Process mining uses event data to examine how a process runs in practice. For a business considering a pilot, the useful question is whether the available records can support a decision: where work waits, which routes repeat and which changes deserve investigation.
The technique does not observe everything employees do. Its view is limited by the systems, events and timestamps included in the data. A successful pilot makes those limits explicit.
For the core concepts and differences between discovery and conformance checking, see process mining. This guide focuses on preparing and evaluating a first project.
Choose one decision to support
Start with a process that has an identifiable outcome and a responsible owner. Examples include invoice approval, order fulfilment, service requests and purchase-to-pay.
Translate a broad concern into a question. “Finance is slow” is difficult to test. “Which approval routes account for the longest elapsed time before payment?” gives the analysis a defined purpose.
Agree what action could follow a finding. If no one can change the rules, staffing, system configuration or workflow, the pilot may produce an interesting diagram without an operational result.
Check whether the data represent the process
A conventional event log needs a case identifier, an activity and a timestamp. Additional fields can describe department, order value or other context, provided their use is justified.
| Field | Example | Question to check |
|---|---|---|
| Case identifier | Invoice ID | Does it consistently identify the same case across systems? |
| Activity | Approval requested | Does the event describe a business step or only a technical update? |
| Timestamp | Time recorded by the application | Does it represent the action, a later batch update or an import? |
| Context | Business unit | Is it accurate, necessary and available for the relevant cases? |
An ERP does not necessarily record every action. Work may happen in email, spreadsheets or conversations. Events may be missing, duplicated, overwritten or recorded at different levels of detail.
Where objects have many-to-many relationships, such as orders, deliveries and invoices, a single case ID may oversimplify the process. Consider the data model before treating a generated map as a complete representation.
Prepare and validate a sample
Extract a limited sample before committing to a larger integration. Inspect missing identifiers, inconsistent activity names, time zones, duplicate events and records arriving out of sequence.
Trace selected cases with the people who handled them. Check whether the event sequence matches their understanding and explain any gaps. This helps distinguish a genuine exception from a logging artefact.
Separate elapsed time from active work. The interval between two events may include a queue, a weekend or work that the log does not show. It should not automatically become a labour-cost estimate. Activity-based costing can provide a complementary model where its assumptions are explicit.
Choose tools against the use case
Tools such as Celonis, SAP Signavio, Microsoft Power Automate Process Mining, Apromore, Disco and ProM provide different environments for analysing event data. Their current capabilities, licences and deployment options need checking against the proposed project.
Assess data import, modelling, conformance analysis, access control, export and the ability to explain findings to process owners. Check whether a feature requires a separate licence or service; do not assume it is included in an existing office-software subscription.
A proof of concept should use representative data and a defined question. A polished demonstration with sample data does not establish that your own event log is usable.
Investigate findings before changing the process
A map can reveal recurring loops, unusual routes and long waiting periods. Those patterns require interpretation. An extra approval may be avoidable rework, or it may be a necessary control for a particular transaction.
Compare similar cases before drawing conclusions. Value, urgency, customer type and seasonality can affect both route and duration. A correlation between a route and a poor outcome does not prove that the route caused it.
For a proposed change, record the explanation, the owner, the expected effect and any control that must be preserved. Test the change in a suitable scope and measure the outcome against a defined baseline.
Evaluate value without inventing ROI
Process mining does not produce a universal percentage saving or payback period. Value depends on data access, the cost of analysis, the changes implemented and whether the organisation can realise the benefits.
Distinguish released capacity from cash savings. Saving staff time can improve throughput or service, but it only reduces expenditure under specific operational conditions. Faster invoicing can affect working capital without producing the same amount of accounting profit.
Include licences, integration, data preparation, analysis, implementation of changes and ongoing monitoring in the cost model. Use a scenario with explicit assumptions, then update it from observed results. See ROI of process automation for the calculation issues.
Protect personal data and employee trust
Event logs can contain names, account identifiers and information about work patterns. Pseudonymisation can reduce exposure but does not automatically remove GDPR obligations.
Define the purpose, lawful basis, access rules, retention and appropriate transparency measures. Prefer process-level questions over unnecessary individual monitoring. Involve the responsible privacy and employee-relations specialists where the analysis affects people.
A practical pilot checklist
- Name the process owner and the decision the analysis will support.
- Define the process boundary and the relevant outcomes.
- Inspect a sample event log and document its limitations.
- Agree privacy, access and retention requirements.
- Validate the model with people who know the work.
- Select a change with an owner and acceptance measures.
- Review results before extending the scope.
Within Q-Journey, an audit can help assess a process and identify appropriate next steps. Process mining is one possible method, chosen where the data and question justify it. Express and Platinum scopes are agreed individually; Platinum also includes Lean and Muda consulting. Talk to QA10 about your process.
FAQ
Can process mining work without changing the ERP?
Sometimes existing exports or APIs provide enough data. Other projects need additional logging, access configuration or integration work. Assess the actual systems before promising an unchanged ERP.
How long does a first project take?
The schedule depends on data access, quality, process complexity and the decision being tested. Agree an initial sample and acceptance criteria before estimating the wider project.
Does process mining show how much time employees spend working?
Not automatically. Timestamps usually show when recorded events occurred. The time between them can include waiting, non-working hours and unrecorded activity. Estimating active labour requires additional evidence.
Are pseudonymised event logs outside the GDPR?
No. Pseudonymised records remain personal data when they can be linked back to people. The analysis still needs appropriate purpose, legal basis, access controls and other applicable safeguards.