Algorithmically Motivated
Daniel Freund is an assistant professor of operations management at the MIT Sloan School of Management. His research focuses on complex decision-making problems in the sharing economy.
Reflecting on the benefits of connecting his research with operational decision-making in practice, Daniel Freund says, “A field that studies operations management only in theory studies a fictional world of operations.” The statement might lead one to assume that he has little interest in developing algorithms that don’t have immediate real-world applications. But the truth is more nuanced. Freund is an assistant professor of operations management at the MIT Sloan School of Management, and much of his research is focused on complex decision-making problems in the sharing economy; all of it stems from a fundamental interest in algorithms.
Freund has collaborated with bike-sharing systems like Citi Bike, Ford GoBike, and Boston Blue Bikes. He also spent a year as a research fellow at Lyft Marketplace Labs. These days, he’s excited about a new collaboration with a delivery platform that deploys various types of supply that result in a unique set of challenges. Consider the supply side of a typical ride-hailing system like Lyft or Uber, which consists of one thing: drivers in cars. This platform, on the other hand, offers delivery via bicycle, motorcycle, car, and even has supply in the form of people getting around the old-fashioned way, on their own two feet. “Managing different types of supply, finding the right mix of them is interesting from a practical standpoint, but also because it relates to mathematical questions that have been explored without this specific application in mind,” says Freund.
This balance of theory and practice is apparent in Freund’s working paper, “Good prophets know when the end is near.” Rather, the story behind the paper illustrates the point. A few years ago, he was working with a company facing an algorithmic challenge. “The details of the challenge don't matter much,” Freund says. What does matter is that he recognized a connection between the issue he was pondering and the work of one of his former collaborators. Fundamentally motivated by theory, the other researcher had developed a tool to make progress on a long-standing problem in revenue management.
Working together, the two researchers were able to provide an algorithmic solution to the company’s problem, which got them thinking further about the tool’s potential to address other algorithmic challenges. They began exploring its limitations in earnest and wound up making progress on other fundamental questions in operations research. “This type of back and forth is the most exciting thing to me,” says Freund. “He started with a purely theoretical interest while I was interested in a specific application. Meeting in the middle, we solved the applied problem and gained insights on theoretical problems. Having an impact on both sides is really what I strive for today.”
Meeting in the middle, we solved the applied problem and gained insights on theoretical problems
As an undergraduate at Warwick University, Freund was a strict adherent of pure mathematics. “Originally, I just wanted to study mathematics for mathematics’ sake,” he says. But somewhere along the line he became fascinated with algorithms, their combinatorial structure, and the idea of using them to solve optimization challenges. However, as he points out, it was still in the name of theory.
Eventually, Freund’s theoretical leanings tipped in the direction of applications. At Cornell University, where he earned his PhD in applied mathematics, he met key figures in the bike-sharing industry, which sparked a newfound interest in using algorithms to solve real-world problems. He was struck by the commitment of these individuals to providing equitable access to transportation options. Their passion inspired Freund, and it occurred to him that his algorithmic skills could help improve their bike-sharing systems.
Working with colleagues at Cornell, Freund used algorithms and advanced analytics to improve the systems of Motivate, then the largest operator of bike-sharing programs in the US. In 2019, they published their findings in a paper titled “Analytics and bikes: riding tandem with Motivate to improve mobility.” From an algorithmic standpoint, says Freund, solving challenges around docking and sizing different stations was particularly interesting. How many docks should each station have? Where do you need more dock capacity? What if you want more dock capacity at a station with spatial constraints? The algorithms he developed answered questions like these for Motivate and are relevant for any bike-sharing program looking to provide a better experience for their users.
Freund has also made important contributions to ride-hailing. In his paper, "Pricing fast and slow: limitations of dynamic pricing mechanisms in ride hailing," Freund takes a deep dive into a problem with current models used to analyze the systems for designing pricing and other control policies. While revenue management has been a mainstay of the airline industry for a long time, ticket prices for flights don’t change minute by minute. Ride-hailing, on the other hand, promised to be one of the first real-world applications of dynamic pricing in real time.
But, in designing their systems, the ride-hailing industry ignored the possibility that rapid price changes might incentivize customers to wait for the price to drop before booking a ride. As Freund put it at the time, “Dynamic pricing mechanisms deployed by companies like Uber and Lyft are limited by the fact that price changes are so fast paced, leading to inefficient supply oscillations.” Lyft eventually implemented a "wait and save" feature, giving customers the option to pay the original (higher) price or pay less and wait longer for pick-up. “Our study suggests that by offering two different streams, the platform gains greater control over its supply than it would under the traditional dynamic pricing scheme in which the price is either high or low, but it changes instantaneously and for everyone. Adding the ‘wait and save’ option benefits all: the riders, the drivers, and the platform.”
These days, it might seem like data-driven decision-making, optimization through algorithms, is the answer to everything. And Freund says he’s seen his fair share of applications and operations that would (or, with his help, did) benefit from a data-driven approach. Take Citi Bike’s method for relocating bikes between stations to meet user demand, otherwise known as rebalancing. In the early days, Citi Bike’s dispatchers made rebalancing decisions based on intuition. Freund developed data-driven tools to help them make better rebalancing decisions, benefiting the company and its riders.
Our algorithms address one part of the question: in this case, the core team composition
Closer to home, he has been working on a project with the MBA office at MIT Sloan. Every year, the MBA office assigns incoming students into core teams, groups of six or seven students that will interact closely as they move through their first semester together. Core teams are meant to be representative of the incoming class, providing students exposure to a rich mix of backgrounds, interests, and experiences. Freund developed an algorithm to help optimize this process.
Using the project as a jumping off point for a larger discussion, Freund points out that optimization is not a solution in and of itself. “Optimization as a tool is often limited by its input,” Freund explains. “Our algorithms address one part of the question: in this case, the core team composition. But the diversity of the student body sets a natural limit on that composition. Algorithms help us identify the optimal solution for the current situation but optimizing one part of a process often leaves larger systemic question open. In a best-case scenario, the optimization will also shed light on these larger questions. But addressing them usually involves wider systemic changes.”