How AI is Helping Mitigate Fraud, Waste, and Abuse in Transportation Benefit Programs


Christopher DeSantis,
Data Scientist at American Logistics

An algorithmic approach is helping ensure equal and fair transportation access to all members while helping Medicare Managed Plan administrators save money.

American Logistics provides Non-Emergency Medical Transportation for Medicare and Medicaid managed benefit plans, providing critical transportation to ensure members can access medical care. Our model incorporates ride-share for healthcare into our offering.

In the past, it's been difficult to uncover fraud, waste and abuse scenarios in this arena – but today technology is changing this.

Fraudulent practices include billing for mileage when a beneficiary is not in the vehicle and billing for services never provided. For example, the owner of an Indiana NEMT company was found guilty of billing for services never provided, including billing for cancelled trips, up-coding some trips, and using patients’ medical identifiers without their knowledge to bill for services they never received. The fraud scheme resulted in a loss of over $1 million to Indiana Medicaid.

At American Logistics, our AI or machine learning approach to combat fraud, waste and abuse started with a data analytics project to understand passenger “no show” rates by county and by healthcare plan. Reports enabled us to investigate scenarios of high volumes of passenger “no shows” by passenger IDs and driver IDs. A pattern of high volume was identified and formed the basis for our first ML model.

To visualize and better understand instances of potential FWA we next plotted the distance of trips in miles and how long it took to service that trip in minutes. We identified the statistical mean of trip duration grouped by mileage, or average time in minutes to complete a trip for a given distance. By standardizing the data, we were able to ascertain that 68% of all trips for a given distance are completed within a given range of a time that is 1 standard deviation above or below the mean, and 95% of trips are within 2 standard deviation above or below the mean.

We focused our attention on trips that were outliers – which means that the time to complete a trip of distance X was so far below the mean that it was faster than 95% of all trips for that distance. Drivers who consistently log trips that are 3 or -3 standard deviations from the mean (or more) are flagged and handed off to our group in charge of investigation. True positive results are re-fed into the outlier detection model to form a collection of data points for the model to base future decisions. The process of detecting fraud, waste, and abuse started as simple statistical reporting but has now developed into machine learning.

Further manual investigation is a vital part of the process because it’s important to be wary of false positives in machine learning models and mathematical output should never solely be used to establish a person’s guilt. There’s always the possibility that there’s a reasonable explanation for a data anomaly – i.e. maybe a driver is incorrectly using the trip pick up and drop off technology. The goal of technology is to pinpoint scenarios that warrant more investigation. The goal of humans is to understand the additional context around the data that may tell a different story.

If we are able to verify a fraudulent act has occurred, we recommend that the driver or passenger is removed from participating in the program.>


As a result of this AI/machine learning approach, we’ve been able to dramatically improve both effectiveness in identifying suspected fraud, waste and abuse, and productivity. We’ve also streamlined the detection process through standardized scripts and algorithms. Previously the process of compiling data took 2 weeks; today it takes 5 minutes.

Fraud, waste, and abuse pickpockets every taxpayer and the perpetrators stand in the way of ensuring equitable resources for all. AI and machine learning are fast becoming vital tools in the fight to put this situation to right.

Subscribe to Industry Era Review