Skip to content
allanninal.dev/build
Collaborate

Part 2 of 7 · Capacity forecaster series ~5 min read

How the real capacity gets measured

The number everybody uses for capacity is headcount multiplied by hours, and it has never once been achieved. Measuring what actually gets delivered is both easy and uncomfortable.

Key takeaways

  • Measure delivered output per period over a year; that is your capacity.
  • The gap is holiday, sickness, setup, rework and everything unbookable.
  • Do not try to eliminate the gap in the forecast. Forecast against reality.
  • Identify the bottleneck by which stage has the least headroom.
  • The bottleneck moves with the product mix, so re-check it.

Where the hours go

Paper capacity against measured capacity for one teamA stacked bar chart with two bars in hours per week. Four series: delivered productive hours in green, holiday sickness and training in purple, setup changeover and waiting in orange, and rework and unbooked time in red. Paper capacity is two hundred and forty hours, all counted as delivered. Measured capacity is one hundred and eighty-two delivered, plus twenty-two hours of holiday and sickness, twenty hours of setup, and sixteen hours of rework. A note says the green bar is your capacity, and the other three are real and are not going away.0100200300400~240Paper~240MeasuredDelivered productive hoursHoliday, sickness, trainingSetup, changeover, waitingRework and unbooked timeThe green bar is your capacity. The other three are real and are not going away.
Fig 1. Paper capacity against measured capacity for the same team. The three coloured bands are the difference and each of them is a normal part of operating.

The temptation on seeing that chart is to treat the coloured bands as waste to be eliminated, and some of it is. But the forecast has to be built on what actually happens, not on what would happen after an improvement programme that has not been done.

The two questions are separate and both worth asking. What will we deliver next January uses a hundred and eighty-two. Whether a hundred and eighty-two could be a hundred and ninety-five is an improvement project with its own timescale and its own uncertainty.

Measuring output, not effort

Capacity is measured in whatever unit the business already tracks: jobs completed, hours billed, units produced, appointments delivered. The important property is that it comes from something already recorded rather than from a new timekeeping exercise nobody will sustain.

A year of that data, expressed per week, gives both the average and the spread, and the spread matters as much as the level. A team that delivers between a hundred and sixty and two hundred hours depending on the week is planning against a different number from one that reliably delivers a hundred and eighty-two.

The bottleneck

How the constraining stage in a process is identifiedA vertical chain of five steps entered by a box labelled The work has stages: quote, build, test and ship. Step one measures capacity at each stage separately. Step two measures demand at each, which is not the same everywhere. Step three asks which has the least headroom, and that is the constraint. Step four forecasts against it, since the others do not bind. Step five re-checks quarterly because it moves, with a side box explaining that the product mix changes. A note says adding capacity anywhere except the constraint changes nothing at all.AWS ACCOUNTThe work has stagesquote, build, test, shipCapacity at eachmeasured separatelyDemand at eachnot the same everywhereLeast headroom?that is the constraintForecast against itthe others do not bindRe-check quarterlyit movesWhythe product mix changesAdding capacity anywhere except the constraint changes nothing at all.
Fig 2. How the constraint is identified. The last box is the one that gets skipped, and a bottleneck analysis from two years ago is often describing a stage that is no longer the problem.
  • Compute
  • App integration
  • Management
  • Analytics
  • Front-end & mobile

The bottleneck moves

This is the part that makes capacity forecasting harder than it looks. A workshop constrained by fitting hours becomes constrained by testing when the product mix shifts towards something that needs more testing, and the forecast built against fitting hours stops predicting anything.

The practical response is to forecast against every stage rather than only the current constraint, and to report which one binds first. That is barely more work and it catches the case where a stage that has never been a problem becomes one.

Capacity that is not people

The same method applies to machines, vehicles, rooms, bays and ovens, and for those the gap between paper and actual has different components: breakdowns, changeover, cleaning, and maintenance from Day 118. A machine available a hundred and sixty-eight hours a week on paper is available considerably fewer in practice, and the deferred maintenance debt is one of the reasons.

Where the constraint is a machine, the lead time to add capacity is a purchase and an installation rather than a hire, which is usually longer, which makes the horizon question in Part 4 more acute rather than less.

Next: why the average is the wrong thing to forecast.

All posts