Skip to content
allanninal.dev/build
Collaborate

Part 3 of 7 · Capacity forecaster series ~5 min read

Why the peak matters more than the average

A business running at seventy per cent average utilisation sounds comfortable and misses deliveries eight weeks a year, because the seventy per cent is made of forty-per-cent weeks and hundred-and-twenty-per-cent weeks.

Key takeaways

  • Average utilisation of seventy per cent routinely contains weeks above a hundred.
  • Forecast a range: a likely case and a busy case, both dated.
  • Seasonality usually explains more than trend at a one-year horizon.
  • Work that can be moved between weeks softens a peak; work that cannot does not.
  • Express the answer as weeks over capacity, not as average utilisation.

The average hides it

Five weeks of demand against a fixed weekly capacityA bar chart with five bars showing hours of demand against one hundred and eighty-two available. Week one: one hundred and twenty. Week two: one hundred and sixty-eight. Week three: two hundred and ten. Week four: one hundred and ninety-six. Week five: one hundred and forty-two. A note says the average is one hundred and sixty-seven against one hundred and eighty-two available, and two of the five weeks were impossible.0100200300400~120Wk 1~168Wk 2~210Wk 3~196Wk 4~142Wk 5Hours of demand against 182 availableAverage 167 against 182 available. Two of the five weeks were impossible.
Fig 1. Five weeks whose average looks comfortable. Weeks three and four exceeded capacity and the average conceals both.

The reporting consequence is direct: the headline number should be weeks over capacity, not average utilisation. “Two of the next twelve weeks are above capacity, peaking at fifteen per cent over in week three” is actionable. “Utilisation is projected at ninety-two per cent” is not.

Some work moves and some does not

A peak matters less when the work can be pulled forward or pushed back. Stock production can be built early; a booked installation on a customer’s site cannot. The forecast is considerably more useful when it distinguishes the two.

The practical version is to tag demand as fixed or movable at whatever granularity already exists, and report the peak twice: as booked, and after smoothing the movable work across adjacent weeks. The gap between those two numbers is the value of being able to reschedule.

A range, not a line

How a demand forecast range is constructedA vertical chain of five steps entered by a box labelled Known work, confirmed orders. Step one adds likely conversions from quotes at their historical rate. Step two adds the usual late arrivals from history, with a side box saying to measure it rather than guess. Step three applies seasonality from three years of data. Step four produces likely and busy cases, two lines rather than one. Step five widens the range with distance, since week twenty is vaguer than week four. A note says the second box is the one people leave out and it is often a third of the work.AWS ACCOUNTKnown workconfirmed ordersPlus likely conversionsquotes, at their ratePlus the usual late arrivalsfrom historyHow muchmeasure it, do not guessApply seasonalityfrom three yearsLikely and busy casestwo lines, not oneWiden with distanceweek 20 is vaguer than week 4The second box is the one people leave out, and it is often a third of the work.
Fig 2. How a demand range is built. Late-arriving work is a measurable quantity in most businesses and omitting it produces a forecast that is consistently low.
  • App integration
  • Management
  • Analytics
  • Front-end & mobile

Late arrivals are measurable

Every business has work that arrives inside the forecast horizon and was not in the order book when the forecast was made: repeat customers, urgent jobs, warranty work. In many operations it is twenty to forty per cent of the total and it is entirely predictable in aggregate.

Measuring it is straightforward: for a given past week, compare what was in the order book four weeks earlier with what was actually delivered. The difference, averaged over a year, is the uplift to apply. It is one of those quantities that everybody knows exists and nobody has quantified.

Seasonality beats trend

At a horizon of a year or less, the seasonal pattern usually explains far more variation than any growth trend, and it is easier to estimate: three years of weekly data gives a seasonal shape that is more reliable than an extrapolated trend line.

A forecast that applies a growth percentage to a flat baseline will systematically miss the weeks that matter, because the weeks that matter are seasonal peaks rather than a gradual rise.

How to state it

The output, in four lines

  • Weeks above capacity: 3 of the next 26, all in January.
  • Peak shortfall: 40 hours in week 4 of January, likely case; 68 hours in the busy case.
  • After smoothing movable work: peak falls to 22 hours, still over.
  • Confidence: based on 3 years of seasonal data and a measured 27% late-arrival uplift.
  • Weeks of notice: 22. Lead time to add capacity: 10.
  • No recommendation. Hiring, contracting and rescheduling are commercial choices.

The fifth line is what turns this into a decision rather than a chart. Twenty-two weeks of notice against a ten-week lead time means there is a real choice available, and the same finding at eight weeks of notice would be a different conversation entirely.

Next: how far ahead is worth forecasting.

All posts