Strategic Path Selection in Service Networks: Leveraging Machine Learning to Estimate Combinatorially Complex Preferences for Consumption and Waiting

73 Pages Posted: 16 Apr 2021 Last revised: 28 Feb 2024

See all articles by Ken Moon

Ken Moon

University of Pennsylvania - The Wharton School

Date Written: October 1, 2023

Abstract

Complex choice sets are challenging for empirical studies of customer decision-making. Within service networks such as marketplaces, shopping centers, and amusement parks, customers routinely choose between combinatorially many network paths to reach their desired stations while evading congestion. Existing methods struggle to accurately estimate the preferences regarding consumption and wait-times that underlie customers' dynamic path selections. We develop new estimators to meet this challenge. Exploiting concepts from discrete optimization, we demonstrate that a limited set of utility comparisons among "local" choices suffices to verify the optimality of a customer's decided selection of stations. Importantly, these relatively few comparisons also suffice for maximally learning customer preferences from their chosen paths. Because neural networks capably exploit such low-dimensional structure, we design estimators that use them to discriminate between and thus identify different customer types from path choice data. We validate our approach using both Monte Carlo experiments and an empirical case study utilizing customer mobile location data from a hypermarket serving over 1M shoppers annually. By consistently and tractably estimating customer preferences, our new empirical methods enable analyses of service network design, capacity management, and customer targeting that account for network complementarities and strategic waiting.

Keywords: Adversarial estimation; Consumer choice; Empirical operations management, Machine learning, Marketplace design, Neural networks, Queueing, Service networks, Structural estimation

Suggested Citation

Moon, Ken, Strategic Path Selection in Service Networks: Leveraging Machine Learning to Estimate Combinatorially Complex Preferences for Consumption and Waiting (October 1, 2023). Available at SSRN: https://ssrn.com/abstract=3819117 or http://dx.doi.org/10.2139/ssrn.3819117

Ken Moon (Contact Author)

University of Pennsylvania - The Wharton School ( email )

Jon M. Huntsman Hall
3730 Walnut St.
Philadelphia, PA 19104-6365
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
291
Abstract Views
1,565
Rank
195,368
PlumX Metrics