Process mining before the robot: why order matters

The question that opens almost every automation project is which process to automate first. The answer usually comes from whoever speaks loudest, whoever owns the most visible process, or whoever already had a pilot on the shelf. None of these criteria are bad in bad faith — they are bad because they rest on perception, and perception does not measure effort.

The designed process and the executed process are different documents

Every mature operation has flowcharts. The problem is not missing documentation: it is that documentation describes intent, not execution. Between one step and the next lies a space where manual fixes, shadow spreadsheets, email approvals and back-and-forth between teams all live. That space appears on no map — and it is exactly what consumes the hours.

What is not measured is not merely invisible: it is systematically underestimated in investment decisions.

Process discovery inverts the order of the conversation. Instead of starting from a hypothesis and looking for confirming data, it starts from observed execution and lets the numbers point where to look. The result usually contradicts expectation: the bottleneck is rarely where the team thinks it is.

Three things that only surface after measuring

• Real effort per step, including time nobody records because it happens outside the system.
• Variation between runs — two identical cases taking completely different paths.
• Risk concentrated in specific people, only evident when the path is mapped by those who execute it.

With those three numbers on the table, prioritisation changes nature. It stops being a persuasion meeting and becomes a ranking: effort on the gain axis, complexity on the cost axis, risk as a filter. What stays at the top is what is worth automating first — and what sits at the bottom gains an explicit reason to wait.

Automating is not the end of measurement work

The baseline established during discovery is what allows the robot to be evaluated afterwards. Without it, automation gain becomes a retroactive estimate — the kind of number that survives the presentation and not the audit. With it, each cycle is comparable to the last, and the decision to expand or stop no longer depends on conviction.

That is why order matters. Not because measuring is nobler than automating, but because measuring afterwards means measuring against a reference that no longer exists.