Health insurer Oscar unveiled its new artificial intelligence chatbot, Oswell, last week. The bot is available to Oscar’s doctors, who can use it to take notes and retrieve patient data, and its members, who can access on-demand support.
Oscar Health started experimenting with using member-facing artificial intelligence agents in August, when it launched its “superagent,” which could answer basic questions about an individual’s insurance plan. But now the company has expanded the bots’ capabilities.
“Oscar needs to be able to deliver better outcomes at a lower price, and you should have AI that helps you with clinical issues,” Oscar founder and chief technology officer Mario Schlosser told Ruth.
Oswell can answer questions about members’ health history, explain test results, request prescription refills, and recommend specific doctors or practices for specialty care.
Schlosser sat down with Ruth to discuss why health insurers should have AI bots for their customers and what he thinks those bots will be able to do in the future.
What’s the difference between Oswell and more general-purpose chatbots like ChatGPT and Claude?
The two things that make Oswell very different than using any other AI tool out there, or Googling for that matter, is that one, we have a lot of context about you: claims, lab results, the drugs you’re taking, diagnosis codes, medical records — all these things that we have about members — the conversations you’ve had with our telemedicine doctors and so on.
And then the other big differentiator is we know how we pay for your health care. We know who’s in the network. We know the doctors that are best for the particular issue you have. We know how much it will cost you to go to those doctors. We have our own virtual primary care physicians and our own nurses. Oswell knows what you talk to customer service agents about.
And why is that good?
A good example is financial pressures. We have members call Oscar and say ‘I cannot pay my bill this month because I don’t have enough money.’ And that does not show up in any claim or any medical records, but it is on the record with Oscar, and so that is a very useful piece of information to have because these things matter for your health care outcomes. If you’re constantly stressed about your finances and you can’t pay your insurance bill, that may be tied to mental health. It would be tied to [why] you didn’t pick up a drug and why you’re not feeling well. And so I think that’s a really good thing, that the bot has that information, because it’s part of your picture.
I think that’s the promise of AI, that it can actually aggregate a lot more about you and then hopefully make [better sense of it than] the health care system as a whole can do that right now.
How do you want to use AI in the future?
One of the things I absolutely want to do is, we have a number of clinical guidelines [for determining whether a service or drug is medically necessary]. When the nurse gets a medical authorization request, the medical record comes in, and then the nurse has to fill out these criteria. Oscar AI [could] refill those criteria.
So the nurse can look and say, ‘Oh, the member didn’t have physical therapy yet. And it says it here in the medical records the member did take this drug.’ Now I can go through and decide, is it, based on the criteria, an approval or denial? We only have that for a couple of clinical guidelines right now, but that enables us, in theory, to simply say, let’s have that whole process done automatically.
