Automating Performance Insights: Suggesting and Computing Process Performance Indicators from Event Logs

The increasing availability of event data tracing the execution of business processes represents an excellent opportunity for organizations to create relevant and measurable Process Performance Indicators (PPIs). PPIs are a tool to assess how well an organization achieves its key business objectives at an operational level and to support informed decision-making. To mitigate the risk of extracting mislabeled and misused PPIs with few or no connection with the available event data, in this paper, we present an approach and an implemented tool called PPIPilot to automatically suggest a list of measurable PPIs against a pursued organizational goal by providing an event log and a business process textual description as inputs. PPIPilot leverages the domain knowledge embedded in large-language models (LLMs) to suggest relevant PPIs from the event log and relies on the PPINAT definition model to compute them from the available data. We report on the results of a qualitative evaluation to investigate the feasibility and perceived usefulness of PPIPilot and a quantitative assessment to measure the extent to which PPIPilot is able to correctly suggest and compute PPIs from event logs.