Enno Siemsen

The algorithms available for retail forecasting have increased in complexity. Newer methods, such as machine learning, are inherently complex. The more traditional families of forecasting models, such as exponential smoothing and autoregressive integrated moving averages, have expanded to contain multiple possible forms and forecasting profiles. We question complexity in forecasting and the need to consider such large families of models. Our argument is that parsimoniously identifying suitable subsets of models will not decrease forecasting accuracy nor will it reduce the ability to estimate forecast uncertainty. We propose a framework that balances forecasting performance versus computational cost, resulting in the consideration of only a reduced set of models. We empirically demonstrate that a reduced set performs well. Finally, we translate computational benefits to monetary cost savings and environmental impact and discuss the implications of our results in the context of large retailers.

Purpose: This paper enhances our understanding of how national culture impacts manufacturing performance (assembly speed, consistency between teams, etc.) during a production process move. The authors also investigate the efficacy of co-location as a strategy to enhance knowledge transfer from one organization to another.Design/methodology/approach: To study the impact of national culture on production process moves, the authors develop and employ a team-based behavioral experiment within and between an individualist society (the United States) and a collectivist one (China). The authors also examine the impact of co-location on knowledge transfer effectiveness within and between these two unique cultures.Findings: Interestingly, co-location has little impact on the performance of US recipient teams. Without co-location, Chinese recipient team performance lags significantly behind the US teams. However, firms can overcome these knowledge transfer challenges by co-locating source and recipient team members. These results suggest that firms should assess the national cultural context when considering co-location to manage their production move. There are contexts where co-location may be incredibly useful to facilitate an effective knowledge transfer (e.g. collectivist cultures like China) and contexts where this approach may not be as valuable (e.g. individualistic cultures such as the United States).Originality/value: This research contributes to the academic literature in several ways. First, while past research demonstrates that national culture can be an essential barrier to information and knowledge sharing, this paper extends these findings showing that co-location may effectively overcome this barrier. After the authors offer and test the merits of co-location, they also establish the boundary conditions of this approach by showing that the effect of co-location on knowledge transfer is contingent on the cultural context. This contribution enhances our understanding of the relationship between national culture and knowledge sharing and has implications for managers developing approaches to transfer knowledge between cultures. Second, the authors develop and execute a novel cross-country experimental design. While cross-country experiments have been done before (e.g. Ozer et al. 2014, Kuwabara et al. 2007, etc.), it is still rare to see such experiments due to them being “technically difficult and costly” (Ozer et al. 2014, p. 2437). This research not only offer insights into how teams of people from individualist and collectivist societies send, receive and comprehend production knowledge. It also documents how these teams convert this knowledge into production results.

Our research examines how to integrate human judgment and statistical algorithms for demand planning in an increasingly data‐driven and automated environment. We use a laboratory experiment combined with a field study to compare existing integration methods with a novel approach: Human‐Guided Learning. This new method allows the algorithm to use human judgment to train a model using an iterative linear weighting of human judgment and model predictions. Human‐Guided Learning is more accurate vis‐à‐vis the established integration methods of Judgmental Adjustment, Quantitative Correction of Human Judgment, Forecast Combination, and Judgment as a Model Input. Human‐Guided Learning performs similarly to Integrative Judgment Learning, but under certain circumstances, Human‐Guided Learning can be more accurate. Our studies demonstrate that the benefit of human judgment for demand planning processes depends on the integration method.

Effective approaches to forecast model selection are crucial to improve forecast accuracy and to facilitate the use of forecasts for decision-making processes. Information criteria or cross-validation are common approaches of forecast model selection. Both methods compare forecasts with the respective actual realizations. However, no existing selection method assesses out-of-sample forecasts before the actual values become available—a technique used in human judgment in this context. Research in judgmental model selection emphasizes that human judgment can be superior to statistical selection procedures in evaluating the quality of forecasting models. We, therefore, propose a new way of statistical model selection based on these insights from human judgment. Our approach relies on an asynchronous comparison of forecasts and actual values, allowing for an ex ante evaluation of forecasts via representativeness. We test this criterion on numerous time series. Results from our analyses provide evidence that forecast performance can be improved when models are selected based on their representativeness. This paper was accepted by Manel Baucells, behavioral economics and decision analysis. Supplemental Material: The online appendix and data are available at https://doi.org/10.1287/mnsc.2022.4485 .

Demand planning is informed by demand forecasts, service level requirements, replenishment constraints, and revenue projections. “Demand forecasts” differ from “demand plans” in that forecasts only represent the distribution (or the most likely value) of product demand. Motivated by common forecasting practices in industry, our research examines whether forecasters recognize this difference between demand forecasts and demand plans. Based on a lab experiment informed by data from two large FMCG companies, we found that forecasters factor service levels into their demand forecasts, even when they are clearly instructed to predict the most likely demand and incentivized to minimize the forecast error. We establish that this result holds for students and practitioners alike, and show that this behavior is driven by the service level information, and not some other anchor. We use data from a recent industry survey to support the external validity of our key findings.

