Date of Graduation

5-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Business Administration (PhD)

Degree Level

Graduate

Department

Supply Chain Management

Advisor/Mentor

John Aloysius

Committee Member

Christian Hofer

Second Committee Member

Nada Sanders

Third Committee Member

Enno Siemsen

Fourth Committee Member

Brent D. Williams

Keywords

Behavioral Experiment, Demand Planning, Field Study, Human Judgment, Machine Learning, Supply Chain Analytics

Abstract

Big-data, analytics, automation, and machine learning are changing the role of managers in supply chain and operations functions. Extant research indicates that effective value creation by analytics is achieved through careful attention to three components: technology, people, and processes. As such, the purpose of my dissertation is to improve the integration of people in current supply chain and operations management systems with ever-changing technologies and processes. I focus on one critical function—demand planning—since it is increasingly influenced by predictive analytics yet can still require human judgment. My dissertation is comprised of three essays as follow.

In Essay 1, I implement an experiment to compare existing methods of human-machine integration with two new machine learning methods. The machine learning methods of integration are classified as supervised machine learning and allow the machine learning algorithm to utilize human judgment inputs to train the model. The findings suggest that while human judgment provides a significant benefit, not all methods of integration are equal. The results indicate that the two new machine learning methods of integration that I propose are the most effective forms of integration vis-à-vis other methods commonly used in practice and studied in the academic literature.

Essay 2 consists of a field study at a large, multinational firm, testing the two machine learning methods of integration introduced in Essay 1—interactive machine learning (IML) and human guided machine learning (HGML). Analyzing the results of over three million datapoints across five product categories reveals that demand forecasts using an appropriate process to integrate machine learning and human judgment — IML and HGML — provide a significant benefit to demand planning.

In Essay 3, I study the behavioral mechanisms driving analytics use. Utilizing a multi-theoretical lens combining Adaptive Character of Thought (ACT) theory and dual process theory, I develop, and test, interventions embedded in training aimed at changing behavior and improving performance. The experiment tests the trainings on two types of demand planning tasks: 1) forecasting using IML (classified as a high-level processing task) and 2) forecasting using HGML (classified as a low-level processing task). Results of an experiment reveal declarative knowledge alone changes behavior and improves predictive performance for low-level processing tasks. In contrast, high-level processing tasks benefit from analytical thinking paired with declarative knowledge in training.

This dissertation contributes to the literature on behavioral supply chain and operations management, specifically with reference to human judgment and predictive supply chain analytics. The three overarching contributions are: 1) A comprehensive study testing the efficacy of integrating human judgment and model-based analytics when humans have contextual information unavailable to the model; 2) A first empirical behavioral study of the integration of human judgment with machine learning, that introduces HGML and IML to the behavioral operations and supply chain literature; and 3) An investigation of behavioral interventions that improve predictive analytic processes, using both theories of ACT and dual process theory. In practice, my findings provide operations and supply chain managers with guidance as to when and how human judgment should be integrated with analytics.

Download

Share

COinS