Hospitals are buying billing software that decides, patient by patient, what to offer. The software looks at what it knows about the person and the bill, predicts how likely the person is to pay, and picks a discount sized to get the bill paid. A chatbot then sends the offer and answers questions, so many patients now bargain with software instead of a billing clerk.
The discount is an offer built to work
One billing company, Cedar, began randomly giving different patients different discounts in 2020 and watched who paid. It found that a discount offered when the bill arrives gets more people to settle without lowering the total the hospital collects, and it trained a model on those results. The model gives a discount to 86% of bills of $800 or more and to 12% of smaller ones, and patients from poorer zip codes get discounts more often.
That is real help for many people, and it is also not generosity. The model's test is whether the discount gets the bill paid, so the discount is the smallest one expected to work. A patient who would have paid with a smaller discount, or who qualifies for a much larger one, never sees the difference.
Why this will spread
Hospitals have a reason to buy it. More than one in four US adults owed medical debt in 2025, and 26% of people with insurance were among them. With millions of people expected to lose coverage, hospitals will provide more care that nobody pays for, and a tool that raises collections without hiring more staff will be easy to approve.
The other half of the bill is also changing. Insurers argue that AI note-taking and coding tools are already raising what hospitals charge, and hospitals dispute it. So the amount on the bill and the discount taken off it can both be set by software, and the patient is the only one at the table without software.
What the airline story says about what comes next
American's advantage did not last. Once competitors caught on, the same method was available to everyone, and the passenger's side learned to shop around. I expect the patient side to catch up the same way, and sooner than it did for flights, because the hard part is reading paperwork, which AI assistants already do well.
A family could ask the hospital for an itemized bill, paste it into an assistant, and ask which charges look duplicated or unusual. They could then ask it to draft the financial assistance application from the hospital's own published policy. At a nonprofit hospital, a patient who qualifies cannot be charged more than what insurers generally pay, and applications are usually accepted for at least 240 days after the first bill. A text offering 20% off, with a deadline on it, does not change that.
The airline lesson carries over: the first number you see is what a model thinks you will accept. Ask for the itemized bill and the assistance application before you answer it. These rules cover nonprofit hospitals; for-profit and government hospitals follow state law and their own policies, so ask them what they offer.