Passage
Forecasts that alter the future
Organisations often ask a forecast to perform two incompatible tasks: describe an uncertain future and authorise one present course of action. From that pressure comes the assumption that a forecast can be judged only by comparing its prediction with the eventual outcome. A precise number can coordinate budgets, but the number may also conceal the scenarios, exclusions and judgements that made it possible.
The question is how forecasting can remain useful without converting a chosen plan into a claim of superior knowledge. A hospital predicted unusually high demand and postponed routine appointments before the expected surge. When admissions remained moderate, the forecast looked exaggerated even though the cancellations had helped create the calmer result.
A forecast also creates behaviour. Once a sales estimate determines hiring or production, the organisation begins changing the conditions against which the estimate will later be judged. Success may make the original prediction appear accurate even when management action caused the outcome; failure may reflect a useful precaution rather than a poor model.
Review should therefore compare not only forecast and result but forecast, decision and consequence. Teams need permission to revise assumptions without rewriting the historical record of what they once believed. Preserving earlier versions is not bureaucratic clutter.
It reveals whether the institution learns from error or merely replaces an inconvenient number with a new one. A retailer announced low stock risks and suppliers accelerated deliveries in response. The warning became inaccurate precisely because people trusted it, illustrating how prediction can participate in the event it describes.
Forecasting acquires its authority partly from presentation. A neat percentage, a smooth curve or a single scenario can make an organisation appear to possess knowledge that the underlying evidence cannot sustain. The danger is not that forecasts are useless.
Decisions about staffing, stock or infrastructure cannot wait until the future becomes observable. The danger lies in allowing one convenient number to hide the assumptions that produced it. In the first case, the scenario range did more than widen the answer: it exposed which variables the team could influence and which remained external.
In the second, revising the forecast against actual outcomes converted error into information about the model instead of a reason to conceal earlier judgement. This makes organisational forecasting a continuing practice rather than a ceremonial annual prediction. A useful forecast should state the decision it is intended to support, the conditions under which its range is plausible and the signals that would trigger revision.
Managers may still need a single operating plan, but that plan should not be confused with the full account of uncertainty. When the assumptions remain visible, disagreement can focus on evidence and consequences. When they disappear, criticism is easily treated as disloyalty to the institution’s chosen future.
Some forecasts concern systems too large or slow to be changed significantly by the people who receive them. Evaluators should record the actions triggered by a prediction and ask whether those actions were desirable, rather than scoring accuracy as though the forecast had been a passive observation. Forecasts should finally be evaluated as part of organisational memory.
If only the latest model survives, staff cannot see whether confidence has narrowed because evidence improved or because inconvenient scenarios were removed. A short revision log can record changed assumptions, decisions taken and outcomes observed. This discourages hindsight from making every result appear predictable.
It also protects dissenting analysts: a rejected scenario may later prove useful without being celebrated as a prophecy. The institution learns when it can compare how different forecasts shaped action, not when it rewards whichever number happened to resemble the outcome. The same principle applies to automated forecasting systems.
Their outputs should not be separated from the team responsible for monitoring drift, since a model without an owner can remain operational long after its assumptions have failed. Forecast meetings benefit from separating likelihood from consequence. A low-probability event may deserve planning because its cost is high, while a likely event may require little intervention; one headline number cannot express both judgements.
A useful forecast may fail as a photograph of the future because it succeeds as an instrument for changing it. This approach makes forecasting demanding in the right way: not by multiplying numbers, but by connecting assumptions, decisions, observed outcomes and responsibility for timely revision.