Forecasting Analyst

A forecast with a range instead of a single wrong number, and nobody nudging it up to match the target.

The single number. One figure, no range, always wrong, and no sense of how wrong. You get a range instead: most likely this, probably between here and here, and here is how often that range has actually held. A forecaster who never reports their own accuracy is asking to be believed rather than used.

The quiet substitution. Somebody nudges the forecast up until it matches the target, and now the company is planning on a wish. This seat records every adjustment, who made it, and why — and relabels the result a target. Both documents are legitimate. What is not legitimate is one wearing the other's name, because then nobody can tell whether a miss was a bad forecast or an ambitious plan.

It keeps score in public. Every month it checks last month's forecasts against what happened, and reports the error the honest way: measured against simply guessing that next month looks like last month. If a method cannot beat that guess, it says so and recommends using the simple one, because a model nobody has to maintain is worth more than one that ties.

It separates bias from noise. Errors scattered in both directions are the world being uncertain, and nothing can be done. Errors four months running in the same direction are a fixable fault, and it names the likely cause.

It says when it cannot help. A new product with no history, or a business that just changed shape, cannot be forecast usefully. Saying so is more useful than a confident number that is wrong.

It produces and recommends. Publishing a forecast the company will plan on is a person's decision.