The event log

Everything starts with an event log: a table where each row records that an activity happened to a case at a time — an invoice was received, matched, approved, paid. Add attributes (amount, supplier, approver, plant) and you have the raw material for discovery. The quality of the log decides the quality of everything after it: missing timestamps, inconsistent activity names and orphaned cases are the usual first week of any project.

Discovery, conformance, performance

• Discovery builds a model of the process from the log — a directly-follows graph or a more formal model — showing which activities follow which, how often and how fast.
• Conformance compares the discovered behavior to a reference model: skipped steps, unauthorized paths, repeated approvals. This is how audit and compliance teams use mining.
• Performance analysis overlays time: where cases wait, which paths are slow, which variants are expensive. Variant analysis ranks the hundreds of real paths against the one that was designed.

The object-centric extension

Classic mining forces every event into one case. Real processes are not shaped like that: one order produces several invoices, each with a delivery, each with a payment. Flattening them into a single case notion either duplicates events or hides relationships. Object-centric process mining (OCPM) keeps the objects distinct — order, invoice, delivery, payment — and records which objects each event touches. The OCEL 2.0 standard defines the format. The result is a picture that matches the business: one order with three invoices and two deliveries is one connected trace, not three logs.

Where mining stops

Mining answers what happened and why, at the population level. It does not, by itself, answer three questions that operations care about most:

• Which of the cases that are open right now are about to go wrong?
• What should be done about each of them, and is it allowed?
• Did the action work, compared with doing nothing?

Many process-mining programs stall exactly here: the diagnosis is excellent, the improvement cycle is slow, and the value is realized by a transformation project rather than by the process itself, day by day.

From mining to intelligence to action

Process intelligence extends mining with context (objects, KPIs, policies, roles, systems) and with prediction (case-level foresight). Orchestration adds the governed action. Operational memory closes the loop by recording each prediction, decision, action and result, so the next round is better than the last. Mining remains the evidence base for all of it — which is why it is the first step of the loop, not a separate product.