Nobody has ever left a board meeting excited about process mining. There's no demo that makes people gasp. It doesn't write poems or talk back.
It is also, in my experience, the single most useful thing a business can do before it spends serious money on automation or AI. So let me try to make the case.
What it is
Every system your business runs keeps a record of what happened. An order was created at this time. It was approved at this time, by this person. It was changed. It was sent back. It was approved again. It was invoiced.
Those records are sitting in your ERP, your CRM, your claims system, your ticketing tool. Nobody looks at them, because they're millions of rows of timestamps.
Process mining reads those rows and draws you a picture. Every route that work actually takes through your business, how often each route is used, how long each step takes and, crucially, how long work sits waiting between steps.
It isn't a workshop opinion or a consultant's diagram. It's what your own systems say happened.
What it shows you
The first time a leadership team sees their process mined, the room goes quiet for a moment. It's usually the same three surprises.
There isn't one process. There are hundreds. The procedure document shows one path. The data shows the happy path handles a fraction of the volume, and the rest goes through dozens of variations nobody designed.
The time isn't where you think. Teams assume the slow part is the work. The data nearly always shows the slow part is the waiting. Something sits in a queue for six days, gets worked on for ten minutes, and sits for another four.
The loops. Work that goes backwards. Sent back for more information, rejected and resubmitted, corrected after the fact. Every loop is cost, and most businesses have never counted them.
Why it has to come before AI
If you automate a process you haven't measured, three things go wrong.
You automate the path in the procedure document, and find it covers a minority of cases. The exceptions still land on people, and now those people also have to supervise a bot.
You have no baseline. Six months later someone asks what the project delivered, and the honest answer is that it feels faster.
And you miss the cheap wins. A good share of what mining finds doesn't need technology to fix. An approval step that adds five days and never rejects anything. A field that's always filled in wrong because the form is confusing. You can fix those next week for nothing.
What it doesn't do
It's worth being clear about the limits.
Mining only sees what systems record. The work done in email, spreadsheets and conversations is invisible to it. That's why it should always be paired with time spent alongside the people doing the job. Take the mined picture to them and ask what's happening in the gaps. Data without people misses the workarounds. People without data gives you opinions.
It also depends on the data being there. You need a case identifier, an activity and a timestamp. Most modern systems have them. Some older ones need work to get them out.
The commercial case
A mining exercise costs a small fraction of an AI programme. It tells you where to aim that programme, gives you the baseline to prove it worked, and usually pays for itself in the no-technology fixes it turns up along the way.
It's the first stage of most ETT engagements. It will never be the exciting part. It's the part that makes the exciting part work.