The Cost in Cooperation

Task mining measures behaviour by watching it, which changes it. The size of the change depends entirely on how the study is read by the people in it.

The observation effect

People work differently while watched, which is not deception but ordinary human response.

They follow the documented process more closely, which conceals exactly the workarounds the study was meant to find.

They avoid personal use, which is fine, and they also avoid legitimate pauses.

They work faster, which distorts every duration.

Expect the first days to be unrepresentative, and discard them.

What makes it worse

Deployment without explanation.

Screenshots, which feel qualitatively different from event counts and produce the largest behavioural change.

Any suggestion of comparison between participants.

Uncertainty about duration, since a study with no end date is read as monitoring.

A previous deployment that drifted, anywhere in the organisation.

What reduces it

A stated question and a stated end date.

Volunteers, genuinely.

Showing people their own data.

Event counts rather than content, and metadata rather than screenshots, wherever the question allows.

A written purpose limitation, honoured in the first hard case.

Reporting the findings to participants first.

The data consequence

A study read as surveillance produces a model of the documented process, which the organisation already had.

Which means the intrusion bought nothing and cost the cooperation needed for the next study.

This is the operational argument, independent of the obligations, and it persuades people who are unmoved by the legal ones.

Measuring whether trust held

Participation rate in the next study, which is the clearest signal available.

Whether participants raised issues during this one, which they will only do if it feels safe.

Whether the model matched what people described in interviews, since a large divergence suggests observed behaviour.

Whether anyone asked to see their own data, which is a healthy sign rather than a suspicious one.

The sustainable arrangement

Occasional, short, question-driven studies with volunteers, whose raw data is deleted.

Not a standing capability. A licensed agent sitting on every desktop, dormant but installed, is read as monitoring whether or not it is collecting, and it has that cost without any of the benefit.

The measures that show whether trust held

Four signals, none of which appear in the capture data.

Participation rate in the next study, the clearest indicator available.

Whether participants raised issues during this one, which they only do if it feels safe.

Divergence between the captured model and what people described in interviews, since a large gap suggests observed behaviour rather than normal work.

Whether anyone asked to see their own data, which is healthy rather than suspicious.

Watch all four, because the capture itself will look fine either way.

Deciding the hard case in advance

Someone will ask to use the captures for something else, and the answer given then becomes the policy.

Name who decides, before deployment.

Write down the answer for the predictable requests: a performance concern, an investigation, a comparison between participants.

If the answer is yes to any, say so in the consultation, because participants are entitled to know before they volunteer.

If the answer is no, the refusal has to hold in the first real case, which is when it is tested and when it becomes either credible or worthless.

The first hard case

Someone will propose using the data for something outside its stated purpose, and that decision becomes the real policy.

Decide in advance who decides, because deciding under pressure produces the wrong answer.

If the answer is yes, everyone learns what the system is for within a week, and the cooperation the analysis depends on ends.

If the answer is no, and the reason is stated publicly, the assurance becomes credible in a way no document achieves.

Record the decision either way, so the next request has a precedent.

Expect the request within the first year. It always comes.