Using process mining to diagnose bottlenecks, detect rework, and identify automation opportunities.
While discovery and conformance checking reveal what is happening, enhancement addresses how to improve. Process enhancement uses the event log to extend or improve an existing process model with performance data — adding timing information, identifying bottlenecks, detecting rework patterns, and quantifying the impact of variability (Van der Aalst, 2016, Ch. 8).
Enhancement is where process mining delivers its most tangible business value. An organisation might already know its order-to-cash process is slow; process enhancement pinpoints which specific step causes the delay, for which customer segments, and with what financial impact. This evidence-based approach to improvement replaces intuition and opinion with data-driven prioritisation.
Bottlenecks are identified by analysing waiting times between activities. If event logs capture both the start and completion timestamps of each activity, the time between the completion of one activity and the start of the next represents waiting or handover time. High waiting times at a specific transition indicate a bottleneck (Dumas et al., 2018).
More sophisticated analysis uses social network mining — examining which resources or departments handle which activities and how work flows between them. Bottlenecks may not be individual activities but rather organisational handoffs: work queues up when one team cannot process output fast enough. Process mining makes these invisible constraints visible.
Rework occurs when an activity is executed multiple times for the same case — for example, an invoice approved, rejected, resubmitted, and approved again. Event logs capture rework naturally because each activity instance has a timestamp. By analysing loop patterns, process mining quantifies the frequency and cost of rework (Van der Aalst, 2016, Ch. 9).
Rework detection is often eye-opening for organisations. Common patterns include approval loops (a manager repeatedly rejects and requests revision), error-correction loops (data entry errors caught downstream and sent back), and exception handling (cases that deviate from the standard path and take significantly longer). Each rework loop has a measurable cost in time, labour, and customer satisfaction.
Not all process steps are equally suited for automation. Process mining supports automation decisions by identifying activities that are highly standardised (low variant count), rule-based (decisions depend on clear conditions), and high-volume (significant labour cost). These characteristics can be extracted directly from event logs (Dumas et al., 2018).
For example, if 95% of invoice approvals follow the same path with minimal variation, that activity is a strong candidate for robotic process automation (RPA). Conversely, activities involving frequent exceptions or judgement calls may be better suited for decision support tools rather than full automation. Process mining provides the data to make this distinction systematically.
Your event log shows that the "Manager Approval" step in a travel reimbursement process has a median duration of 8 days, while the actual approval takes 2 minutes. The remaining 7 days 23 hours is waiting time in the manager's queue.
What interventions could reduce this bottleneck? Would you prioritise automation, delegation, or process redesign? How could you measure the impact of your intervention using the same event log data?