices. He points out that PBM (Pharmacy Benefit Manager) pricing is locked down so tightly that when Elizabeth Warren asked what Express Scripts charges the military through Tricare, the answer was a flat no. Proprietary. Cuban compares it to Fight Club. “The first rule of PBM contracts is that you don't talk about PBM contracts.”

And my favorite detail: when he's run these through an LLM, he's found actual errors. Misspellings. One section says 36% while another says 46%, at a multi-hundred-billion-dollar company, and nobody caught it. These aren't sacred documents. They're long documents, which turn out to be a very different thing.

If you ever watched the “Big Short”, this is where we would shift to a movie star, singer or celebrity chef to explain what a PBM is for the reader. So… 

What is a Pharmacy Benefits Manager (PBM)? A Pharmacy Benefits Manager is basically the middleman's middleman: a company that sits between your health plan, drug manufacturers, and pharmacies, negotiating drug prices and deciding which meds get covered - while somehow keeping their own cut a mystery to everyone involved.  

In the "benefits stack," think of them as the bouncer at the drug counter: your insurance carrier hired them to keep costs down, but they also collect rebates from the same drug companies they're supposedly negotiating hard against. So PBMs are less a helpful referee and more a toll booth that also owns stock in the highway. 

What's actually hiding in the fine print? 

The trick Cuban describes is worth understanding even if you never do anything about it.

The body of the contract makes a clean, generous promise. Something like: we pass through 100% of rebates. Sounds great, easy to nod along to in a renewal meeting. Then the definitions, sitting somewhere else in the document, quietly establish that "rebate" doesn't include specialty drugs or a handful of fees. Which, as Cuban puts it, is all of them.

Then the part that actually matters: the PBM decides what counts as a specialty drug. 

So the promise is real, it applies to almost nothing, and the other side holds the pen on which nothing. Nobody stumbled into that. It's built. And it survives because it requires reading page 12 against page 140 and holding both in your head, which is precisely the task humans are worst at and machines are unreasonably good at.

Worth one caveat here, and Kara Swisher caught it in the interview. The person and the prompt matter enormously. Cuban agreed. AI is remarkable at reading documents and terrible at being trusted with arithmetic. If a model hands you a savings figure, that number came from a prediction, not a calculation, and you should not put it in a board deck. Make it show you the clause and the page. 

Why does this land on HR? 

Cuban said something about HR that I've been chewing on for days.

He said HR departments, tech especially, are chronically understaffed. So, when an employee's claim is denied or they're trapped in a pre-authorization loop, they go to HR. And HR, by his estimate, spends most of its time on exactly that. This leaves no capacity at all to ask whether the plan is any good in the first place. As a former Chief People Officer who valued data and was always AI-first, this one hit me hard. This is not a knock on HR. It is a diagnosis.

We've built People functions to absorb the pain the healthcare system generates, then act surprised that nobody's auditing the contracts. You can't run a claims-denial help desk and a vendor audit out of the same two people. Something gets dropped, and it's always the one nobody's yelling about today. 

Why won’t companies act on what AI finds? 

Here's Cuban's explanation for why large companies won't act even after he shows them millions in savings.

They're afraid of their employees. Not the vendor. Not the contract. The employees. Because acting on it might mean changing the logo on the insurance card, and nobody wants that fight. For once in tech history, he says, they're scared of their own people.

I don't love the framing, but I don't think he's wrong. In fact, I have personally experienced outrage, frustration, and sometimes backlash from employees when I changed healthcare benefits for what I thought would be a better, cheaper experience. And it's worth naming, because the fear is mostly misplaced.  

The biggest money in these arrangements isn't in the network at all. It's in the contract terms, the claims that were paid incorrectly, and the dependents who were never eligible. None of that touches anybody's card, doctor, or experience. Whoever's telling you this is a disruption conversation is having the wrong conversation. 

Can AI actually get the money back? 

Now the honest part, because I think Cuban skips past it a little.

An LLM reading your contract gives you a suspicion. A very good, very specific, deeply uncomfortable suspicion. What it doesn't give you is a dollar savings.

Turning that into money means auditing claims across the full population rather than the 250 the TPA sampled. It means knowing what market terms actually look like so you can tell whether your deal is bad or just ugly. It means a re-procurement or a renegotiation with real leverage behind it and documentation that holds up when someone asks how you made the decision. Cuban's own success story is telling here. The company that saved millions after taking his advice still had to run an audit to collect the refund. Somebody did that audit, and it wasn't the chatbot.

That's the gap. The reading got cheap. The rest of it still takes people who've seen fifty of these.

One more thing worth saying plainly: none of this replaces your broker. A good broker's value was never in reading the document. It's in the market relationships, the carrier leverage, and the negotiating room they can create that a contract review alone can't touch. AI can tell you where a contract might be working against you. It can't get you a better renewal, and it can't be the relationship that gets a carrier to actually move. That part still takes someone who's done fifty renewals and knows exactly who to call.

That second half is the work we're building at livingHR. Your Form 5500 is public, and we (with the help of AI agents we built) can usually tell you something interesting about your plan from that alone before you send me a single confidential document.

So read the contract. Genuinely, go do it, with or without us. 

Just don't stop at knowing.