What to do when decision-makers deviate from model recommendations? Empirical evidence from hydropower industry

• Published in: European journal of operational research Vol. 278; no. 3; pp. 869 - 882
• Main Authors: Käki, Anssi, Kemppainen, Katariina, Liesiö, Juuso
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Decision support systems; Empirical research; Energy-related operations; OR practice; Decision support systems; OR practice; Energy-related operations; Empirical research
• ISSN: 0377-2217; 1872-6860
• DOI: 10.1016/j.ejor.2019.04.021

•Studies deviations between optimization tool recommendations and actual decisions.•Demonstrates benefits of combining objective and judgmental process feedback.•Explains value of input-adjusting human-model interaction in system improvement.•Presents a general intervention process for improving decision models. Decision makers do not always follow recommendations from model-based decision support systems. We suggest that analyzing the differences between decision recommendations produced by prescriptive models and the behavior of the decision makers provides valuable insights that can be utilized to improve model-based decision support processes. Specifically, we develop an intervention process in the context of hydropower production planning to study the motivations of decision makers and the ramifications of their behavior. The analysis is based on deviations between recommendations of an in-house optimization tool and actual decisions, enhanced by planner feedback collected from a daily web-survey. We find that even though the planners make some adjustments with positive financial impact, their actions mainly worsen the performance of the production plan. Using the collected data, we identify several reasons for the deviations and recommend multiple enhancements to the planning process. For example, we propose a shift from output-adjusting to input-adjusting interaction between human planner and model. Altogether our facilitated modeling project shows that combining objective and judgmental process feedback is superior for recognizing corrective actions and systematically improving model-driven decision processes. Furthermore, the intervention process developed for this case gives structure for the lifecycle management of model-based decision support systems.