Activity (process mining)

What is an activity in process mining?

An activity is the type of step a process event describes. Order created, Credit checked, and Invoice paid are activity labels. Each real occurrence of one of those steps is a separate event.

In an event log, events that share the same label share the same activity. The case ID tells you which process instance the event belongs to, and the timestamp tells you when it happened.

The activity label sets the level of detail on your process map. A label that is too technical fills the map with system updates. A label that is too broad hides the decisions and waiting time you actually want to study.

Activity versus event

An activity is the kind of step. An event is one concrete occurrence of it. Invoice approved is the activity. The approval of invoice F-1042 on Tuesday at 14:03 is an event.

The relationship is like a category and its members. Thousands of invoices can carry the same activity label, but every event has its own case, timestamp, and attributes.

Process discovery reads those labels across many events to find recurring patterns and build a process model. The activity is the vocabulary that discovery works with.

Choosing the right granularity

Say an employee updates a customer address. Underneath, the system might record Validation started, Record loaded, Street changed, Postcode changed, and Record saved.

For a business analysis, Address changed may be all you need. For an investigation into postcode validation errors, the finer steps suddenly matter.

Pick the granularity from the question you are asking. Activities should be recognisable to the people who run the process, and they should separate steps whose order, duration, or outcome makes a difference.

Do not change the granularity quietly halfway through a measurement period. If three labels are later merged into one activity, your variants and timing figures shift, and old and new numbers stop being comparable.

Activity versus status and task

An activity says something happened. A status says how a case or object sits at a given moment. In progress can hold for days, while Handling started is a single activity event.

A task is work that still needs doing. Task assigned and Task completed can be activities around that task, but not every task is modelled as one clean event.

A useful habit is to name recorded events with a verb in the past tense, such as Document received. It signals that the label describes observed behaviour, not an open instruction.

The lifecycle of an activity

An activity that takes time can have both a start event and a completion event. XES, the IEEE standard for event logs, carries lifecycle information so you can tell the start and the completion of the same activity apart.

With both timestamps you can separate processing time from waiting time. A single completion timestamp is enough to put events in order, but not enough to measure how long the work itself actually took.

Check that starts and completions can be matched to each other correctly. When several instances of the same activity run in parallel, you need an extra identifier to pair them.

Managing naming and mapping

Keep an activity dictionary that records each label, what it means, which source fields feed it, how its timestamp is defined, and who owns it. Write down your filters and merge rules there too.

Different source systems often name the same step differently. Only normalise labels when the business meaning is genuinely the same. Approved by system and Approved by manager may stand for two different controls, and collapsing them hides a real distinction.

Test your labels with the people who do the work. They will often spot that a tidy technical mapping has merged two separate actions into one.

Version your mapping and keep the data lineage behind it. An activity is an analytical definition layered on top of your event data, not a property that comes for free with every source table.