Somebody Always Adjusts It
In almost every organisation the published forecast is not the system output. Who adjusts it, why, and why the adjustment is usually invisible.
In almost every organisation the number that gets published is not the number the method produced. Somebody has adjusted it — a planner, a manager, a committee — and in most cases there is no record of by how much or why.
The adjustment process in “Somebody Always Adjusts It” should preserve the unaltered model result, the human change and the reason for it. A team considering open the detailed product overview for monitask pricing can add operational time and project context, while the forecast owner remains responsible for documenting the override and reviewing its effect.
The adjustment is not the problem. The absence of the record is, because it makes the forecast unassessable and the judgement unimprovable.
Additional context for “Somebody Always Adjusts It” is available from Vena planning resources. External material is most useful when it prompts a specific validation check rather than replacing local judgement.
Who adjusts
The planner, applying knowledge the model does not have. The operations manager, who knows about something next month. The finance business partner, reconciling to a budget. And occasionally the person who will be measured against the number.
- The planner, applying knowledge the model does not have.
- The operations manager, who knows about something next month.
- The finance partner, reconciling to a budget.
- The person who will be measured against the number.
Four quite different motivations producing the same edit.
Why they adjust
Because models do not know about the release on the fourteenth, the contract that was signed last week, or the fact that the comparable period last year was distorted. Much adjustment is information arriving through a channel the data does not have.
That kind of adjustment usually improves the forecast, which is why abolishing override is the wrong response.
Why it is invisible
Because the adjustment happens in the same cell. The model's figure is overwritten and the file contains one number.
Nobody decided to conceal it; the spreadsheet simply has one cell, and the discipline of a second column has never been introduced.
The two columns
System and published, side by side, with a reason column. Three columns where there was one.
That is the entire apparatus required to make judgement assessable, and introducing it is a ten-minute change with years of payoff.
What the columns reveal
Typically: that a few large adjustments were excellent, that many small ones were noise, and that there is a consistent direction to the adjustments which nobody had noticed.
The direction finding is often the most valuable, because it is usually unintentional.
The adjustment that is information
A release date, a cancelled campaign, a contract start, a known outage. These are facts the model could not have.
In the long run they should be moved into the event calendar so the model does have them, which turns a recurring override into a modelled driver.
The adjustment that is a feeling
"It feels low." Sometimes right, often not, and it cannot be examined because there is nothing to examine.
The useful response is not to forbid it but to record it as such, so that after a year there is evidence either way.
The adjustment that is a negotiation
A separate thing, dealt with elsewhere here: a change made because a number is unwelcome rather than because it is wrong.
The record does not distinguish these by itself, but the reason column usually does, because the reasons given are noticeably different.
The size distribution
Worth plotting. In most operations the adjustments are many and small, with a few large ones. The small ones are usually the ones that hurt, because they are made without specific information.
A threshold — adjustments below five per cent require a stated reason or are not made — often improves accuracy by itself.
The adjustment nobody made
Also worth noticing: periods where the model was obviously going to be wrong and nobody intervened. These are failures of the same system.
A review that only examines the adjustments made sees half the picture.
Where the record lives
With the forecast, in the same file, published. Not in a separate governance document.
Published, it is read; separate, it is maintained for a while and then not.
The adjustment made downstream
Worth looking for: the forecast is published, and the rostering team applies their own uplift before staffing to it, because they have learned it runs low.
Two adjustments in series, neither aware of the other, and the net effect unknown to everybody. It is a common arrangement and it is invisible without asking.
Why people do not say
Usually because adjusting feels like an admission that the forecast is not trusted, and nobody wants to say that in a meeting.
Making the system column a normal published field removes the implication, which is most of why people start recording honestly.
The first time the columns are shown
Expect discomfort, and expect the first quarter's record to be incomplete. Both pass.
What helps is introducing it as a measurement of the model's blind spots rather than of anybody's judgement, which is also the truthful framing.
What changes once it exists
The conversation about judgement becomes evidence-based. Within a year it is possible to say whether adjustment has helped, which kinds have helped, and whose have.
None of those questions can be asked today in most organisations, and all of them are worth asking.