Detecting and Mitigating the Event Log Mutability Problem: At the Dutch Employee Insurance Agency (UWV)

(a) Situation faced: UWV is the Social Security Institute of the Netherlands. At UWV, event data was used to gain insight into the unemployment benefits process. Traditionally, event logs have been conceptualized as immutable append-only data streams. However, in the real world, process mining implementations are generated a posteriori from business process execution systems which may not store all temporal data changes. This contrast brings forth silent risks caused by the mutability of recorded process execution data. In order to investigate this phenomenon, six event logs were created covering the same observation period from different reference timestamps. The logs showed significant differences in the form of inserted, updated, and deleted events. (b) Action taken: Once the event log mutability problem was identified and sufficiently analyzed, three potential mitigation strategies were devised. Instead of using the monthly data mart for event log creation, an intermediate data layer was constructed based on weekly mutations from the information system. Though the possibility of unobserved events was not completely ruled out, the solution provides sufficient transparency into the process and its changes over time. (c) Results achieved: By structurally creating data extracts and event logs for the same observation period at different reference timestamps, we were able to identify the event log mutability problem. Having a much more detailed account of what happens to each entitlement has already been shown to have additional analysis benefits. (d) Lessons learned: For analyses to be trustworthy, they should be repeatable, reliable, and reproducible. The event log mutability problem that results from current event log creation practices may prevent this. Three risk mitigation strategies were proposed as inaction is not a viable option. Since our practice also identified the event log mutability problem in other processes and organizations, we integrated it into the core responsibilities of the process mining team.