The Number That Became a Target
What happens when a forecast is used as a target, why the forecast stops being informative within two cycles, and how to keep the two numbers apart.
A forecast used as a target stops being a forecast, usually within two cycles. Once the number somebody is measured against is the number somebody produces, the production changes — not dishonestly, just sensibly — and what comes out is a negotiating position with a decimal point.
The decision in “The Number That Became a Target” should be written before additional workforce data is collected. For organisations assessing remote employee productivity monitoring in relation to remote employee productivity monitoring, time and project reports can inform the decision, but assumptions, uncertainty and human review still need to remain explicit.
This is the single most destructive thing that happens to workforce forecasting, and it happens quietly, without anybody deciding to do it.
As a separate benchmark for “The Number That Became a Target”, see Rob Hyndman forecasting research. Keep source data, model, override and resulting decision visible as distinct steps.
How it starts
Innocently. The volume forecast is accurate, so it gets used to set expectations. Expectations become commitments. Commitments get reported against. Within a year the forecast is the plan, and the plan is what people are held to.
A forecast says what is expected to happen. A target says what somebody undertakes to achieve. They are allowed to differ, and the gap between them is the plan.
Nobody introduced a target. The forecast was simply the only number available.
What changes in the forecast
It gets safer. Forecasters who are judged on being exceeded forecast low; forecasters judged on efficiency forecast high. Both are rational and both destroy the information.
The giveaway is a forecast that is wrong in the same direction almost every period, with a size that does not vary much.
Why it is hard to see
Because the numbers still look reasonable and the error still looks acceptable. A consistent 6% bias is invisible in a mean absolute error and invisible to anybody not plotting the direction.
The series looks healthy right up until a decision is taken on it.
The two numbers
A forecast says what is expected to happen. A target says what somebody undertakes to achieve. They answer different questions and they are allowed to differ, often substantially.
Keeping them as two rows on the same page, with different names and different owners, is almost the whole remedy.
Where they legitimately differ
The target is above the forecast when a deliberate effort is being made. The target is below when capacity is being withdrawn on purpose. The gap is the plan, and it is the interesting number.
Collapsing them hides exactly the thing management most needs to see.
The forecaster who is asked to change the number
It happens in most organisations and it is not always improper: new information arrives in conversations before it arrives in data. The distinction is whether what is offered is information or a preference.
The practical rule: a change that comes with a reason goes into the forecast and into the record. A change that comes with a wish goes into the target row.
Recording the request
Not to catch anybody. Because six months later the only way to tell whether judgement improved the number is to know what the number was before and why it moved.
Two columns — "system" and "published" — and a one-line reason. That is the entire apparatus.
What happens to accuracy once this is in place
It usually gets worse on paper for a quarter, which is worth warning people about in advance. The safety margin comes out and the honest error appears.
The honest error is larger and it is the first error anybody has actually been able to act on.
Incentives that make it come back
Measuring the planner on forecast accuracy sounds reasonable and reintroduces the problem immediately: the fastest way to be accurate is to influence the outcome or to forecast something easy.
Better to measure the planner on whether the forecast was published on time, with its band and its error record intact, and to measure the decision on its own terms.
The version that is already embedded
Where a forecast has been the target for years, the series itself is contaminated: past outturns reflect behaviour that responded to the target, not demand.
That is worth saying out loud when the method is chosen, because it limits what any amount of modelling can recover. The series starts being informative again from the date the two numbers were separated, and not before.
The target that is a forecast in disguise
The reverse case, and it exists: a target set by taking last year and adding a percentage, with no undertaking behind it and nobody doing anything differently to achieve it.
That is a forecast wearing a target's name, and it is worse than either — it carries neither the information of one nor the commitment of the other.
What to say when the separation is resisted
That nobody is being asked to abandon the target, only to stop pretending it is a prediction. The target stays, with its owner and its commitment; what changes is that there is now also a number that says what is actually expected to happen.
Most resistance evaporates at that point, because the objection was never to the forecast — it was to losing the commitment.