Towards enhanced sales and operations planning : Using machine learning for decision support in an engineer-to-order context

Abstract: All companies deal with tactical planning questions and decisions, for example balance demand and supply, to be able to create an acceptable delivery ability without too much inventory or resources/capacities. For that, some companies use Sales and Operations Planning (S&OP) as their tactical planning process. The ongoing customization wave applies to more and more products and there is a general displacement from standard products, manufactured to stock, towards more customized ones where the product is either assembled-, manufactured-, or engineered-to-order (ETO). This displacement brings an increased complexity into tactical planning questions and decisions, which might be new to a company and must be handled efficiently. The use of S&OP in an ETO context is, however, rarely documented.The possibility for companies to store large amounts of data and the availability of technologies such as Machine Learning (ML) to make predictions, opens up for an improved decision support for S&OP. ML models are normally trained with large datasets, and this is a challenge in an ETO context since there are normally small datasets to work with. Moreover, the use of ML in S&OP and ETO contexts are rarely documented.The purpose of this thesis is, thus, to explore where and how ML can be a useful tool for tactical planning, such as in S&OP in an ETO context. This thesis takes the first steps toward using ML as a decision support for S&OP in an ETO context. Three studies have been performed to map the current state of ML in S&OP in ETO contexts, to understand the challenges and tasks connected to S&OP in an ETO context, and to explore some of the considerations required when implementing ML in S&OP in an ETO context.The main findings indicate that implementing ML in an ETO context with the purpose of improving S&OP requires an understanding of challenges and related tasks before starting any ML implementation projects. Further, considerations are required before starting to understand available data and to build data models. Tasks for ML must also be understood and agreed. Mechanisms behind occurring challenges need to be understood as well. What is driving trust for a technology and the business process is also important to understand and prepare for, ahead of an ML implementation. The results of the studies are (i) a model presenting the different parts of S&OP in an ETO context, (ii) specific challenges and related tasks, (iii) a model of critical aspects of trust connected to the process, the technology, and the combination of the two, and finally, (iv) a model for assisting in understanding the mechanisms behind capacity and load in engineering.