We were eleven million light points nobody at anything. Split four ways, the same eleven million names the department that can act on it.
A single variance figure is arithmetically true and operationally useless. It confirms that the month was worse than planned and says nothing about why, which means the meeting that follows is an argument rather than a decision.
The number always has at least four causes inside it, and they belong to different people. Volume, which is operations. Realised price, which is marketing and basis. Cost rate, which is procurement and contracts. Cost usage, which is operations again but a different question from volume.
Left aggregated, each department can construct a story in which the variance was somebody else's. All of those stories are consistent with the single number, which is precisely why the same argument recurs every month and never resolves.
The undecomposed number also destroys the feedback loop the forecast needs. If you cannot attribute a miss, you cannot tell whether the model needs refitting or the asset had a bad month, so nothing gets corrected and next month's forecast carries the same error.
Open last month's pack. Is the variance split by cause, or is it one line with commentary underneath? Commentary is not decomposition.
Check whether the splits reconcile. A decomposition that does not add back to the total exactly has a residual bucket, and the residual is where the uncomfortable part usually got put.
Ask whether last month changed anything. If the same variance appears three months running and no parameter was refitted, the reporting is describing the problem rather than feeding back into the forecast.
Decompose four ways, every period. Volume, price, cost rate, cost usage. Each one names an owner. The conversation changes from whose fault it was to which of these four we are fixing.
Reconcile to the total exactly. No residual bucket. If the pieces do not sum, the decomposition is wrong and should be fixed rather than plugged.
Feed it back into the model. A variance that persists across periods is a parameter that has drifted, not bad luck. That is the signal to refit, and it is the whole point of measuring in the first place.
An Asset Health Check finds this one on a producing asset in two weeks, fixed fee.
The Asset Health Check is the cheapest possible way to find out whether we are worth the larger engagement. Two weeks, fixed fee, and a written answer either way.