Process mining finds what the workshop was never going to tell you
Interview-based process mapping produces the process people believe they follow. The event log produces the one they actually run. The gap averaged 34% of activity in our last twelve diagnostics.
I spent fifteen years running plants before I did any of this, and in that time I sat through a lot of process mapping workshops. They produce a genuinely useful artefact: a picture of how experienced people believe the process works, which is roughly the intended design plus the workarounds they are willing to admit to.
What they cannot produce is the actual path work takes.
The gap, measured
Across our last twelve operational diagnostics we mapped the process both ways — workshops first, then event logs from the underlying systems. On average, 34% of activity by volume followed a path that appeared nowhere on the workshop map.
Some of that is trivial: minor variants, edge cases nobody thought to mention. But the recurring pattern is more interesting. The high-volume hidden paths are almost always rework loops — work returning to an earlier step — and rework is systematically invisible in workshops because the people describing the process think of it as an exception rather than a step.
At one client, an approval step that the workshop described as taking "a day or two" had a median duration of 1.2 days and a 91st percentile of 19 days, because 23% of cases went back for information and re-entered the queue at the bottom. Nobody was wrong in the workshop. The median was a day or two. The cost was entirely in the tail, and the tail was invisible.
People describe the happy path accurately and the exception path not at all. The money is nearly always in the exception path.
What this does not mean
It does not mean skip the workshops. The event log tells you what happened; it cannot tell you why, and it has no opinion about which variants are legitimate. We have twice found "rework loops" in the data that turned out to be a correct and necessary control, obvious to anyone who does the job.
The sequence that works is: mine first, workshop second, and bring the variant analysis into the room. Asking "why do 23% of these come back?" gets you a far better conversation than asking people to describe their process from memory — and it takes an afternoon instead of a fortnight.
The practical obstacle
The real barrier is not analytical, it is access. Getting transaction-level event data out of four ERP instances takes longer than the analysis does, and it usually requires an executive to overrule a system owner who would rather not. Budget six weeks and get that sponsorship before you start, or the diagnostic becomes a workshop by default.