How airport passenger flow is like hospital patient flow….or translating to your own industry

I went to two talks this morning related to my work in health care that might on the surface seem completely irrelevant. One was “Forecasting Airport Passenger Flow Using Machine Learning and Real-Time Data,” by Yael Grushka-Cockayne of the University of Virginia’s Darden Business School. She was a very enthusiastic speaker and talked about a big project with Heathrow Airport’s Operation Center, which is a kind of command center that co-locates various airport operational functions, not unlike the Patient Logistics Center we have at my employer, UNC Health Care System. Ours was modeled after the Capacity Command Center at Johns Hopkins Hospital, which I am touring on Wednesday. Their task was to predict in real-time passenger flow through the airport, and their initial focus was predicting connection times for passenger transfers, so they could help the airlines determine when/if to hold planes if enough people are arriving soon, as well as to help services throughout the airport predict the right staffing levels at the right points at the right time. They chose a regression tree model because of its interpretability and also built a simulation model of arrivals. They were excited that Heathrow liked their initial proof of concept, which is now being implemented by a consulting firm.

The other talk was “Aligning Analytics and Culture: Is Your Organization Ready for the Industry Digital Transformation,” by Michael Bentley of Revenue Analytics, on revenue management for the cargo industry. Michael had experience doing revenue management for airlines and hospitality but cargo was new for him. They had plenty of data, so he thought they’d be in good shape, until he realized it wasn’t fit for analysis. They had big cultural changes to work through, since an operational-focused culture meant an emphasis on utilization, not the commercial side of the business. For these reasons and others he said something I’m fond of saying, which is the math is the easy part. He had a clever metaphor at the beginning comparing the complexity of a Formula One car to that of speed bump, which is simpler and costs far less but can still take out the car. So he warned about speed bumps that can cause troubles.

Predicting behavior related to passengers flowing through an airport has similarities to patients flowing through a hospital, which is why I attended Yael’s talk. She even talked about the open question of whether co-locating operational services together will offer improvements, which is one question we’re asking of our Patient Logistics Center. I attended the second talk because health care has some similarities to revenue management problems: we have perishable product (appointment slots), relatively fixed capacity, and accept reservations in advance. We will not use price as a control mechanism, but we’ve been asking how we can look at other revenue management techniques to help us shape demand so that we increase patient access to appointments while improving utilization. I took note of some use of statistical modeling they had done to estimate unconstrained demand, since like us they lack good data to track this metric.

It is important to learn how to translate problems from other industries to see how you can apply them in your own. On the surface, health care bears no resemblance to cargo pricing or airport passenger predictions, and yet structurally the problems have similarities. If it weren’t a conflict I would have gone to another session by Lehigh University on “The Inmate Assignment & Scheduling Problem: Application to the PA Dept. of Corrections,” because we’re working on some assignment optimization problems at UNC Health Care. Too often people only look for solutions in their industry, but good work is being done in many places. In fact, tonight at dinner I learned about an entertainment company facing a similar problem to one we face related to patient transport assignment optimization. It is learning from a diversity of industries that is one reason I appreciate conferences in general and this one by INFORMS in particular.