A Free Startup Idea For You

Using AI to Avoid Bankruptcy From Medical Debt

Here's a free startup idea for you:

I've been thinking about this idea for some time, but I'm already preoccupied with my current project. I also don't have the specific domain expertise to fully execute this idea. Nonetheless, I think it's a large opportunity that, in the right hands, could be both profitable and massively beneficial for consumers.

As you may know, medical debt is one of the largest causes of personal bankruptcy in the United States. Quite obviously, the system is overtly complex with unnecessary administrative costs embedded within the system. Those costs are passed directly on to consumers.

Even worse is that healthcare is one of the few places where you know what you owe after you purchase the product or service. And even that initial price isn't set in stone. But even if the insurance company won't budge, consumers have options.

That leads to the idea. While the numbers aren't exact, it's estimated that billions of dollars of healthcare claims are denied every year. A 2023 KFF study estimated a whopping $260 billion in inpatient claims that were denied. Regardless of the precise number, healthcare companies are in the business of denying claims.

But consumers can fight back. They can do so through the appeals process.

With that said, it isn't easy. The appeals process is extremely complicated and can dramatically vary based on your health insurance plan and state regulations. It's also terribly old school (think faxing medical information to healthcare companies).

Because there is so much friction, patients often don't appeal. In fact, fewer than 1% of denied claims are appealed every year.

Therein lies the opportunity. With well-constructed prompts trained on the legally dense plan documents that patients often don't read, LLMs can demystify the appeals process. They can evaluate the denied claim, understand the appeals procedure for fully insured versus self-funded plans, and even gather studies that would support a successful appeal.

I'm not saying this would be easy. ProPublica has done some fantastic work on how complex the appeals procedure is. That said, appeals can be successful. There are some fantastic stories of six-figure claims paid in an excellent book called Never Pay the First Bill (thankfully, I haven't had to experience this firsthand).

Ultimately, let me know what you think! It's just another way that we can use LLMs to take large sets of unstructured data, find unique insights in that data, and make people's lives better.

Prompt of the Week

Using this theme of business ideas, I wanted to test ChatGPT and ask it about contrarian ideas that are hiding in plain sight. For this prompt to work, I had to have it use context about me and my prior chat history.

This may be more of a meta prompt, but I was curious how an LLM would respond—considering that it is trained on data from the Internet. Try it out for yourself!

"Considering what you know about me (specifically, my areas of professional expertise), I want you to give me some contrarian ideas that are hiding in plain sight. No idea is too controversial or off the wall. Importantly, these ideas can't be contrarian for the sake of being contrarian. They need to be contrarian and have a high probability of being right."