Three columns are enough to start process mining
Case id, activity, timestamp. Everything else is convenience. What this minimal contract buys you, and what skipping it costs.
A usable event log fits in three columns: which case we are talking about, what happened to it, and when. From there, an engine can rebuild the real paths, measure waiting times and spot rework.
The contract
- case_id: what follows the case end to end: an order number, not a line number.
- activity: the name of the step, stable over time.
- timestamp: when the step happened, with its time zone.
The case_id traps
The choice of identifier decides what the analysis can see. Taken at order-line level, it splits one case into as many cases as it has items, and the measured lead times lose their meaning. Taken too high, at customer level, it lumps unrelated cases together.
The right granularity is the one the business uses when talking about the case. If the order desk says “order 4412”, the order carries the case.
