UCLA

When we think about Operations Management and Business Analytics, we think about optimization, efficiency, algorithms, optimality, profits and costs, exact quantitative analyses, etc. However, the field was created by humans and for humans. So what is the role of humans in the process of operational decision-making? In this dissertation, we study two aspects of the interface between humans and operational decision making, in the context of the apparel industry. The first part (Chapter 1) is related to social responsibility, i.e., how operational decision making affects workers, communities, and society. The second part (Chapters 2 and 3) shows how human biases affect operational decision-making.

In the first chapter, we study unauthorized subcontracting, i.e., when suppliers outsource part of their production to a third party without the retailer's consent. This practice has been common practice in the apparel industry and it is often tied to non-compliant working conditions. Since retailers are unaware of the third party, the production process becomes obscure and cannot be tracked adequately. We present an empirical study of the factors that can lead suppliers to engage in unauthorized subcontracting. We use data provided by a global supply chain manager with over 30,000 orders, of which 36% were subcontracted without authorization. Our results show that there are different factory types, ranging from factories that used unauthorized third parties for all of their orders to factories that used none. Moreover, the degree of unauthorized subcontracting in the past is highly related to the probability of engaging in unauthorized subcontracting in the future, which suggests that factories behave as if they choose a strategic level of unauthorized subcontracting. At the order level, we find that state dependence (i.e., the status of an order carrying over to the next one) followed by price pressure are the key drivers of unauthorized subcontracting. Buyer reputation and factory specialization can also play a role, whereas the size of an order shows no effect. We find that the main effect (state dependence) is tied to factory utilization. Finally, we show that unauthorized subcontracting can be predicted correctly for more than 80% of the orders in out-of-sample tests. This indicates that retailers can use business analytics to predict unauthorized subcontracting and help prevent it from happening.

In the second chapter, we study the adherence to the recommendations of a decision support system (DSS) for markdowns during clearance sales. The DSS was implemented at Zara, the Spanish fast fashion retailer. Managers' initial adherence was low, which motivated two interventions: 1. showing a revenue metric; and 2. showing a reference point for that metric. We use data collected by Zara during seven clearance sales campaigns to analyze the effect of the two interventions and the behavioral drivers of managers' adherence decisions. Intervention 1 did not significantly alter managers' adherence, but Intervention 2 increased it, and also decreased their likelihood to mark a product down when DSS recommended keeping its price unchanged. Managers were more likely to adhere to the DSS's recommendations when the suggested price was aligned with the heuristic they followed before the DSS was implemented. Managers' decisions were consistent with inventory minimization, as opposed to revenue maximization. Higher salvage values were related to higher adherence, but also to larger deviations when managers did not adhere. Managers were minimizing the number of different prices to set and basing their pricing decisions on metrics that were aggregated at the group level, instead of at the individual product level. These findings can be explained by preference for the status quo, salience of the inventory (compared to a revenue forecast), loss aversion, and inattention. Some of these biases were mitigated after the interventions. Our findings provide insights on how to increase voluntary adherence that can be used in any context in which a company wants an analytical tool to be adopted by its users.

In Chapter 3, we continue to study pricing decision making by country managers at Zara. We aim to disentangle managers' degree of loss aversion from other behavioral biases by building a structural model to replicate managers' price decision making process and fitting it using data collected by Zara prior to the DSS's implementation. In our model, managers choose prices to maximize their utility over the whole season, subject to a number of constraints given by the firm's pricing rules. The utility function consists of a revenue component and a loss aversion component that depends on a loss aversion parameter. Both components include demand uncertainty and contain all products of the same type (shirts, pants, etc.). This model is, therefore, a dynamic program over a finite horizon with a large state space and uncertainty set. We use a certainty equivalent for the demand function, and discretize the problem, given that the set of available prices is discrete. Our model thus becomes a mixed integer linear program. We find, for each value of the loss aversion parameter, its corresponding set of utility-maximizing prices, and then pick the value of the parameter that best fits the prices that managers implemented. We then compare their degree of loss aversion across product groups (e.g. fashion or basic products) and across country managers. Preliminary results using a very small dataset suggest that country managers at Zara are, indeed, loss averse, but some managers are more so than others. There seem to be no clear pattern on what product types trigger loss aversion in managers. Our model explains managers' observed prices 11\% better than if we assume they are pure revenue maximizers, and several times better than if we model them as inventory minimizers.