Matching inventory and demand in a Fast Moving Consumer Goods company: A Decision Support System

Abstract

Demand fulfilment and its core, order promising, is a key business function that directly impacts the company’s ability to generate revenue through sales. For companies operating under a Make-To-Stock (MTS) production strategy, it is tantamount to matching both on-hand and projected stock to customer demand. The prevalent manner to carry out this task is to use either First-Come-First-Served policies, or simple allocation rules embedded in commercially available software. However, order promising is, in general, quite complex due to uncertainties both in demand and supply, and it can become extremely challenging when demand exceeds the stock. Therefore, several contributions have established the advantages of order promising using optimisation models and/or sophisticated algorithms. However, these works assume that customers can be clustered according to their profitability/importance and it remains open to test whether this also holds for homogeneous customers. Furthermore, due to the exploratory nature of most studies, they assume rather stylised scenarios (such as e.g. single product-orders, partial shipments) that do not clarify how model-based order promising would perform in a realistic setting.In this paper we present a decision framework for order promising in a Fast Moving Consumer Goods (FMCG) company in the brewery sector where customers cannot be clustered and that incorporates many realistic features (such as e.g. multi-product orders, on-full delivery, or perishability constraints). The decision framework relies on several iterations of a MILP model that allocates the stock to customer and reserves some portion of the stock for orders arriving later on. These functionalities are embedded in a Decision Support System (DSS) that also helps with the renegotiation of the orders that cannot be served. Results reported from the usage of the DSS show its ability to obtain high service levels and to favourably compare to other order promising policies. Furthermore, in order to analyse the instance sizes that can be solved in reasonable time, and the parameters that influence the performance and results of the MILP model, an extensive computational experience has been carried out using a testbed that covers a wide range of business scenarios.