Dwell Time and Why It Misleads
How long someone spent at each stop looks like a productivity measure. What it actually contains, and what happens when people are ranked by it.
Once geofence events exist, the duration between them is one subtraction away. That subtraction is where most of the harm in this field comes from.
What dwell time contains
The work itself.
Waiting: for the customer to answer, for access, for a part, for a lift.
Travel within the site, which on a large campus is substantial.
The customer talking, which is frequently the job rather than an interruption.
Diagnosis, which is invisible and looks like standing still.
Breaks, which are legal entitlements and appear as long dwells.
Position error and geofence size, which add or remove minutes at both ends.
Six of those seven are not the person's doing, and the measure does not separate them.
What happens when people are ranked by it
Jobs get rushed. The measure rewards leaving, not fixing.
Difficult jobs get avoided, because they wreck the number.
Callbacks rise, which costs more than the time saved and shows up in a different report that nobody connects.
Customer conversations get cut short, which is the part customers actually value.
Breaks get skipped or not recorded, which is a working-time problem.
And the number improves, which is why the practice persists.
The comparison problem
Rounds differ. Urban and rural, residential and commercial, new installs and repairs.
Allocation differs. Whoever gets the hard jobs has worse dwell figures by construction.
Which means a league table measures the work assigned, not the person doing it.
What to use instead
Job outcome: completed first time, callbacks, customer response.
Aggregate round analysis, not individual: does this round consistently overrun, and is it sized correctly?
Travel against on-site time in total, which is a routing finding rather than a person finding.
Ask the engineers. The people doing the round know why it overruns, and a fifteen-minute conversation produces better information than a quarter of dwell data.
Where dwell time is legitimately useful
In aggregate, for planning: how long does this class of job actually take, so rounds can be sized honestly.
For finding systematic waiting, which is an access or scheduling problem to fix.
For spotting a site that consistently consumes more time than quoted, which is a pricing finding.
All three are about the work, not about the worker, and all three survive the objections above.
The line to hold
Aggregate and by job type: useful.
Per person, compared: not supportable, and the data does not become more accurate by being averaged over more jobs, because the biases are systematic rather than random.
If dwell time is being requested per person, ask what decision it would inform — and whether job outcomes would inform it better.
Ask what decision it informs
The test that ends most requests for this measure.
If the answer is "we would know who is slow", ask what would be done.
Usually: a conversation. Which could have happened without the measure and would have produced the reason.
If the answer is round sizing, that is legitimate — and it needs aggregates by job type, not a per-person table.
The decision determines the shape of the data, and a measure with no decision behind it is being collected because it was available.
Check the difficult case
Use this professional-services case to frame one representative test for this issue. The useful evidence is the record created when a worker challenges an event, a manager reviews it and an administrator exports it.
Independent reference
For an external point of reference, see Google Maps. A widely used mapping service can provide a comparison point, but arrival estimates still need operational context.