Process mining is an X-ray for operations. It shows the process as it runs, with every detour and every wait, using data the company already has. Like an X-ray, it tells you where the problem is. It does not operate.
How does process mining work?
Every business system keeps records. When an order is created, approved, shipped or invoiced, a timestamped entry lands in a database. Process mining collects those entries into an event log, where each row needs three fields: a case identifier, an activity and a timestamp. From that log, algorithms rebuild the process.
The field's work falls into three groups:
- Discovery builds a process model from the log alone, showing the paths cases actually took.
- Conformance checking compares the log against an intended model and flags where cases deviated.
- Enhancement, also described as performance analysis, adds performance data, such as processing times, waiting times and costs, to the model.
The term was coined by the Dutch computer scientist Wil van der Aalst, and the field emerged at Eindhoven University of Technology around 1999.
What can process mining tell you about where AI fits?
Mining output answers questions that decide where automation and AI belong:
| What the mining shows | What it suggests |
|---|---|
| A step every case passes through, done by a person, with consistent outcomes | A candidate for a model or a rule |
| Cases looping back to an earlier step | Missing information upstream, a redesign problem |
| Long waits between two steps, short work at each | Handoff and queue design, often no AI needed |
| Many rare variants with high cost each | Exceptions that should stay with skilled people |
| Deviations from the intended process | Either a control problem or a better process nobody wrote down |
That is why mining belongs to the See step of the AI implementation method, before any decision about what to build.
What can't it tell you?
Mining sees only what systems log. It cannot see the email thread where a supervisor approved an exception, the call where a customer explained the real problem or the spreadsheet where an analyst reconciles two systems every Monday. It cannot tell you why a step exists or whether a decision was right. A process can look clean in the data and still depend on one person's knowledge.
A mining tool does not change the process either. A dashboard of variants sits unused until someone redesigns the work and builds the change, whether that is an internal team or an AI automation services provider. And mining cannot tell you what the process should be. That takes judgment, the job of business process reengineering and redesign.
How does it relate to process intelligence and RPA?
Process intelligence applies mining continuously and adds business context. RPA sits at the other end: it does the work instead of observing it. The process mining vs RPA comparison explains why the order matters.