Speaker: Dr. Kleanthis Karakolios
Date: Tuesday, 28 April 2026, Time: 15:00
Place: Microsoft Teams meeting https://teams.microsoft.com/meet/348968333477102?p=xGaW2v8S2IohVAOFyO, Meeting ID: 348 968 333 477 102, Passcode: hV7DK7BK
Abstract
Interference between treated and untreated units is a pervasive source f bias in marketplace experiments. This presentation focuses on pricing interventions in which a platform lowers base prices to stimulate demand. In matching marketplaces, such interventions raise a fundamental algorithmic design question: should treated and untreated units be matched differently to account for price differences? We show that standard estimation methods yield biased estimates, with the direction of bias depending critically on this design choice. To address this, we introduce the shadow price estimator, derived from the optimal dual solution to the platform’s supply–demand matching problem. We further propose a matching design in which the platform deliberately ignores price differences during the matching process, and show that this design substantially reduces estimation bias.
Speaker Bio:
Kleanthis Karakolios works as a strategic consultant at Ernst & Young – Parthenon (head of the AI Go-To-Market lab). He is a graduate of the Georgia Institute of Technology (PhD in Machine Learning ’25, MBA ’24, MS in ECE ’22), Columbia Business School (MS in Operations Research ’18), and the Aristotle University of Thessaloniki (Diploma in ECE ’13). His research activity focuses on Causal AI, Experimentation, Optimization, Stochastic Modeling, and Business Analytics. He has worked as a researcher, engineer, and data scientist at Microsoft, Amazon Web Services, JPMorgan Chase, Bank of America Merrill Lynch, and CERTH/ITI.
This lecture is part of the seminar series of the MSc Program in Supply Chain Management and Logistics of the Department of Mechanical Engineering.