Expediting Analytics Research

Selecting an advertising slot for a large online retailer. Creating a model to predict the salability of a product. These business challenges may seem to have little in common with societal imperatives like minimizing the impact of the opioid crisis or providing underserved communities with access to fresh produce. However, Georgia Perakis, the William F. Pounds Professor of Management and codirector of the interdepartmental Operations Research Center, designs algorithms that address all of these problems.

Her research is focused on developing computational models and algorithms for pricing, supply chain management, and demand prediction, among other areas. Speed is key in the practical application of these algorithms. Information for thousands of products in thousands of stores must be parsed and processed within seconds. The resulting information allows companies to create personalized pricing and targeted promotions and provides scholars such as Perakis opportunities to better understand trends and develop new ways to predict demand. “What we build combines machine learning, econometrics, and optimization,” she says.

Historically, retailers have relied on gut feelings and experience to decide what products to promote and when. “There is a lot to be said for this experience, but it has limitations,” Perakis says. “For example, a grocery store typically has approximately 2,000 stock-keeping units (SKUs) on promotion at any one time. The human brain cannot parse this amount of data, so using analytics represents an important market edge.” She combines business experience with data to create new computational models. “This method allows managers to run many ‘what if’ scenarios and still utilize their intuition,” she adds.

Perakis recognized that this methodology also has compelling societal applications. For example, she is investigating how it can be used to help nonprofits best serve economically disadvantaged communities. She’s also researching how it can help identify which patients treated for substance use disorder are most likely to relapse, providing an opportunity to put targeted clinical supports in place.

“Computing is core to all of my work,” says Perakis. “Solving important problems, difficult problems, with computational methods and testing them in real-world practice is what I love to do.”

This story was originally published in January 2020.

Share this story

Computing Intelligence