Automating Operational Business Decisions Using Artificial Intelligence: an Industrial Case Study

Abstract

The process of making business decisions is increasingly reliant upon analyzing very large data-sets. Due to the amount of decisions having to be made on a daily basis, this becomes time-consuming and expensive to carry out manually. The purpose of this thesis was to determine whether using Artificial Intelligence to automate business decisions is feasible. This was done by carrying out a proof of concept project at IFS World, a software company developing Enterprise Resource Planning systems. Procurement decision making was chosen as a case for this study. Automating these decisions can not only result in speeding up the decision making process, but also in making more accurate decisions. To achieve this, three machine learning algorithms were proposed. Their goal was to learn preferences from historical procurement data and apply this knowledge to new situations. Prototyped versions of the algorithms were developed, tested and compared using both real-world and artificial datasets. The results showed that after a short period of supervised learning, two algorithms were able to make decisions automatically, with a low error-rate. Furthermore, sensitivity analysis showed that the algorithms are robust enough to recover from errors in the training data. The study also revealed several constraints and prerequisites related to feature selection, data freshness, and completeness. It was concluded that automating operational business decisions using Artificial Intelligence is achievable if certain preconditions are met. It can provide several advantages over manual decision making: it will speed up the decision making process, and can, in certain scenarios, improve the quality of the decisions.