Definition
Process mining reads the timestamps your systems record, such as created, assigned, updated, and closed, then reconstructs the actual paths cases took, how long each step really lasted, and how many variants of the process exist.
Evidence instead of recollection
Interviews give you what people remember and believe. Process mining gives you what happened. Both are useful, and only one of them can tell you that 34% of orders passed through an approval step nobody mentioned, or that the average handling time your team reports is a median hiding a long, expensive tail.
The mechanism is unglamorous: your systems already stamp a time on things. Ticket created. Assigned. Status changed. Closed. Sequence those events by case ID and the real process emerges, showing every path taken, how often, and how long each took.
The findings are consistently uncomfortable in the same three ways. There are more variants than anyone thought. Rework loops exist that nobody had named. And most elapsed time is waiting, not working.
What it reveals that mapping cannot
True variant count. A process people describe as having two paths routinely turns out to have forty. Most are rare; a handful matter and are consuming real capacity.
Rework loops. Cases that go backward, whether returned for more information, reopened, or reassigned, are usually invisible in interviews because each person only sees their own segment.
Actual durations, with the distribution. Not the average, which hides everything. The tail is where the cost is.
Where the time really goes. Almost always queueing rather than effort, which redirects improvement work away from making steps faster and toward removing handoffs.
What it requires, and why many businesses cannot use it
Process mining needs three things from your data: a case ID that persists across systems, timestamps on activities, and enough volume for patterns to be statistically real.
That third condition is easy. The first is where it falls apart. If the order becomes a ticket which becomes an invoice, and none of those share an identifier, you cannot follow a case across the systems it actually passed through. Assembling that linkage is often the bulk of the work, and if you undertake it, you get something valuable beyond mining, since you have made your operational data traceable end to end.
For lower-volume processes, or those that live mostly in email and spreadsheets, process mapping is the better tool. Mining needs digital footprints, and not all work leaves them.
How Automathing approaches it
We check whether the event data can actually support mining before recommending it, because the answer for many mid-sized businesses is not yet, and a mining project that becomes a data-linkage project without anyone saying so is how budgets get spent without a result. Where the data does support it, we use mining to find where the time goes and mapping to find out why, because only the second one tells you what to change.
Frequently asked questions
What is the difference between process mining and process mapping?
Mining reconstructs what happened from system logs: objective, at scale, and blind to anything not logged. Mapping documents the process through observation and interviews: subjective, slower, and able to capture the reasoning and workarounds no system records. Mining finds the problem; mapping usually explains it.
Do we need expensive software for process mining?
Dedicated platforms are genuinely useful at enterprise scale and are often overkill below it. For a single process, exporting event logs and analysing them with standard data tools answers most of the important questions. The expense that matters is rarely the licence. It is assembling clean, linkable event data in the first place.
How much data do we need?
Enough cases for a pattern to be meaningful rather than anecdotal, which means a few hundred at minimum for a single process, ideally covering several months so seasonal variation is visible. More important than volume is completeness: partial logs produce confident, wrong conclusions about paths that were simply not recorded.
What if our process runs on email and spreadsheets?
Then mining will not see it, because those leave no structured event trail. Map it instead. This is also a signal worth noticing: a core process with no digital footprint is one you cannot measure, audit, or improve with evidence, which is usually a stronger argument for changing where it runs than for buying a mining tool.