Reimagining Medicare Advantage plan design using AI

As private insurers have entered Medicare Advantage (MA) over the past decade, seniors now confront a dizzying volume of plans to choose from—more than 40 for the average Medicare-eligible U.S. resident. Meanwhile, payers face significant headwinds in the form of increased utilization and growing rate pressure from the Centers for Medicare & Medicaid Services (CMS), forcing payers to do everything they can to make sure their plans stand out in the marketplace.

At the same time, we know that benefit design drives roughly 30% of annual enrollment shift in Medicare Advantage plans, putting product design teams on the hot seat to devise benefits packages that drive member growth and retention while also improving plan-level profitability. AI offers significant promise to help them do so more efficiently and, most importantly, more effectively.

While it’s true that Medicare Advantage providers are already using AI here and there to improve plan design, they’re currently missing the biggest opportunities to build competitive advantage through these new technologies. This is because health plans, like others across the healthcare ecosystem, have so far approached AI with the mindset of greenlighting a hodgepodge of pilots in service of boosting productivity and breeding comfort with AI.

Creating true competitive advantage through improved Medicare Advantage product design, however, will require a more coordinated approach—one that combines generative AI, classical AI and traditional analytics into a string of interconnected use cases that add up to a harmonious whole.

MA product design is made up of four main categories of interrelated business processes connecting product strategy, product design and activation. To reinvent how health plans design products requires breaking the overall process into its constituent parts and individually reengineering each with a view of the whole. Here’s how:

Category 1: Product strategy - Portfolio optimization

With local Medicare Advantage markets becoming increasingly heterogenous, health plans must start by deeply understanding customer segments in each market and then tailoring their portfolio of MA plans to these segments’ known preferences. Today, this process is more art than science. Here’s how it can be improved using a mix of traditional analytics and classical AI.

• Align plans to key county-level segments. MA organizations have been busy adding new plans with incremental features, creating confusion among members and insurance agents alike, leading to slower sales and higher administrative expenses. Advanced analytics can identify counties with overlapping plans or plan gaps, alerting product design teams where to focus their time and resources.
• Simulate the effect of adding or dropping plans. For counties with coverage gaps or plans targeting the same customer segment, classical AI can remove uncertainty by simulating the portfolio-level impact on enrollment and retention by generating what-if scenarios based on adding or dropping plans.

Category 2: Product strategy - Product strength mapping

To know where to focus attention, MA providers must understand how their current products compare with those of their competitors. Today, this process relies on a mix of classical AI, advanced analytics and a whole lot of manual work. Using generative AI to curate unstructured data not only can save time but also can significantly improve classical AI models that quantify relative product strength.

• Use classical AI to identify which benefits drive enrollment for specific demographics. Traditionally, when insurers seek inputs to drive product design, they rely on a mixture of subjective intelligence from insurance agents, sales teams and syndicated market reports to gauge member preferences. Today, they can use product and enrollment data published by the CMS, as well as county-level social drivers of health (SDOH) data, to power models that pinpoint benefits and other variables that drive enrollment for specific member segments.
• Compare plans using generative-AI-curated data sets. Knowing which benefits drive enrollment and retention allows carriers to compare their plans against those of their competitors in each geography. To do this, they traditionally have used county-level contract and enrollment data published by the CMS to find every plan in each county, download the summary of benefits and coverage for each and then manually compare them. Today, they can use generative AI to pull all the relevant benefits and coverage data from publicly available documents and then use this data to build a side-by-side comparison view. This can then be used to compare plans much more rigorously than was possible before.
• Use advanced modeling to see where your MA plans rank in each geography. To aid decision-making, executives want to see a visual representation of how their plans rank compared with their competitors’ plans in each geography where they are present. Using data on the relative impact of every coverage and benefit variable on enrollment, companies can then use classical AI to map company strength at the county level. This identifies those counties where a company is positioned to win—and those that need attention.

Category 3: Product design - Plan design simulations

Once executives have a clear picture of the relative MA plan strength in every geography in which they operate, it’s time for product designers to adapt plans to optimize their portfolio for each county. This involves making difficult trade-offs related to benefit choices, a process that has been largely ad hoc up until now.

Using AI models, however, allows product design teams to simulate an infinite number of MA plan design iterations to arrive at the optimal plan design in counties targeted for attention. By adjusting metrics like financials, plan type and supplemental benefits, AI can project the expected enrollment lift for every combination of coverage benefits, helping designers make informed decisions about the following year’s CMS bids.

Category 4: Activation - Co-pilots for information gathering and internal and external communication

With the proliferation of MA plans, members aren’t the only ones overwhelmed by the volume of plan information they need to sift through to make informed decisions about their coverage. Product designers, insurance brokers, inside sales and senior executives also need to be able to drill down to understand the specifics within a health plan’s portfolio. Generative AI can help.

• Chatbots for plan queries. Whether it’s a product designer informing their work or a broker sitting with a member shopping for an MA plan, insurers need a way to make complicated, detailed plan information available to anyone in the most frictionless way possible. Using the structured database of plan information already scraped from the CMS and rival plans’ benefits and coverage summaries, companies can build intuitive user interfaces on top of these inputs that deliver detailed plan information that can inform users as they make decisions.
• PowerPoint co-pilots. Product and analyst teams spend a lot of time producing decks to align internal stakeholders around a course of action. Co-pilots can take analysis graphs or commentary and import these directly into PowerPoint, saving teams from the time-consuming process of creating first drafts of PowerPoint slides and removing some friction from leaders’ decision-making process.
• Delivering personalized insights. With so many MA plans and plan providers in every geography, insurance agents and broker relationship managers can benefit from personalized insights highlighting specific motivators for brokers. Similar to “next best action” triggers that pharmaceutical companies send to field sales reps, these nudges can surface specific plan features most likely to appeal to a given broker based on their past preferences, the demographic of their customer base or both.

Many MA carriers today have the technical capabilities to build and apply classical AI, generative AI and advanced analytics to the product design process. The secret sauce, so often lacking in any change management exercise, is in how companies build and deploy tools that ease comfortably into existing workflows and therefore encourage adoption.

Consider the example of the algorithm that pushes alerts directly to sales teams so they can surface personalized insights for insurance agents or members in the field. The model that generates insights is only effective if the sales teams find its outputs valuable and put them to use. These simple guidelines can help:

• Build tools directly in the platforms your target audience is already using. If you force users to create new ways of working by having to go outside their existing platforms, you create friction and hobble adoption.
• Include a mechanism for users to offer feedback on tools’ outputs. If they don’t find them valuable, they won’t use them.
• Feed quantitative and qualitative data back into models. By continually refining models, they improve over time and become even more valuable to users.

Given the speed at which AI is advancing, models themselves are quickly becoming commodified. The value is in how companies choose to apply them to solve larger business problems and the degree to which companies can embed them into existing workflows. The health plans that most effectively apply these lessons to the critical issue of MA benefit design will deliver key advantages now and in the future.