Proactive prevention of on-time delivery KPI violations

Case company, ABB, measures its success in order fulfillment process with on-time delivery KPI. Trend of the KPI is followed on high level and past orders can be analyzed with company’s process mining tool. Despite the current analytics tools that are available, company’s stakeholders feel that there is room for improvement with regards to order management and analytics. Problems in KPI performance are being addressed reactively, rather than proactively. Reactive approach doesn’t prevent delivery failures, which might lead to negative outcomes. In the worst case, delivery failures could even lead to loss of customers. This master’s thesis evaluates current order performance analytics tools for case company and seeks to develop analytics processes and tools that will shift case company’s analytics focus from past to present. Theoretical framework for this thesis is built from the past approaches on KPI violation prediction. However, due to complexity of both data and processes as well as the global scope of research and tools to be implemented, found approaches are adapted to suit case company’s needs. Tree-based machine learning algorithms are used to implement predictive analysis that quantifies the risk of late delivery in terms of probability predictions. This analysis is supported by enhancements on company’s current diagnostic approach with association rule mining methods. Proactivity was also addressed with implementation of notification workflow, aimed to engage stakeholders for proactive analysis and adaption actions. Results of this study show that KPI violation predictions can be successful even in such a complex and global context. Emphasis on predictive analysis was placed on capturing KPI violations, which xGBoost algorithm was able to do sufficiently. Company’s KPI analysis focus could be turned from analysing past deliveries on to managing upcoming deliveries. Data mining approach on diagnostic analysis was found to complement predictive analysis but also to be hard to interpret on its own. Although we were able to implement a thorough data analysis process that shifts the analysis towards upcoming deliveries, our empirical research covered only the development of the analysis and related tools without proper deployment to business practices. Therefore, the overall success of the approach couldn’t yet be evaluated from practical perspective