Publications

You can also find my articles on my Google Scholar profile.

Peer-Reviewed Journal Articles:

• R Gürlek, DS KC, P Letizia. Impact of Temporary Store Closures on Online Sales: Evidence from a Natural Experiment. Accepted at Manufacturing & Service Operations Management.
  • Honorable mention in the 2023 POMS College of SCM Best Student Paper Competition
  Abstract

  
  Problem definition: This paper examines the impact of retail store closures on omnichannel sales and consumer shopping behavior in the context of the COVID-19 pandemic. To explain the likelihood of store closure, we develop a novel instrumental variable motivated by varying geopolitical responses across the US to the pandemic.
  Methodology/Results: Using data from a luxury fashion retailer, we find that when a store is closed, the volume of online orders originating from its location increases by 24%. Furthermore, when the retailer closes 10% of its stores, the omnichannel total sales (offline + online) decrease by 5.5%. Notably, our findings indicate that the online channel enables the retailer to recover 11% of offline sales that would have otherwise been lost due to store closures. We also show that compared to existing e-shoppers, new e-shoppers are more likely to order popular product models in an effort to mitigate the heightened mismatch risk associated with online transactions. For new e-shoppers, the likelihood of ordering a popular model stands at 70%, whereas it is 45% for existing online consumers. Additionally, the conservative behavior of favoring popular models reduces the likelihood of returns by new e-shoppers.
  Managerial implications: Even for luxury apparel often associated with in-store purchases requiring “touch and feel” and customer tryout, the option to purchase online proves immensely valuable. The tendency of new e-shoppers to limit product mismatch risk by choosing popular products may create an opportunity for retailers to strategically target these inexperienced online customers with advertisements, product promotions, or virtual fitting rooms, all geared toward reducing online shopping risk of product mismatch.
• N Chen, R Gürlek, DKK Lee, H Shen 2024. Can customer arrival rates be modelled by sine waves?. Joint issue in Service Science and Stochastic Systems
  Abstract

  
  Customer arrival patterns observed in the real world typically exhibit strong seasonal effects. It is therefore natural to ask, can a nonhomogeneous Poisson process (NHPP) with a rate function that is the simple sum of sinusoids provide an adequate description of reality? If so, how can the sinusoidal NHPP be used to improve the performance of service systems? We empirically validate that the sinusoidal NHPP is consistent with arrival data from two settings of great interest in service operations: patient arrivals to an emergency department and customer calls to a bank call centre. This finding provides rigorous justification for the use of the sinusoidal NHPP assumption in many existing queuing models. We also clarify why a sinusoidal NHPP model is more suitable than the standard NHPP when the underlying arrival pattern is aperiodic (e.g., does not follow a weekly cycle). This is illustrated using data from a car dealership and also via a naturalistic staffing simulation based on the call centre. On the other hand, if the arrival pattern is periodic, we explain why both models should perform comparably. Even then, the sinusoidal NHPP is still necessary for managers to use to verify that the arrival pattern is indeed periodic, a step that is seldom performed in applications. Code for fitting the sinusoidal NHPP to data is provided on GitHub.
• ÖG Ali, R Gürlek 2020. Automatic interpretable retail forecasting with promotional scenarios. International Journal of Forecasting.
  Abstract

  
  Budgeting and planning processes require medium-term sales forecasts with marketing scenarios. The complexity in modern retailing necessitates consistent, automatic forecasting and insight generation. Remedies to the high dimensionality problem have drawbacks; black box machine learning methods require voluminous data and lack insights, while regularization may bias causal estimates in interpretable models.
  The proposed FAIR (Fully Automatic Interpretable Retail Forecasting) method supports the retail planning process with multi-step-ahead category-store level forecasts, scenario evaluations, and insights. It considers category-store-specific seasonality, focal- and cross-category marketing, and adaptive base sales while dealing with regularization-induced confounding.
  We show, with three chains from the IRI dataset involving 30 categories, that regularization-induced confounding decreases forecast accuracy. By including focal- and cross-category marketing, as well as random disturbances, forecast accuracy is increased. FAIR is more accurate than the black box machine learning method Boosted Trees and other benchmarks while also providing insights that are in line with the marketing literature.

Under Review

• R Gürlek, F de Véricourt, DKK Lee. Boosted Generalized Normal Distributions: Integrating Machine Learning with Operations Knowledge. Submitted.
  Abstract

  
  Applications of machine learning (ML) techniques to operational settings often face two challenges: i) ML methods mostly provide point predictions whereas many operational problems require distributional information; and ii) They typically do not incorporate the extensive body of knowledge in the operations literature, particularly the findings that characterize specific distributions. We introduce a novel methodology, the boosted Generalized Normal Distribution (bGND), to address these challenges. bGND leverages gradient boosted trees to flexibly estimate the parameters of the GND as functions of covariates, and can be used to model a wide range of parametric distributions encountered in operations. We establish bGND's statistical consistency, thereby extending this key property to special cases studied in the ML literature that lacked such guarantees. Using data from a large academic emergency department in the U.S., we show that the distributional forecasts of patient wait and service times can be meaningfully improved by leveraging findings from the healthcare operations literature. Relative to forecasts from the distribution-agnostic ML benchmark, bGND can potentially improve patient satisfaction by 9% and increase hospital earnings by \$120,000 per 10,000 visits. Our work underscores the importance of integrating ML with operations knowledge to enhance distributional forecasts.
• R Gürlek, M Baucells, N Osadchiy. Optimal Design and Pricing of Sequenced Bundles in the Presence of Satiation.
  • Finalist at the 2024 POMS CBOM Junior Scholar Paper Competition
  Abstract

  
  Problem definition: Sequencing of consumption has significant implications for enjoyment of experiential goods and derived ex-post utility due to psychological and physiological effects, such as satiation, habituation, or memory decay. In this paper, we examine the effect of consumption sequencing on ex-ante valuations of bundles, with the goal of increasing consumer surplus and revenue.
  Methodology/results: We conduct a lab experiment to elicit preferences and willingness to pay for three bundles of goods consisting of two high (H) type products and one low (L) type product that vary only in the position of low type product. We find that consistent with the satiation model, 53% of subjects prefer the HLH sequence, placing an approximately 2.7% greater ex-ante valuation over the second-best LHH sequence. The LHH sequence is optimal under the acclimation, and memory decay model and preferred by 31% of subjects. The front-loaded HHL sequence, optimal under the discounted expected utility model, is preferred by 16% of subjects. We estimate the parameters of the satiation model and find that satiation effects are significant with a half-life of 17 hours.
  Managerial implications: Using the calibrated satiation model, we optimize consumption for each period and find that, compared to a bundle with equal consumption in each period, the optimal bundle has a greater selling probability and achieves a revenue lift in excess of 4.5%.

Work-In-Progress:

• Designing and Comparing Custom Interventions to Mitigate Product Returns
• Modeling Customer Asset Balances: A Parametric Machine Learning Approach