Process Mining: how BPM went from manual labor to data science
Technologies for automatically extracting process models from the event logs of enterprise systems — Celonis, ProM, Disco — and their effect on the analyst's profession.
For decades process analysis remained a craft of manual labor. The analyst interviewed process participants, drew diagrams, discussed and corrected them, then checked reality through sample interviews and observations. The cycle took weeks and months, and the result always remained subjective.
Process mining — a technology born in the early 2000s and gone mainstream by the 2020s — radically changes this. The real picture of a process is extracted directly from the data of enterprise systems, bypassing interviews.
What process mining is
Process mining is a family of techniques based on one idea: every operation in a modern enterprise system leaves a trace in the logs. SAP, Oracle, ERP, CRM, ITSM — all of them log events with timestamps, object identifiers and attributes.
If you gather these logs in the right format, an algorithm can automatically reconstruct the process — which steps were performed, in what order, how often, and with what deviations from the expected path.
The basic data format
The minimal format needed for process mining is a table with three columns:
| Case ID | Activity | Timestamp |
|---|---|---|
| INV-001 | Request created | 2025-01-15 09:14 |
| INV-001 | Approved | 2025-01-15 11:32 |
| INV-001 | Paid | 2025-01-15 14:05 |
| INV-002 | Request created | 2025-01-15 09:22 |
| INV-002 | Returned for rework | 2025-01-15 13:18 |
| INV-002 | Approved | 2025-01-16 10:44 |
| INV-002 | Paid | 2025-01-16 16:20 |
From such a table the algorithm can build a process model and enrich it with statistics: the average time between steps, the frequency of different routes, the distribution of delays.
Three types of analysis
Process mining is traditionally divided into three types:
Discovery
Reconstructing the “as-is” process model from the logs without any prior model. The result is a notation (usually Petri nets or BPMN) reflecting the real behavior of the process.
Conformance checking
Comparing the real execution of a process with a reference model. Finding deviations: which steps are skipped, which are performed in the wrong order, which “extra” ones are added.
Enhancement
Augmenting an existing model with information from the logs: timing statistics, the frequency of different paths, bottlenecks.
Platforms
The process-mining market is consolidated around a few key players:
- Celonis — one of the largest commercial players, which grew into a “unicorn” in the 2010s
- UiPath Process Mining — acquired from ProcessGold
- SAP Signavio — the Signavio platform, absorbed by SAP
- IBM Process Mining — the acquired myInvenio
- ProM — an academic open-source tool developed in Eindhoven
- Disco (Fluxicon) — a compact commercial tool for analysts
The effect on the profession
Process mining changes the role of the business analyst. Where the main skill used to be conducting interviews and drawing diagrams, now the following are added:
- Working with large data — SQL, basic ETL
- Knowledge of enterprise-system formats (SAP, Oracle, Salesforce)
- An understanding of process-discovery algorithms
- Interpreting statistical distributions
This shift moves the profession closer to a data analyst than to a traditional business analyst.
Limitations
Process mining is a powerful technology but not a panacea:
- The quality of the result depends directly on the quality of the data in the logs
- Loosely structured processes (case management) are less amenable to automatic discovery
- Processes spread across several systems require complex data integration
- Analysis requires context — the algorithm sees the data but not the reasons
The future
By 2030 process mining will be a standard part of the BPM stack alongside modeling and automation. AI assistants already help interpret analysis results, generate hypotheses about the causes of deviations and suggest optimizations.
This does not eliminate the analyst’s role but fundamentally changes it. Whoever, by 2030, commands both classic BPM skills and data-analytics tools will be at the peak of the profession.