Expion Health revamps its RFP process with AI

Process automation and improvement is a perennial CIO agenda item, and the call for business process optimization is only getting louder — especially for those processes directly tied to the bottom line.

For Expion Health, a cumbersome manual process to determine what rates to quote to potential new customers had become a cap on the healthcare cost management firm’s ability to grow its business.

On behalf of insurance carriers, pharmacy benefit managers, and other healthcare payers, Expion negotiates prices with pharmacies and medical practices based on volume discounts and other factors. The company typically gains new customers through requests for proposals, but responding to an RFP was a slow process involving detailed analysis by a trained underwriter performing calculations on a large volume of variables in an Excel spreadsheet.

RFPs are typically requested every time a health insurance carrier, pharmacy benefit manager, or third-party administrator begins to cover a new employee group. In response to an RFP from a pharmacy benefit manager, for example, the underwriter would need to project future medication use in the covered population, wholesale cost increases, potential patent expirations on medications, and cost changes in medication rebates. Calculations not only looked at past medication use among the covered population, but also involved predictions about several factors that will affect drug prices in the future.

That process typically took six to eight days and was subject to miscalculations that could potentially cost Expion millions of dollars because the company guarantees cost savings for customers.

“Underwriters used to have at least 10 versions of the Excel sheets, and then they’re pumping this data into the Excel sheet and just doing the math,” says D.S. Suresh Kumar, chief transformation officer for mergers and acquisitions at Expion. “We take the financial risk for this, which means that if there is anything that’s misrepresented, the money comes from our pocket.”

And with just six underwriters on staff, Expion could respond to only about 200 RFPs per year, limiting the company’s ability to bring in new business.

Machine decisions

In August 2023, the Expion IT team, with support from executive leadership, launched an internal project to automate the complex underwriting process for prescription cost management.

In about six months, Expion’s two AI developers created ExpionIQ Advisor, a tool that uses linear regression, multiple algorithms, and custom-built AI models to automate prescription RFPs. The IT team plans to further enhance the application using the XGBoost machine learning software library for forecasting medication use in covered populations.

ExpionIQ Advisor has cut the time to calculate the numbers needed for an RFP to a few hours, instead of days, Kumar says. The new application also allows employees to check its work, he adds.

“Now I know exactly what math we are using and what risk are we taking, and even if the model is predicting something wrong, we know exactly what to do to go fix that issue,” Kumar says. “In the past, anybody could make a mistake easily. Now, if the model is drifting away, we get alerted automatically, and we fix it.”

A web-based .NET application on the front end of ExpionIQ Advisor allows underwriters and other users to log in to the system and check the numbers. “The .NET application brings it all together and does the final computation to present that data in an easy-to-digest manner as well as provide a printout to our end customers,” Kumar says.

New revenue potential

While ExpionIQ Advisor is just past the pilot stage, the project is already profitable, Kumar says. Expion hasn’t yet calculated the potential new business created, but the tool will save the company the cost of about 1.5 data analyst FTEs.

In addition, several of Expion’s customers have expressed interest in licensing the application for their own healthcare-related projections, Kumar says. Expion projects about $300,000 in sales this year.

“External clients are now asking, ‘Why don’t you give this to us?’” he says. “Most of the companies have this entire [underwriting] infrastructure in Excel sheets. Nobody has a clean, organized way to do it.”

The ExpionIQ Advisor project has earned Expion Health a 2024 CIO Award for IT leadership and innovation.

To complete its work on ExpionIQ, the IT team spent an “enormous amount of time” with the underwriting team to better understand the intricacies of the various underwriting processes in practice, Kumar says. As a result, Expion AI developers were able to tune ExpionIQ Advisor to handle a range of underwriting scenarios and variables, including drug utilization, insurance exclusions, and past RFP responses. The two teams also conducted focus sessions to validate the logic and finalize the model, he adds.

Expion’s AI developers also made use of data modeling techniques and machine learning algorithms to ensure ExpionIQ Advisor could make decisions even when addressing incomplete data sets. In many cases, clients provide Expion with partial prescription claims from the previous year to generate new price quotes, but the company needs to use multiple factors to determine pricing, Kumar says.

“Though clients imply their utilization will be the same, it never is the same,” he adds. “ExpionIQ Advisor factors in a few factors to project a reasonable estimate.”

Automating complex projections and other calculations in the insurance industry is a vital undertaking, says Jeffrey Rivkin, research director for healthcare payer IT strategies at IDC Health Insights. The Expion project demonstrates a new automation use case to him, and he sees the benefits of speeding up the RFP process.

“Something like this that gets you a number in a cost-savings situation and saves the auditors from doing the number crunching makes a lot of sense,” he says.

Automation, and generative AI in particular, can transform the insurance industry, he adds. Insurance companies can use AI to summarize long medical charts, to classify documents, and to find patterns in unstructured data, he says.

In the Expion case, the company is “cranking large sets of data to make complex decisions,” he says, with the potential for great benefits.