Skip to main content

Cornell Engineering

Dynamic Modeler

Dynamic Modeler

From football to ambulance deployment, Matt Maxwell Ph.D. ’11 OR predicts the future

By Robert Emro

maxwellWhen you call for an ambulance, you want it ASAP. But how can an ambulance company ensure the fastest possible response time when previous calls have sent ambulances all over town?

This is the question Matt Maxwell Ph.D. ’11 OR answered in his thesis. “As the day progresses, and ambulances are called out, that creates holes in coverage,” he explains, “so you want to move them around dynamically so they can serve calls rapidly.”

Maxwell used approximate dynamic programming to improve a simulation-based algorithm developed by a previous student. “I was able to boil it down to a very simple algorithm so you don’t need a lot of additional complexity to run it,” says Maxwell. “I was able to speed up running the simulations by up to 100 times faster.”

With that speed, Maxwell could run many weeks’ worth of simulations, giving him the data he needed to fine-tune the algorithm, cutting response times by a small, but significant amount. “These performance increases are a big deal for ambulance companies,” says Maxwell. “If they can take one ambulance and crew off the road, they can save about a million dollars a year.”

Maxwell has co-authored several papers on the work with his advisers Professor Shane Henderson and Associate Professor Huseyin Topaloglu. One was a finalist for best paper at the 2009 Winter Simulation Conference. And he says he has two more in the works.

No ambulance company has yet to implement Maxwell’s work, but there has been some interest, and fire and police redeployment present similar problems.

Maxwell’s skill with algorithms caught the attention of Andrew Daines ’10, an undergrad philosophy major and Cornell Daily Sun columnist Maxwell knew from church. Daines and Emily Cohn ’10, a communications major and Sun editor, were working on a social app, but they needed help.

Daines came up with the idea at a New York Yankees game in June of 2009. By the eighth inning, the score was heavily lopsided in the Washington National’s favor and he wondered how to keep fans interested in the game. When he looked around, he saw that many were on their smart phones checking stats and scores. If fans could use them to compete with their friends at predicting the outcome of the next play, he thought, they would remain engaged no matter how bad the blowout.

Daines sought out Maxwell and PrePlay Sports was born. “He needed someone to come up with a probability-based scoring algorithm,” says Maxwell. “That’s how I got involved. And because of my computer science background, I also got involved in some of the server-side programming.”

Because real-time stats on past plays were more readily available from the NFL, the young entrepreneurs switched their focus to football. Maxwell devised an algorithm that, given a team’s current field position and score, could estimate the probability of outcomes, like an incomplete pass, a hold, or a run over 5 yards. It wasn’t so different from ambulance deployment, according to Maxwell.

The PrePlay app lets users compete with their friends at predicting the outcome of NFL plays.  
The PrePlay app lets users compete with their friends at predicting the outcome of NFL plays.

“In both situations you come up with a fundamental model of what you think can describe it, but then you have to make it fit,” he says. “You can look through previous history and not find an exact match, but maybe something a lot similar. The key here is to group similar outcomes to get a good estimate. So you have to define what is similar and what is the right amount of data.”

With some startup capital from family and friends, PrePlay Sports hired a development firm to create a prototype. It was greeted enthusiastically by investors and the company raised enough money to create a beta version. After launching in the fall of 2010, the free app was ranked number one on Apple’s list of hot sports apps.

Today the company employs around 10 people in New York and Paris. It has raised $2 million in venture capital and boasts of users in the hundreds of thousands. During the 2011 season, they made about 600,000 guesses and the average number of guesses per weekend hovered around 60,000. Subway sponsored the PrePlay 2012 Playoffs contest, in which users competed for prizes including a 32” flat screen TV, Apple TVs, and gift certificates to Dick’s Sporting Goods.

The company also has expansion plans. “Our company has attracted interest from sports and entertainment companies and we are currently building technology for them,” says Maxwell.
But for now, PrePlay Sports remains a sideline for Maxwell, whose day job is at the SAS Institute, the North Carolina-based business analytics services and software company, working on a revenue management product for hotels.

Maxwell still finds time to play the app, though he says he’s not the best at predicting plays. “I think the most fun part of it is actually holding it in my hand and watching the game and saying ‘Hey, that number is there because of something I did,’ and that’s a pretty neat experience,” he says.