Skip to content
allanninal.dev/build
Collaborate

Part 4 of 7 · Capacity forecaster series ~5 min read

How far ahead is worth forecasting

The horizon is the design decision that determines whether a capacity forecast changes anything, and it is usually set by whatever the reporting tool defaults to.

Key takeaways

  • The minimum horizon is the longest lead time to add capacity, plus decision time.
  • Different responses have different lead times; report against the relevant one.
  • Beyond a certain distance the range is so wide it stops being informative — say so.
  • Report weeks of notice next to every shortfall.
  • A forecast inside the lead time is still useful, for a different set of actions.

Lead time sets the horizon

Lead times in weeks for five ways of adding capacityA bar chart with five bars showing weeks from decision to available capacity. Overtime: one week. Agency staff: three weeks. Subcontract: four weeks. Hire and train: ten weeks. Buy a machine: twenty weeks. A note says the horizon has to exceed the longest option you want to keep open.010203040~1Overtime~3Agency staff~4Subcontract~10Hire and train~20Buy a machineWeeks from decision to available capacityThe horizon has to exceed the longest option you want to keep open.
Fig 1. Five ways of adding capacity and how long each takes. A forecast horizon shorter than one of these removes that option before anybody has considered it.

The chart makes the design decision obvious. A twenty-six week horizon keeps every option available; a twelve-week horizon quietly removes machine purchase from the set of possible responses; a four-week horizon leaves overtime and agency staff, which are the two most expensive options per hour.

That is worth stating explicitly because it is a cost consequence of a reporting choice. Businesses that forecast short systematically pay peak rates for capacity they could have arranged more cheaply with notice.

Decision time is part of it

The lead time to hire is ten weeks after somebody decides to hire, and deciding takes time: a conversation, a budget approval, a job description. Adding two to four weeks of decision time to every lead time is realistic and is what makes the difference between a warning that is actionable and one that is theoretically actionable.

Where it stops meaning anything

How the useful limit of a capacity forecast horizon is determinedA vertical chain of five steps entered by a box labelled How far out, considered week by week. Step one measures order book coverage, asking how much is booked. Step two asks whether it is below about a third, meaning mostly extrapolation; if not it exits to Still forecast it, with a wide range. Step three asks whether the range is wider than capacity, so the answer is maybe; if not it exits to Useful, report it. Step four says so plainly: we cannot tell yet. Step five revisits when booked, since coverage grows weekly. A note says a forecast whose range spans both sides of capacity has not answered the question.AWS ACCOUNTHow far out?week by weekOrder book coveragehow much is booked?Below about a third?mostly extrapolationStill forecast itwith a wide rangenoRange wider than capacity?the answer is 'maybe'Usefulreport itnoSay so plainly'we cannot tell yet'Revisit when bookedcoverage grows weeklyA forecast whose range spans both sides of capacity has not answered the question.
Fig 2. How the useful end of the horizon is determined. The test in the third box is the honest one: if the range straddles capacity, the forecast has not said anything.
  • App integration
  • Machine learning
  • Management
  • Analytics

Saying ’we cannot tell yet’

This is a legitimate and underused output. A forecast for week thirty-eight that ranges from sixty per cent to a hundred and forty per cent of capacity has not answered anything, and presenting it as a line at a hundred per cent is worse than saying so.

It also has an action attached: the week to look again. Order book coverage grows steadily, and the useful statement is “we will be able to answer this in about six weeks”, which is itself a plan.

Notice period on every shortfall

The four numbers reported with every forecast capacity shortfallA horizontal row of five boxes. Shortfall in week twenty-six: forty hours. Notice: twenty-two weeks from today. Hire: ten weeks, plus three to decide. Nine weeks of slack: a real choice. Report all four, not just the first. A note says only the fourth one tells anybody whether this is a decision or an announcement.THE FOUR NUMBERSShortfall in wk 2640 hoursNotice: 22 weeksfrom todayHire: 10 weeksplus 3 to decide9 weeks of slacka real choiceReport all fournot just the firstOnly the fourth one tells anybody whether this is a decision or an announcement.
Fig 3. The four numbers that belong together. Reporting the shortfall alone leaves the reader to do the arithmetic that determines whether it matters.
  • Machine learning
  • Management
  • People

Inside the lead time

A shortfall discovered with less notice than the lead time is still worth reporting; it just changes which options are available. Overtime, rescheduling, subcontracting and talking to customers about dates all have short lead times and all are more expensive or more painful than planning ahead.

Reporting the slack as negative — “shortfall in six weeks, hiring takes thirteen, six weeks short of the lead time” — makes that explicit and stops the conversation starting with a suggestion that cannot work.

Next: finding out whether the forecast was any good.

All posts