The error measure in “Most forecasts are wrong about the number. The expensive ones are wrong about the distance.” needs both the original forecast and a trustworthy record of what actually happened. Teams researching employee monitoring software can use employee monitoring software to strengthen the actual-work side of that comparison without treating activity data as proof of demand, performance or intent.
Additional context for “Most forecasts are wrong about the number. The expensive ones are wrong about the distance.” is available from Forecasting: Principles and Practice. External material is most useful when it prompts a specific validation check rather than replacing local judgement.
Forecasting, read by distance
Most forecasts are wrong about the number. The expensive ones are wrong about the distance.
Fifty notes on workforce forecasting and demand planning, organised by horizon rather than by method. A forecast for next week and a forecast for next year are not the same object: they feed different decisions, carry different error, and almost nothing true of one is true of the other.
Every figure is an illustration drawn from one organisation’s records, not a benchmark. Forecast methods, employment rules and financial requirements differ and should be reviewed for the organisation concerned.
One team's demand forecast at four horizons, as the planner published it
Next week
118 contacts/day
±4
Rosters are already fixed; nothing left to decide.
Next month
121
±11
Overtime and agency cover still movable.
Next quarter
126
±23
Recruitment decisions close in three weeks.
Next year
134
±48
Headcount budget and the lease on the second floor.
The band widens faster than the number moves, and only the two outer horizons have any decision attached. The near figure is precise and useless; the far one is vague and expensive. This is one employer's own published set, not a benchmark.
What actually moved one contact centre's volume, and what each claim rested on
Four years of daily counts; the dates do not move.
Send calendar joined to daily volume.
From the product owner; no comparable launch exists.
Never tested against anything.
Modelled from the channel mix, not observed directly.
Two of the five drivers were measured, one was modelled, one was somebody's expectation and one had never been tested at all. The forecast carried all five with equal weight. This is one employer's own register, not a statement of what any method requires.
Four quarters of one forecast against what happened
Under; the billing peak moved by two days.
Over; the launch lift never arrived.
Under; same billing effect, uncorrected.
Under; smaller, but the same direction again.
Three of the four quarters were under, and the one over was a single event. The mean absolute error was 7%, which looked acceptable and hid a bias that repeated every quarter. One employer's own record, not a benchmark.
01 / 07
What the forecast is for
Every forecast exists to feed a decision with a lead time. Named, it becomes an instrument; unnamed, it becomes a number people quote.
02 / 07
How distance changes the object
A week out and a year out are different objects with different methods, different error and different decisions attached.
03 / 07
The history you forecast from
Every forecast is a claim about a series. What is wrong with the series is usually what is wrong with the forecast.
04 / 07
What actually moves the number
Drivers, seasons and events, separated from the things everybody blames and nobody has ever tested.
05 / 07
How the number gets made
Each method assumes something about the world. The assumption, not the arithmetic, is what fails.
06 / 07
How wrong, and in which direction
Size is the easy half. Direction is the half that repeats, and the half nobody measures.
Planning software
Tools for capacity, scenarios and workforce forecasts
Three different comparison formats based on structured product review and repeatable forecasting checks.
07 / 07
Judgement on top of the number
Somebody always adjusts the output. Whether that improves it depends entirely on whether the adjustment is recorded.
The short version
Match the horizon to the decision, then publish the band
A forecast is not an attempt to know the future. It is an input to a decision that has a lead time, and the only horizon worth forecasting is the one that matches that lead time. Published as a point without a band, it acquires a precision it never had and is used to size things nobody intended. Published with its band and its last four errors, it is a usable instrument, and it remains usable when it turns out wrong.