Problem definition: Firms increasingly use augmented reality (AR) devices to improve their production ramp-up processes. These devices appear useful, yet little is known about their broader impact on worker productivity and behavior. Academic/practical relevance: Efficient production ramp-ups are particularly important when product life cycles are short. An ongoing debate among academics and practitioners pertains to how Industry 4.0, and AR devices in particular, can accelerate the ramp-up. The current study provides empirical evidence related to AR in the production ramp-up context, examines the strengths and weaknesses of AR, and tests four hypotheses, leading to a more nuanced view of AR use in the manufacturing ramp-up. Methodology: A framed field experiment in a manufacturing plant provides a test of how quickly workers can perform new tasks with and without AR support and how the use of AR affects their ability to suggest process improvements. Results: When faced with a new task, workers instructed by AR smart glasses use 43.8% less time to complete the task compared with a control group that relies on paper-based instructions. However, workers that use AR glasses consistently use 23% more time than the control group when both groups repeat the task without either AR or paper-based instructions. Task difficulty moderates this relationship; workers assigned to a more difficult task benefit the most from AR instructions. After the devices are removed, workers instructed based on paper improve their productivity faster through learning than those instructed by AR. In addition, the former group suggests better process improvements than the latter one. Managerial implications: Although these results indicate substantially higher productivity resulting from AR devices, they also support the view that, once instructed through AR devices, workers come to rely on this new technology without fully internalizing the task. This failure to internalize their task then leads workers to suggest less useful process improvements.

Problem definition : Transshipment/inventory sharing has been used in practice because of its risk-pooling potential. However, human decision makers play a critical role in making inventory decisions in an inventory sharing system, which may affect its benefits. We investigate whether the opportunity to transship inventory influences decision makers’ inventory decisions and whether, as a result, the intended risk-pooling benefits materialize. Academic/practical relevance : Previous research in transshipment, which is focused on finding optimal stocking and sharing decisions, assumes rational decision making without any systematic bias. As one of the first to study inventory sharing from a behavioral perspective, we demonstrate a persistent stocking-decision bias relevant for inventory sharing systems. Methodology : We develop a behavioral model of a multilocation inventory system with transshipments. Using four behavioral studies, we identify, test, estimate, and mitigate a demand-side underweighting bias: although inventory sharing brings both a supply-side benefit and a demand-side benefit, players underestimate the latter. We show analytically that such bias leads to underordering. We also explore whether reframing the inventory sharing decision reduces this bias. Results : Our results show that subjects persistently reduce their order quantities when transshipments are allowed. This underordering, which persists even when a decision-support system suggests optimal quantities, causes insufficient inventory in the system, in turn reducing the risk-pooling benefits of inventory sharing. Underordering is evidently caused by an underweighting bias; although players correctly estimate the supply-side potential from transshipment, they only estimate 20% of the demand-side potential. Managerial implications : Although inventory sharing can profitably reduce inventory, too much underordering undermines its intended risk-pooling benefits. The demand-side benefits of transshipment need to be emphasized when implementing inventory sharing systems.

Behavioral operations research has proliferated greatly over the decade since its first formal review in 2006. The growth of the field warrants an objective mapping of contributions to the literature and the identification of trends. We conduct a systematic review of the literature of behavioral operations and supply chain management (BOSCM) across eight key operations and supply chain management journals, with publication dates through the end of June 2018. Collected articles are categorized into 12 operations contexts as well as emerging topic considerations. Key research trends, theoretical foundations, and methodological choices are discussed in each context. The results show that supply chain management, inventory management, and procurement/auctions have been the most popular operations contexts for BOSCM researchers. The results of our co‐citation analysis shows that the fundamental research areas that have informed and shaped the field include supply chain risk management, marketing, cognitive psychology, and social psychology. Based on these findings and a survey of the most prolific authors in the field, we discuss possible avenues for future research.

The field of behavioral operations has produced a rich tradition of experiments in newsvendor decision‐making. Our study provides a meta‐analysis of 24 papers in this research area; we confirm that the pull‐to‐center (PtC) effect, where average order quantities in a sample lie in between average demand and optimal order quantities, is a very stable observable phenomenon across studies. However, we also show that the asymmetry in the PtC effect between high‐margin and low‐margin conditions varies significantly from study to study; factors that allow predicting the extent and direction of the PtC asymmetry include the likelihood of obtaining losses, the way underage costs are presented to subjects, and the existence of a decision support system. We validate these factors using a controlled experiment. In conclusion, the PtC asymmetry appears less as a fundamental attribute of human behavior, but rather as a function of the design aspects of the experiment.