What the Analysis Cannot Tell You

An analysis that answers everything asked of it is answering some things badly. Defending the boundary is what keeps it trusted on the rest.

Why something happened

Why not: the log records sequence, not reason.

Use instead: asking the people who handle those cases, which is also how you discover what the log got wrong.

Whether a deviation was wrong

Why not: conformance measures difference from a model, and the model may be the thing that is wrong.

Use instead: a conversation about whether the standard applies to those cases. Frequently it does not.

How much effort a step took

Why not: duration includes waiting unless start and end timestamps exist separately.

Use instead: better logging, or a scoped observation study. Do not present duration as effort.

Whether the outcome was correct

Why not: the log shows completion, not quality. Errors that pass every check are invisible.

Use instead: downstream signals — complaints, returns, corrections — joined to the case.

What happens outside the systems

Why not: email, spreadsheets and conversations produce no events.

Use instead: asking, observation, or a narrowly scoped task mining study for a specific named gap.

Whether a person is performing well

Why not: case duration reflects case difficulty, queue position, handover partners and system availability far more than the individual.

Use instead: normal management. Keep the mining output out of it, which is both an obligation question and the thing that preserves cooperation.

What a change will achieve

Why not: simulation output inherits every assumption the analyst supplied, and the assumptions usually dominate.

Use instead: a small trial with before-and-after from the same log, which is the technique's real strength.

Whether the process is fully covered

Why not: coverage is measured against the log, which is itself incomplete.

Use instead: state the model coverage as a proportion, name the known unrecorded steps, and stop claiming completeness.

How to decline

Name what the log does support, which is usually adjacent and substantial.

Name what would answer the question and roughly what it would cost.

Write the limitation into the report, not the covering note, because the report is what circulates.

Do not produce a weak number with a caveat, since the caveat is dropped in the retelling.

What it does support

Which paths cases actually take, and in what proportion.

How long each step takes and how long cases wait between steps.

Where cases go backwards, and how much that costs.

Where the intended controls were not applied, with case numbers.

How many handovers a case passes through.

Whether a change moved any of the above.

Six things answered reliably is worth the programme.

The declaration attached to every figure

Five lines that keep the reporting honest.

The boundary: which activities start and end the case.

The period, and the exclusion rate.

Whether incomplete cases are included.

Whether durations are elapsed or working time.

The mapping version.

A figure circulated without these will be compared against another figure computed differently, and the disagreement will be attributed to the data rather than to the definitions.

Declining a request

The specific skill that keeps the programme trusted.

Name what the log does support, which is usually adjacent and useful.

Name what would answer the question and roughly what it costs.

Offer the alternative method: asking, observation, better logging.

Do not produce a weak number with a caveat, because the caveat is dropped in the second retelling.

Write the limitation into the report, not the covering note, since the report is what circulates.

Declining a request properly

The skill that keeps the function trusted on everything else.

Name what the log does support, which is usually adjacent and substantial.

Name what would answer the question and roughly what it would cost.

Offer a range where a range is honest.

Offer the alternative method: observation, interviews, a bounded study.

Write the limitation into the report, not the covering email, because the report is what circulates and the caveat is dropped by the second retelling.