Suitability check — is my case log fit for a process analysis?
A process analysis run on an unfit log produces figures nobody can stand behind. The customer pays for a statement about their procurement process, receives an invented one, and only notices once they have acted on it. That is why this check comes before the analysis, not after — and why no governance conformance on case data is offered without a usable result.
You will find it under Governance Mining → Suitability check. The tab appears only for tenant admins and platform administrators.
What you need
Section titled “What you need”An extract from the system whose value stream you want analysed — a ticketing system, a procurement application, a document management system. Four columns, in this order, separated by semicolon, comma or tab:
| Column | Content | Example |
|---|---|---|
| Identifier | the case number — what holds all events of one case together | A-1042 |
| Activity | what happened | Approval granted |
| Timestamp | when | 2026-06-01T09:12:00Z |
| Actor | who — an identifier is enough, the real name is not needed | m.huber |
A header row is detected and discarded. Below 100 events there is no verdict: that would be a statement about the sample, not about the log.
What is measured — and what falls away without each measure
Section titled “What is measured — and what falls away without each measure”| Measure | Without it, this is impossible |
|---|---|
| Case identifier coverage | everything. Without a case there is no case view, only a list of events |
| Activity variety | any statement about the value stream. A log with two activities describes a status, not a value stream |
| Timestamp granularity | any statement about sequence |
| Case completeness | any statement about completeness — a case consisting of a single event has no value stream |
| Actor resolvability | any statement about accountability and segregation of duties |
The most common pitfall: the format looks fine, the values do not
Section titled “The most common pitfall: the format looks fine, the values do not”Timestamp granularity is determined from the values, not from the format. A system can emit impeccable ISO 8601 with seconds and still set every value to midnight — because it records the posting date and not the posting time. Granularity is then one day, and the sequence within a day cannot be established. With financial and approval data this is the rule, not the exception.
The three verdicts
Section titled “The three verdicts”| Verdict | Meaning | What becomes possible |
|---|---|---|
| Conformance possible | identifier, variety, granularity, completeness and actor are all sufficient | statements about sequence, actor and rule: “runs as modelled in 63 % of cases, without approval in 21 %” |
| Metrics only | everything except the actor is sufficient | frequencies, value stream variants, cycle times per case type — but no statement about accountability |
| Not meaningful | one of the four load-bearing measures is insufficient | The check names which one. Often another extract fixes it |
“Metrics only” is not half a no. It is the most common real case, and the statements that remain are the most frequently used ones. There are three levels rather than two precisely so that this case does not collapse into “unfit”.
What happens to your data — and what does not
Section titled “What happens to your data — and what does not”The sample is not stored. It is evaluated in memory and then discarded. The result contains measures and sentences about the sample, but no case identifier, no activity label and no actor from it. No data holding is created, no log entry about the content, and no transfer to third parties — not to an AI provider either.
This is not frugality but the purpose of the design: the check must be able to take place before an instruction, a works agreement or an impact assessment exists. A log entry reading “tenant X checked a recruitment log” would already be a record of an intention.
Four case types are blocked
Section titled “Four case types are blocked”Not switched off by default — not selectable:
- Recruitment procedures — concern people outside the employment relationship and permit no resolution to roles
- Health-related cases, including absences — special categories under Art. 9 GDPR
- Disciplinary cases — relate to individual employees; a process analysis of them would be behavioural monitoring
- Whistleblowing and grievance procedures — any analysis could point back to the reporting person
If you declare such a case type, the check is refused with a reason — without evaluating the sample. This is expressly not a red light: a red light means “with better data it will work”, this refusal means “this type of case is not analysed, regardless of data quality”. The block cannot be lifted, either in your tenant or by us.
If a term from this list appears in your activity labels, a hint is shown, not an abort. “Complaint management” is a legitimate value stream name; a hard abort would lead to renaming rather than reflection.
And then?
Section titled “And then?”The report is a result even if nothing follows from it: “our log carries no case identifier in 40 % of cases, and the timestamps are rounded to the day” is an insight into your own systems that nobody had before.
If the verdict is green or amber and you do want to set up the analysis, that is a separately commissioned service. It engages co-determination under sections 96 and 96a of the Austrian Labour Relations Act (ArbVG) and, as a rule, a data protection impact assessment under Art. 35 GDPR — the report from this check makes a good first annex for it.
Related topics
Section titled “Related topics”- Governance Mining — findings about your own model, without an external source
- What is analysed and what is not — the full exclusion list
- My data in analytics — the self-disclosure