סמינר בניהול טכנולוגיה ומערכות מידע

First speaker: Maximilian Lowin, Faculty of Economics and Business Administration, Goethe University Frankfurt.

Title: "Using Expert-in-the-Loop Causal Machine Learning to Improve Decision-making during Disasters”

Abstract: Disasters challenge decision-makers as they require them to take actions under extreme conditions such as time constraints, uncertainty, and information and resource sparsity. Because disasters can lead to severe economic damage and loss of lives, comprehensive management is crucial. To support disaster managers in their complex endeavors of making the right decisions that can mitigate the adverse effects of upcoming or ongoing disasters, we propose a disaster type- and disaster management phase-agnostic Expert-ML decision-support framework. We combine expert knowledge with Machine Learning (ML) to leverage the strengths of human and machine intelligence while avoiding the shortcomings of a purely human or ML-driven system. We rely on causal knowledge, which we implement with Causal Bayesian Networks, to increase transparency in the decision-making process. Further, by arranging the human and machine interaction in a loop, we can ensure the frameworks' flexibility to cope with the dynamic developments of disasters. As part of the team for Evidence-Based Pandemic Management of the German University Medicine (“German National Corona Task Force”), we evaluated our framework on the disaster response phase during the second and third wave of the COVID-19 pandemic. We predict the probability of a patient’s ICU transfer at hospital admission, which is on the aggregated level an important key indicator for the allocation management of ICU resources. Based on our proposed frame-work, we demonstrate the superiority of a hybrid Expert-ML causal machine learning approach versus purely expert- or ML-based decision systems.

Second speaker: Moritz von Zahn, Faculty of Economics and Business Administration, Goethe University Frankfurt.

Title: “Reducing Net Product Returns through Green Nudges – Insights from a Randomized Field Experiment and Causal Machine Learning”

Abstract: As free customer deliveries are becoming a standard in E-commerce, product returns pose a growing challenge to online retailers and society. For retailers, product returns create considerable costs associated with transportation, labor, disposal and infrastructure to manage returns. From a societal perspective, increasing product returns contribute to increased pollution, additional trash, and often a waste of natural resources. Due to these costs, companies and society are interested in reducing product returns. However, despite strong entrepreneurial and public interest in minimizing product returns, retailers on a micro level possess only very few effective instruments to minimize product returns without harming customer demand and net sales. In this work, we propose a novel product return prevention instrument (Smart Green Nudging) that leverages Causal Machine Learning (CML) and the availability of rich customer and contextual data sources to identify and nudge selected customers towards better shopping choices that will yield reduced product returns without diminishing customer demand and net sales. We evaluate the performance of the proposed returns prevention instrument with real-world data from the German online shop of a large European retailer. We demonstrate in a randomized field experiment with ~1 million visitors that showing a green nudge to all customers (Naïve Green Nudging) can reduce product returns but also incurs a decrease in demand, which however ultimately translates to higher net profits. Further, we demonstrate the
superiority of Smart Green Nudging over Naïve Green Nudging in terms of both product returns and firm profits. Specifically, in our randomized field experiment,  when compared to Naïve Green Nudging, the Smart Green Nudging strategy curtails product return by additional +1.23% and increases profits by additional +3.61%. Overall, this paper demonstrates the efficacy of using state-of-the-art CML to customize minimally invasive behavioral nudges in the digital environment as a means to tackle business and societal problems.