The Middle Horizon, Where Most Decisions Live
The four-to-sixteen week range is where most workforce decisions are actually taken, what it needs from the forecast, and why it gets the least attention.
Twelve weeks of one team's forecast, with the levers priced beside it
Weeks 1-2
118
±5
Roster published; nothing movable.
Weeks 3-4
120
±9
Overtime at 1.5x covers about 4%.
Weeks 5-8
124
±15
Agency at 2.1x covers about 9%; two weeks' notice.
Weeks 9-12
127
±21
Short-lead recruitment still possible.
Only the last three rows have a lever attached, and the levers get dearer as the band gets narrower. The cheapest decisions are the ones taken furthest out, which is the opposite of how the pack was ordered. One employer's own figures, not a benchmark.
The middle horizon — roughly four to sixteen weeks — is where most workforce decisions are actually taken, and it is the horizon most organisations serve worst. It is too far out for the near method and too close in for the annual one, so it tends to be produced as a by-product of whichever of those exists.
The horizon in “The Middle Horizon, Where Most Decisions Live” should determine which current records are relevant and how much detail the forecast can honestly support. A planning team evaluating see how the workflow is configured for download time tracking software can connect recent time and project context with short-term capacity questions, while keeping longer-range assumptions in a separately governed model.
It is also the horizon where improvement pays most, because the decisions here are frequent, consequential and reversible enough to benefit from a better number.
The assumptions behind “The Middle Horizon, Where Most Decisions Live” can also be tested against Planful financial-planning resources. Record which guidance applies to the available series and which does not.
What is decided here
Overtime commitments. Agency bookings. Training course starts. Recruitment for short-lead roles. Project start dates. Shift pattern trials. Holiday approval policy for a known peak.
- Overtime commitments and agency bookings.
- Training course starts and short-lead recruitment.
- Project start dates and shift pattern trials.
- Holiday approval policy for a known peak.
Seven decision types, all with real money attached, all taken on whatever number happens to be available.
What the horizon needs
Weekly granularity, a stated band, and the known calendar already folded in — campaigns, product dates, school holidays, public holidays, anything booked.
A monthly figure is too coarse: most of the decisions here are about which weeks, not which months.
Why it falls between two stools
The near forecast is built for rostering and does not extend. The annual forecast is built for budgeting and does not disaggregate. The middle is served by whichever one is stretched, and stretching either is a known failure mode.
The remedy is to build it, not to derive it.
What it is built from
Trend and season from the long series, the weekly shape from the recent data, and the event calendar from whoever maintains it. Three inputs, combined explicitly.
The combination is not technically demanding; the work is in getting the event calendar, which usually lives in somebody's head.
The event calendar is the hard part
Marketing sends, product launches, billing dates, regulatory deadlines, system changes, known outages. None of this is in the demand history and all of it moves the number.
Obtaining it is a standing arrangement with three or four people, not a data problem, and it is the single highest-value thing a middle-horizon forecaster does.
The band at this horizon
Typically eight to twenty per cent, widening across the range. Narrow enough to be useful for most of the decisions listed above, wide enough that publishing a point alone is misleading.
This is the horizon where the band earns its keep: the decisions are reversible, so the right response to a wide band is often to wait and re-forecast rather than to commit.
Updating rather than re-forecasting
Weekly rolling is right here. Each week the horizon moves forward one week and the near end firms up, which is exactly how the decisions get taken.
A monthly refresh at this horizon means decisions are taken on numbers up to four weeks stale, which at twelve weeks out is a third of the distance.
What the decision maker should be shown
The weeks, the band, and the lever prices: what overtime costs, what agency costs, what each would cover. The forecast without the lever prices is only half an input.
This is the horizon where the planning function is closest to being useful, and the lever prices are what close the gap.
The commitment that outlives the information
A six-week agency booking made at twelve weeks out, on a figure that has since moved. The booking is rarely revisited because nobody owns revisiting it.
A standing review at the point where each commitment becomes cancellable is worth more than most modelling improvements.
The error record here
Weekly, with the direction, and split between the level and the distribution across weeks. Getting the four-week total right and the distribution wrong is a specific failure with a specific cause, usually a misplaced event.
Where the middle horizon fails quietly
When the event calendar lapses. The method keeps working, the error widens by a few points, and because nobody is attributing error to causes it looks like general deterioration.
Checking the calendar is the first thing to do when middle-horizon accuracy drifts, and it is almost never the first thing anybody does.
Who owns this horizon
Frequently nobody, which is the root of the neglect. The rostering team owns the fortnight and finance owns the year, and the twelve-week view belongs to whoever happens to need it.
Naming an owner is usually the whole intervention required, because the inputs and the method are not difficult.
What it is worth building properly
In most operations, more than the other two horizons combined, because the decisions are frequent and the levers are real. It is also the cheapest to improve, because the inputs exist and are simply not being collected.