Inventory and Service Optimization for Self-serve Kiosks

Abstract

In the retail industry, labor costs constitute a big chunk of total operating costs and retailers are advancing towards process automation to minimize their operating costs and to provide reliable services to their customers. One such example of technological advancement is self-service kiosks that are becoming an integral part of our life, whether it be for cashing a cheque, self-checkout at retail stores, airports, hospitals, or checkout-free stores. Although self-serve kiosks are cost-effective due to low setup and operating costs, the technology is relatively new and poses new research questions that have not been studied before. This thesis explores and addresses strategic and operational challenges associated with self-serve kiosk technology.

The first part of the thesis is based on a collaboration with MedAvail Technologies Inc., a Canada-based healthcare technology company, developing self-serve pharmacy kiosk technology to dispense over-the-counter and prescription drugs. MedAvail faces several challenges related to assortment and stocking decisions of medications in the kiosk due to its limited capacity and the thousands of drugs being ordered in various quantities. We address these challenges by analyzing pharmaceutical sales data and developing a data-driven stochastic optimization approach to determine optimized kiosk storage capacity and service levels and recommend assortment and stocking decisions under supplier-driven product substitution. A column-generation based heuristic approach is also proposed to solve the models efficiently.

The second part of the thesis extends the self-serve kiosk inventory planning problem to a robust optimization (RO) framework under fill rate maximization objective. We propose a data-driven approach to generate polyhedral uncertainty sets from hierarchical clustering and the resulting RO model is solved using column-and-constraint generation and conservative approximation solution methodologies. The proposed robust framework is tested on actual pharmacy sales data and randomly generated instances with 1600 products. The robust solutions outperform stochastic solutions with an increase in out-of-sample fill rate of 5.8%, on average, and of up to 17%.

Finally, the third part of the thesis deals with an application of pharmacy kiosks to improve healthcare access in rural regions. We present a mathematical function to model customer healthcare accessibility as the expected travel distance when multiple pharmacy location (store and kiosks) choices are available to customers. Customer choice behavior is modelled using a multinomial logit (MNL) model where customer utility for a pharmacy location depends on travel distance which is not exactly known but rather depends on kiosk fill rate. We model the problem as a newsvendor problem with fill-rate dependent demand to decide on kiosk stock level (or capacity) to minimize the weighted sum of expected travel distance and total cost. Sensitivity analysis over modelling parameters is carried out to derive insights and to determine problem settings under which pharmacy kiosks improve customer accessibility.