Some measurements are available and should not be taken. Ruling them out explicitly at design time is cheaper than removing them after deployment.
Time away from the workstation
Technically easy and the most damaging single measure in this field.
It cannot distinguish a toilet visit, a safety check, helping a colleague, fetching equipment, or a legitimate break.
Measuring it produces people avoiding necessary breaks, which has direct health consequences and is under regulatory attention in several jurisdictions.
Where idle time genuinely matters operationally, measure it at zone level and investigate the process cause rather than the person.
Bathroom and welfare area activity
Should be technically excluded from every system, not merely policy-excluded.
No operational question requires it.
In several jurisdictions, monitoring it is unlawful, and everywhere it is indefensible.
Exclude the areas at the sensor configuration level so the data does not exist.
Continuous individual rate with automated consequences
The arrangement most associated with injury and with regulatory action.
Rate data is legitimate for planning at team level.
Automated warnings escalating to discipline remove human judgement from a decision that requires context, and in several jurisdictions automated decisions with significant effects require human involvement and a route to contest.
Behavioural and emotional inference
Video analytics claiming to detect attitude, engagement, fatigue or stress from posture or expression.
The underlying science is contested, the error rates on real workforces are poor, and the errors are not evenly distributed.
No legitimate operational question requires it.
Decline it, including when a vendor demonstrates it convincingly.
Biometric identification for productivity
Facial recognition of employees for anything other than a narrowly justified access control purpose.
Strongest legal restrictions in most jurisdictions.
No operational justification in a facility where people already carry credentials.
Private communications and personal devices
Content of conversations.
Personal phone activity, including presence of a personal device.
Social media, in any form.
None of these has an operational purpose and collecting them is a serious breach of the boundary between the employer's interest and a person's life.
Health inference
Deriving health conclusions from movement, pace or posture data.
Health data is a special category with stronger requirements in most regimes, and consent obtained from an employee by their employer is frequently not considered freely given.
Where ergonomic risk assessment is the purpose, do it as an assessment by a qualified person, with the worker's participation, rather than by inference from surveillance data.
The design-time test
For each proposed measure, three questions.
What operational decision does this inform? If none, drop it.
Can the same decision be informed at team or zone level? If yes, do that.
Would you be comfortable explaining this measure, in detail, to the people it applies to? If not, that discomfort is the answer.
Apply the test before deployment, because removing a measure after it exists requires admitting it should not have been collected, which is a conversation most organisations avoid by keeping it.
Excluding areas technically
Policy exclusions fail; configuration exclusions hold.
Welfare facilities, break rooms, medical rooms, changing areas excluded at the sensor level.
Not filtered in the reporting software, where the data still exists and can be unfiltered.
Camera fields of view configured so the area is not captured at all.
Positioning zones defined to omit them, so no trace exists.
Publish the exclusion list, which is what makes the assurance credible and which invites correction if something was missed.
The design-time refusal
Ruling a measure out before deployment is far cheaper than removing it afterwards.
Removing a measure later requires admitting it should not have been collected, which most organisations avoid by keeping it.
So the decision must be made at design time, in writing, with reasons.
Record what was ruled out and why, which also answers the question when a vendor or a manager proposes it again next year.
Include it in the consultation, because a list of what will not be measured is more reassuring than any list of what will.