Most process diagrams in a company show how a flow was meant to run. Process mining shows how it actually runs. It reads the log data of the systems involved, the trail of timestamps and case numbers that every click in ERP, CRM or ticketing leaves behind, and reassembles the real path of each individual case. In the end you do not see the clean target picture from the workshop but the real course with all its loops, back-jumps and waiting times.

That is powerful because it is incorruptible. An interview gathers opinions about how a process works, and people remember the normal case better than the special one. The logs remember everything. They show the case that was handed back and forth between two departments three times, and the approval that sits for four days on average. Those invisible patterns are often the real bottleneck.

Process mining shows you where your process bleeds. It does not put on a bandage.

The catch sits in that last sentence. Process mining is a diagnosis, not an intervention. It measures and makes visible, but it improves nothing on its own. Whoever buys the tool and then expects improvement has skipped the middle part: the decision about what to do with the finding, and the work of carrying it out.

The value of process mining stands or falls on three conditions, and all three must come together. The first is volume. A process that runs a hundred times a day has statistically solid patterns. One that runs twice a month has anecdotes. The second is variance. If every case runs the same way, you need no data analysis to know the path. Value arises where cases take different routes and nobody quite knows why or how often anymore.

The third condition is the hardest: usable logs. Process mining can only reconstruct what the systems recorded. If part of the flow runs by word of mouth, email or a spreadsheet on a shared drive, the trace is missing exactly there. Then the analysis shows a process with blind spots, and the blind spots are often the places you actually care about.

There is a whole range of flows for which process mining is simply too big. A process that three people handle and that you can watch from start to finish in one afternoon needs no data reconstruction. You see the bottleneck with your own eyes. Putting a tool in between costs licences, data connections and weeks, and delivers in the end the same insight a conversation would have given in an hour.

The second case is the process with no digital trace. If the flow half happens in heads, phone calls and notes, process mining has nothing to read. The analysis would show only the digital remainder and leave out the decisive part. Here the honest tool is not the software but observation, backed by a simple measurement of times and error rates.

Even when process mining fits in the end, it rarely has to be the first step. The cheaper entry begins with a clear question: what do you actually want to know? Are you looking for the bottleneck in one specific flow, or do you want to map your whole process landscape? The first question is often answered by a targeted measurement, the second is a programme you do not start on a hunch.

Start with what you can find out without a licence. A documented baseline, a few interviews with the people who handle the process every day, and a look at the numbers you already have take you surprisingly far. If you still have a question afterwards that only the logs can answer, say why a quarter of the cases take a detour, then you have a real reason for process mining and a clear scope to go with it.

• Start with the question, not the tool: what exactly do you want to find out?
• Measure the baseline on the real flow before you connect a platform.
• Check the logs: do the necessary traces even exist in full?
• Bound the scope to one process with volume and variance, not everything at once.
• Plan the step after: mining finds the diagnosis, you make the improvement.

Core sentence: process mining is a strong diagnostic tool for large, tangled flows with clean data, and an expensive detour for anything you could just look at directly.

Process mining improves the process.

It measures and shows, no more. The improvement is a separate decision with its own work behind it. A finding alone has never lowered a cycle time.

More data automatically means more insight.

Only if the data captures the relevant part of the flow. If the trace is missing at the interesting spot, a bigger dataset only produces a detailed picture of the wrong section.

A process diagram replaces the analysis.

The drawn target picture shows intent, not reality. The gap between the two is exactly why process mining exists at all.

Process mining, Business process management, Baseline, Data quality.

From what size does process mining pay off?

It is not company size that decides but the process: if it runs often, has real variants and full logs exist, the analysis can pay off even in a mid-sized firm. A rare, uniform flow never justifies it, no matter how large the house.

What if our data is patchy?

Then process mining shows a patchy picture, and the gaps often sit exactly where it gets interesting. In that case the first step is to make the process digitally traceable, not the analysis itself. Otherwise you analyse the part that already runs cleanly.

Do we need an expensive platform for process mining?

Not necessarily as the first thing. Many questions can be answered with a targeted evaluation of existing log data. The platform pays off when you need the same view regularly and across several processes, not for a one-off diagnosis.

No external market figures or unapproved client results are asserted as verified. The scope of an analysis is checked against the real data before implementation.