Special guest speaker Professor Marlin Ulmer from Technische Universität Braunschweig presented “Urban Services with Demand and Resource Uncertainty” on November 20, 2020.
In his seminar, Ulmer discussed controlling demand and resources to provide punctual and cost-effective delivery service.
All instant delivery and transportation services experience late or unreliable service from time to time. Too much demand, not enough resources, and overly optimistic delivery estimates are the most common causes.
Typically, drivers receive customer requests, pick up orders, and deliver within the promised time. The priority is minimizing delay for customers. From a tactical perspective, companies can:
- control demand by choosing who they service;
- control resources by choosing how they service demand; and
- control service by changing what they offer customers.
Ulmer chooses to focus on the first two options: controlling demand and controlling resources.
One demand control mechanism is service area sizing: adjusting travel time radius around a place like a restaurant. Keeping a fixed radius is unwise because of demand peaks like lunch and dinner. A good solution captures expected demand over time, and the radius should mirror the demand to reduce congestion in the system. This leads to time-dependent service area sizing which is regularly updated. Through methods like continuous approximation and value function approximation, Ulmer observed that dynamic service area sizing was able to serve up to 22% more customers.
Resources in crowdsourced services can be uncertain due to driver availability, starting time, shift length, and customer demand along with high customer expectations. Similar to service area sizing, driver availability should mirror demand. However, uncertainty in workforce compromises service quality, so companies schedule company drivers. From testing on real data from Iowa City, Ulmer saw that having the same number of drivers every hour cost 42% more than using anticipatory workforce scheduling. Furthermore, crowdsourced drivers are less effective than company drivers, but the value of one crowdsourced hour increases as they work longer in the system because of more consolidation i.e. combining more orders into one trip.
Ulmer notes that crowdsourced drivers should be incentivized to stay in the system longer and pre-announce their arrival so that the scheduling system can anticipate orders for them. To help provide optimal service, demand can be controlled via dynamic service area sizing and crowdsourced resources can be supported by scheduled drivers.
Marlin Ulmer is Professor of Prescriptive Analytics at the Business Department of the Technische Universität Braunschweig in Braunschweig, Germany. He studied Mathematics in Göttingen and Swansea. His PhD thesis at TU Braunschweig focused on the interface between Management Science, Mathematics, and Computer Science. It was awarded, amongst others, by the INFORMS TSL Society and the German Operations Research Society. Marlin’s teaching and research interests comprise applications from the entire field of Urban Mobility and Logistics, Operations Research, Stochastic Optimization, Dynamic Decision Making, Approximate Dynamic Programming, and Machine Learning. He works with leading experts of these fields all over the world.
View videorecording of “Urban Services with Demand and Resource Uncertainty” by Professor Marlin Ulmer, Technische Universität Braunschweig, November 20, 2020 on YouTube.
Presented by University of Toronto ITE Student Chapter, UT-ITE.