Why a forecast should come with a range
20 November 2026
A single forecast number hides how sure the model is. A range lets the person planning stock, staff or cash decide how much risk to carry.
A forecast is usually presented as one number: next month's demand, next quarter's churn. The number looks precise, and that is the trouble. It says nothing about whether the model is nearly certain or barely better than a guess.
The range is the useful part
We present forecasts as a low, a likely and a high value. A buyer deciding how much to order does not need the single most likely figure; they need to know how bad a shortfall could be, and how much stock would cover most cases.
The range also widens the further ahead the forecast looks. Next week is narrow, three months out is wide. Showing that honestly stops a distant estimate being read with the confidence of a near one.
Check the range against the past, without peeking
A range is only worth showing if it is right about as often as it claims. We test that by replaying past periods one step at a time, giving the model only the data it would have had on that day, and counting how often the real figure landed inside the range.
A simpler method goes through the same test. If a more complex model cannot clearly beat last year's same week, the simpler one is used.
Keep the planner's judgement in the loop
People who plan for a living know things the data does not: a supplier is unreliable, a promotion is coming. When they override a forecast, the reason is recorded. Reviewed later, those reasons show what the model is missing and what should be added to it.