Confidence-based optimization for inventory management at a local grocery store

Every morning, grocery store managers must order products to meet a demand that fluctuates strongly from one day to another. Managers face opportunity costs, i.e. lost business, if demand exceeds supply, and wastage cost if demand is low. In the literature this problem is known as the “Newsvendor” problem. Existing approaches to this problem assume that the probability of observing a certain demand throughout the day is fully or partially known; based on this information they identify an optimal order quantity that maximizes expected profit. However, in practice, managers of small retail shops track sales data and do not know this probability. Furthermore, managers do not usually trust automated procedures to make their decisions. They would rather prefer to know which decisions, based on available data, can be claimed suboptimal at a given confidence level. Then, for each remaining decision, they may want to analyse the associated confidence interval for the profit. Confidence-based optimization is a novel multidisciplinary approach that bridges this gap between theory and practice by blending statistics, operations research and computer science. In this project we aim to investigate the effectiveness of confidence-based optimization as a managerial decision support tool at a local grocery shop.

This is a one year project now close to completion.
1. A relational database has been produced for six months worth of data at a local grocery store; the database covers a large variety of products and has been used extensively for research-led education purposes.
1a. Two UG dissertations based on this dataset were produced. Results stemming from one of these dissertations were discussed during an interactive workshop that involved the retail catering operations manager and the system manager of the University of Edinburgh catering services.
1b. The dataset was also employed in the context of an MBA project.
2. A prototype decision support system (available online) based on the software package Mathematica implementing a confidence-based optimisation algorithm for stochastic inventory management has been produced.
2a. A PG dissertation on confidence-based optimisation has been developed.
3. A working paper (now published) on confidence-based optimisation has been utilised as a case study in the context of module MATH11010.
4. The PI delivered an invited plenary talk on confidence-based optimization at the 9th International Statistics Day Symposium (IGS 2014)
5. The PI delivered a talk at the IFORS 2014 conference
6. The PI published one research article on confidence based-optimization:
R. Rossi, S. A. Tarim, B. Hnich and S. Prestwich, "Confidence-based optimization for the Newsvendor problem", European Journal of Operational Research, Elsevier, Vol. x(x):xxxx-xxxx, 2014 (ABS 3*)
The PI also published other 5 related ISI research articles and 1 rigorously peer reviewed conference paper that benefited from the research supported by this grant.