Your first process map

Open Promenade. The start screen offers a demo log — take Order-to-Cash, a conventional event log of about 2,800 orders. Nothing is uploaded; the file is parsed on your own machine and stored in your browser’s local storage.

Import, then look before you mine

After import you get an artifact: the log itself, sitting in the tree on the left. Before discovering anything, open it and look at the summary — number of cases, number of activities, the time range, and the most frequent variants.

Two minutes here saves an hour later. If the time range starts in 1970, some timestamps failed to parse. If there are 40,000 distinct activities, the activity column probably contains an ID rather than a name.

Discover a model

With the log selected, run Discover → Inductive model. This gives you a process model in a few seconds.

What just happened: the miner read the order of activities across every case, found which activities follow which, and folded that into a model of the process as a whole.

Why the Inductive Miner is the default

It always returns a sound model — one with no deadlocks, where every activity can be reached and every case can reach the end. Other miners can produce models that look plausible but describe something no case could actually do. Starting from a sound model means anything strange you see is a fact about your data, not an artifact of the algorithm.

Now run a different miner on the same log

Run Discover → Directly-Follows Graph on the same log. You get a different picture: every activity that ever directly followed another, with counts on the arrows.

Compare them. The DFG is closer to the raw data — it hides nothing, which also means it shows every rare path and can look like a hairball. The Inductive model is an interpretation — it generalises, which makes it readable but means it may allow paths no case actually took.

Both are correct. They are answering slightly different questions:

• DFG — “what transitions occurred, and how often?”
• Inductive model — “what process would produce behavior like this?”

Getting used to holding both is most of what separates a first-week analysis from a confident one.

Your results form a chain

Notice the tree on the left: the models appear underneath the log. Promenade records where every artifact came from — which action produced it, from which input, with which parameters. Change the noise threshold and re-run, and you get a second model alongside the first rather than silently replacing it.

That matters later, when someone asks how you produced a number in a slide.

Next

You have a model. Learn to read it — including the parts worth distrusting.