When payment integrity provider Milliman wanted to improve how it detects fraud for its healthcare payer customers, they collaborated with Mastercard® Healthcare Solutions to conduct a pilot using Mastercard’s proven AI solutions. They had no idea how successful it would be.

Healthcare insurers, payment companies and employers across the globe work with Milliman Inc. to improve their healthcare systems, manage emerging risks and advance financial security. The company uses data-driven insights and subject matter expertise to help transform health and financial security for organizations.  Among those services, Milliman’s payment integrity saves its clients millions of dollars each year. Always striving to do better, Milliman wanted to introduce AI technology to boost its success.

Milliman’s history of healthcare fraud detection

Using the Milliman Payment Integrity tool and their MedInsight data warehouse and analytic solution, the Milliman team analyzes 100 percent of its clients’ claims data, including medical, dental, vision and pharmacy claims. With the Payment Integrity tool, Milliman is looking not just for individual claims errors, but patterns of claims billing and other aggregations that lead to cost savings for their clients.

“We have reporting on things we could test directly like duplicate claims, or more complex situations such as the unbundling of claims, or benefit related issues, all with the ultimate goal of earning more reimbursement by providers,” says David Cusick, Milliman principal. “We knew back then, as we know now, there is between three and 10 percent of total claims expenditures that could be considered fraud-based reviews, depending on who you ask.”

The real innovation in the Payment Integrity tool was not just the testing, but the combination of existing testing with MedInsight’s reporting capabilities and their experience in focusing on the important areas to support clients’ success.  This laser sharp focus was aimed directly at the three to 10 percent unnecessary spend.

In a recent webinar, FWA: Payment Integrity Powered by Mastercard AI, Cusick said that Milliman engaged in a pilot with Mastercard® Healthcare Solutions to solve a problem that is difficult to solve without using AI. “The problem is simply not knowing what you already don’t know,” he says.

AI: the next step in detecting healthcare abuse

“Bad (healthcare) providers are always learning what schemes will be detected and then tailor their behavior to avoid that detection,” Cusick explains. “Using artificial intelligence to find new schemes that are yet to be found through manual auditing or otherwise takes a different mindset, because the problem is not the same as with traditional detection methods.”

He adds that bad actors are committing more organized healthcare fraud than ever, taking advantage of governmental programs, telemedicine and COVID-related claims. The events of the past year have put added pressure on payers and payment integrity providers to prevent and detect complex, well-executed schemes.

Milliman brought Mastercard two challenges: help identify the growing fraud problem while implementing technology that integrates with the Milliman Payment Integrity tool.

An AI model customized to detect health provider fraud


Using historical data from a regional health plan (with consent), Mastercard’s AI team built an “ensemble” model, a collection of sub-models that work together to compose the ultimate AI model.

The first sub-model builds a provider risk score based on provider behavior. It gains an understanding of how providers submit claims, the volume, their average number of patients and number of claims submitted, the amount of money billed, and what’s paid and what’s denied. This model creates a clear understanding of each provider’s behavior.

The claim evaluator looks at specific claim details. The various data points create one-to-one profiles and establish if claims are “normal.” Its role is to examine the one-to-one profiles, determining if the number of units billed by providers are appropriate given the patient diagnosis, procedure billing codes and the patient’s medical history.  

The third sub-model, a decision enhancer, evaluates each provider’s compliance to the National Council on Compensation Insurance (NCCI) rules and edits. It creates a profile for providers with their overall adherence to coding rules, guidelines and principles.

Provider risk score sets investigation baseline

Milliman risk scoresThe three sub-models combine to create the ensemble model, establishing a risk score per provider to allow payers to take action as they see fit. Score bands can be adjusted in any way the customer wishes.

For Milliman, Mastercard established a range of 1 to 1,000, with 800 being the predetermined score for further actions.

Anything above 900 is considered very risky; providers with a risk score of 700-899 are moderately risky.  With the predetermined risk score of greater than 800, the model quickly created a list of potentially problematic healthcare providers.

For the riskiest providers, the payer could suspend future payments, initiate investigation proceedings or request records from the highest scoring claims to help substantiate the score and confirm the denial.

For moderately risky providers, the model could kick them out into a different workflow for triage purposes, requesting records and/or prioritizing periodic reviews of their claims. If the payer does find something unacceptable, they can initiate an investigation at that time.

And for the lower risk providers, with scores between 500 and 800, the payer can prioritize periodic reviews of their claims, which may expose fraudulent patterns.

AI pilot identifies more than $235 million in healthcare fraud

The results were astounding. The new model uncovered $235+ million in potential savings for fraudulent claims, identified 2,700 high-risk providers and had a three times uplift in claim-level detection.

“The key to success using AI for provider fraud detection is in the combination of the information on abusive billing practices with suspicious claim patterns found through the use of AI,” Cusick says.

Consider, for example, a cardiologist who frequently prescribed medication known to cause heart attacks, strokes and blood clots at a “primary care center” that turned out to be a regular residence, and a laboratory that billed for unnecessary screening tests.

With the pilot successfully completed, Milliman’s customer is reviewing files to seek recovery of lost funds. Milliman and Mastercard are in agreement negotiations to deploy the ensemble model to make it available to all of Milliman’s current and prospective clients.

Learn more about how Mastercard’s AI helps Milliman detect significant healthcare provider fraud.