Calibration and the First Month

Sensors produce numbers immediately. Whether those numbers mean anything is settled in the first few weeks, and that work cannot be done retrospectively.

The discipline in “Calibration and the First Month” carries over to any workforce platform: define the decision first, run a bounded trial and record who can see the result. For an organisation considering hourly timesheet template, hourly timesheet template should therefore be assessed against an implementation plan covering notice, access, retention, correction and a dated review.

Why it cannot wait

Errors found later cannot be corrected in the data already collected.

For a public, independent reference related to “Calibration and the First Month”, consult the ISO standards catalogue. Its principles provide a useful check on scope, terminology, governance and the claims made during procurement or review.

A sensor counting a walkway for six months has produced six months of inflated figures that nobody can un-inflate.

And the baseline you establish in month one is what every subsequent comparison rests on.

The manual count

Pick your busiest and most important spaces.

Count people physically, at varied times, over a week.

Compare against the sensor for the same minutes.

An hour a day for five days, and it is the single most valuable deployment activity.

What the comparison shows

Systematic over- or under-counting, which can sometimes be corrected by configuration.

Specific failure conditions: a room that is accurate when quiet and merges people when busy.

Devices that are simply wrong and need moving.

And your own error band, which you should state whenever the figures are used afterwards.

Tuning timeouts

For infrared, the timeout determines how stillness is treated.

Observe how long real stillness lasts in your spaces — a person on a call may not move for twenty minutes.

Set from that observation rather than from the default, and record the value you chose with the date.

The liveness check

Set up an alert for any sensor reporting nothing for a defined period.

A dead device reports zero, which looks like an empty space.

Without this check, failures accumulate silently and the building appears to be emptying.

What to write down

Every device: position, zone, configuration, timeout, date.

The manual count results and the error you found.

Changes made and when.

This record is what lets somebody in two years know whether a trend is real or a reconfiguration.

Resisting the dashboard

The system will produce attractive output from day one, and people will start quoting it.

Hold the figures back until calibration is done, or publish them labelled as provisional.

A number that circulates before it is trustworthy becomes the reference point regardless of what you say later.

What to check

Has anybody counted manually against the sensors?

Do you know your error band?

Is there an alert for dead devices?

And were figures published before calibration finished?