How Berkeley Haas research fueled a company that could save Medicare patients from costly mistakes  

Choosing a Medicare plan is both complicated and consequential. Berkeley Haas research has fueled a new company that simplifies the process, promising to save money and improve health for millions of people.

Photo of a woman with gray hair in a ponytail smiling as she looks at a laptop screen showing the Healthpilot website.
Photo courtesy of Healthpilot

It’s Medicare open enrollment season, and the tens of millions of retirees who rely on the government program for health care are grappling with an abundance of options, a dearth of information, and no good way to personalize or compare plans.

“It is not a well-functioning market,” says Jon Kolstad, an associate professor at UC Berkeley’s Haas School of Business who studies the economics of health care. “And yet the choices people make have high financial stakes—consumers are typically on fixed incomes—and critical health implications.”

Kolstad has been studying this challenge since he was in graduate school and, with several academic collaborators, recently helped turn a broad foundation of research into a company centered on helping people make better decisions when choosing a Medicare plan. Healthpilot, which launched in late 2020, uses machine learning to compare Medicare plans and suggest options that are personalized to each person’s current circumstances, projected health needs, and risk tolerance. It’s also free to use.

“We recognized the power of this highly predictive algorithm developed to solve a critical unmet healthcare need for seniors who are evaluating Medicare plans,” says Healthpilot CEO Seth Teich. “We believe that Healthpilot’s platform is a transformative technology that empowers consumers to easily navigate through the complexity of plan choices to find, and enroll in, their best coverage option.”

A desperate need for innovation

People with questions about Medicare enrollment have long relied on phone calls to private agents who walk them through the decision. But there’s a problem: “In reality, these brokers, who are supposed to be experts, do no better at selecting a plan than the average person,” Kolstad says.

In a 2021 paper, Kolstad and several colleagues, including UC Berkeley economist Ben Handel, demonstrated that brokers are prone to the same flawed judgment as everyone else. For instance, they place too much weight on a plan’s premium while overlooking other costs, such as out-of-pocket expenses. The result is that consumers working with brokers pay $1,260 more per year on average than they would if they enrolled in the best plan.

This is not the result of brokers’ bad intentions but simply because solving which plan is best for which person “is a very complex computational problem,” Kolstad says. In fact, the 2021 paper found that when brokers were provided with an AI assistant to help them suggest a plan, they saved consumers about $300 per year. (This is a conservative estimate.)

Healthpilot’s promise lies with its ability to use AI to properly weight the many fixed and projected costs of every available plan, sifting carefully—and impartially—through these multidimensional relationships. The algorithm operates by comparing every Medicare enrollee with millions of similar people and then forecasting the likelihood of different medical complications. By pairing this forecast with known information—including, with users’ permission, secure access to the medications people take and the doctors they see, along with information they provide directly—Healthpilot then determines which plan is ideal and ranks the alternatives.

The algorithm also considers individual appetites for risk. “Some people have very little risk aversion, and they would rather have low payments now and gamble on how they fare,” Kolstad says. “The plan that gets recommended to this kind of person should be different than the plan that gets recommended to someone who is very risk averse, who wants a high premium now in order to know that they’ll be covered.”

Benefiting individuals and the marketplace at large

The financial benefits for individuals are straightforward: Choosing the right Medicare plan generally means better coverage and less expense. These savings also accrue to the federal government, which finances Medicare.

A subtler benefit are gains in well-being and even lower mortality rates. One working paper co-authored by Kolstad found that people who have to pay more out of pocket cut back on important health care services. A related paper, co-authored by Ziad Obermeyer of Berkeley Public Health with researchers at Stanford and Harvard, found that a $100 bump in per-month cost-sharing for drugs—exactly the kind of mistake people make without good guidance—increases mortality by 13.4%, as people forego essential drugs such as blood pressure medication.

Healthpilot is able to deliver these financial and health-related benefits to consumers for free because of the structure of the Medicare market. Since Medicare is a valuable source of revenue for insurance companies, the companies pay commissions to brokers for each person that they enroll. If Healthpilot sends someone to Humana, Humana pays; if instead the enrollee goes to Blue Cross-Blue Shield, then Blue Cross-Blue Shield pays.

That’s a crucial point: Because these commission amounts may vary by carrier, human agents may be biased in their plan recommendation based on the commission they are paid. Healthpilot’s algorithm does not factor commissions into its recommendations and does not steer people toward any particular plan or company based on financial incentives. “There’s no distortion in the platform or plan recommendation, which is unique in the industry,” Kolstad says.

This also has the potential to inspire greater innovation and efficiency in the insurance market as a whole—one of Kolstad’s main interests. As a point of comparison, consider the tech market: When a company like Apple creates a product that people like, they buy it; when it creates a product people don’t like, they don’t buy it. This is quickly reflected in the company’s revenue and share price.

Because insurance products are so much more complicated, consumer decisions rarely reflect clear notions about quality; people often enroll, and stay enrolled, in plans that don’t deliver value. Healthpilot, by sorting people into plans that genuinely benefit them, could bring much greater transparency into the marketplace and produce meaningful information for companies to build better plans, Kolstad says.

“For better or worse, we rely on competing private plans in Medicare. That’s the approach we’ve taken because we believe that a private market will offer innovation,” Kolstad says. “Giving customers a greater ability to match with plans that give them the coverage they want and need will reward innovators. That means Healthpilot isn’t just a digital enrollment solution for consumers but can be a tool to make the whole market function more as it should.”