There is a sales pitch that goes like this: install our process-mining platform, point it at your ERP, and the automation candidates will reveal themselves. The pitch is true at the high end. It is also wrong for many mid-market clients, where the platform cost can dwarf the engagement it is supposed to inform.
What you actually need
- An event log with a case ID, an activity name, and a timestamp. Three columns, anything else is bonus.
- An experienced analyst who can read the data and the business: process mining without business context returns noise.
- A short, fixed window. The most useful findings show up early; the rest of the time is for sharpening the pitch.
The five questions a spreadsheet can answer
- What is the actual sequence, in frequency order? (Pivot by activity-pair.)
- Where do cases stall? (Time-between-events distribution per stage.)
- How many distinct paths exist? (Sequence hashing: if the answer is 'thousands' for a should-be-linear process, you have a finding.)
- Which exceptions consume the most time? (Group by deviation, sum elapsed.)
- Which step has the highest rework rate? (Self-loops, repeat events.)
When the platform is right
Celonis, UiPath Process Mining, and the rest are real value above three thresholds: very large data volumes, ongoing (not one-off) monitoring, and an automation program already mature enough to consume continuous discovery output. Below those thresholds, the platform becomes a status object: bought, deployed, rarely opened.
The output that actually matters
Whether you used a spreadsheet or a platform, the deliverable is the same: a ranked backlog of automation candidates with frequency, elapsed time, and exception rate per candidate. That is the artifact the rest of the program runs on. Anything else is supporting evidence.