Estimating how long a job will take

Duration estimates sit underneath the entire maintenance operation. They determine how many jobs fit in a day, whether the planned schedule is achievable, what a contractor should charge, and whether the department is adequately staffed. For a separate product-side perspective on rounding and payroll time calculations, this payroll guide provides additional detail.

They are also, in most organisations, inherited numbers that have never been compared to what actually happened.

Estimate the whole visit

The commonest error is estimating the task rather than the visit. Changing a belt takes twenty minutes; the visit takes ninety. For a wider operational and compliance reference, consult the Better Buildings O&M guide.

A duration estimate that will survive contact with the week has to include the parts of the visit that are not the task:

• Travel to the location, which on a multi-building site is substantial and on a multi-site portfolio can exceed the task itself.
• Access — finding the key, locating the person who holds it, waiting for a room to be free.
• Isolation, lock-off and any permit process.
• Collecting parts and tools, unless staged in advance.
• The task.
• Reinstatement, testing and proving.
• Recording: closing the job, noting what was found, filing a certificate.

Use the median and the spread

An average duration is a poor planning input because durations are skewed: most jobs cluster near a typical value and a minority take far longer, pulling the mean upward.

The median describes the typical job. The spread describes the risk. A task with a median of forty minutes and a range of thirty-five to fifty can be scheduled tightly; the same median with a range of twenty-five to four hours cannot, and the schedule needs to be built around that variability rather than around the middle of it.

Find out what drives the long tail

When a task has a wide spread, something is causing the outliers, and it is usually identifiable: a particular asset location, an access constraint, a specific failure mode, one plant room where everything takes longer.

Splitting the task into two estimates — the standard case and the known difficult case — is more accurate than a single wide estimate and immediately more useful for planning. It also surfaces problems worth fixing, since 'everything in that plant room takes twice as long' usually has a physical cause.

Capture actuals as a by-product

Comparing estimate to actual requires actuals, which requires time to be recorded against jobs. This is the point at which the exercise usually stalls, because asking technicians to record time feels like an imposition added to the job.

It is far more acceptable when the recording is a by-product of work the technician already does — starting and completing the job in whatever system they are using — rather than a separate timesheet. And it is only acceptable at all if the data is visibly used to improve estimates rather than to assess individuals.

Review estimates once a year

For the twenty or thirty tasks that make up the bulk of the volume, compare the estimate to the median actual. Adjust the estimates that are consistently wrong.

This is a short annual exercise with a disproportionate effect: every downstream calculation improves at once, and the schedule stops being systematically over-committed in the same places it has been over-committed for years.

Estimate the unknown honestly

Some jobs genuinely cannot be estimated before someone has looked — an intermittent fault, an unfamiliar asset, a symptom with several possible causes.

The right response is a two-stage job: a bounded investigation with its own duration, then an estimate for the work. Forcing a single number onto an unknown produces an estimate that is wrong, a job that overruns, and a data point that corrupts the average for every similar job afterwards.