Variance
Variance is the gap between what was planned and what occurred: estimated effort against actual hours, or allocated hours against logged hours. Expressed as a difference or a percentage, it is a measure of how well a plan predicted reality rather than a measure of performance.
Variance is a diagnostic. One figure alone says very little. Any single piece of work can go its own way, for reasons nobody could have seen coming.
The signal is in the pattern. A gap in the same direction, on the same kind of work, month after month, is a bias in how the team estimates. That can be fixed. Finding it is the only reason to track the measure at all.
Direction matters as much as size. Steady overruns point at optimism, or at scope moving. Steady underruns usually mean padding, and padding costs just as much. It holds time the firm could have sold.
One thing kills the measure: punishing a miss. Once a miss costs somebody something, the estimates change rather than the estimating. Then the number stops meaning anything.
Group the work before reading it. Variance across everything a firm does averages to roughly nothing, and hides the one kind of job that always runs long. Sorted by type of work, or by who scoped it, the same figures start pointing somewhere.
Where this shows up in PeopleMuster
PeopleMuster holds both sides: planned hours on the capacity grid and logged hours in the recap reports, per project and per person, for the same period.
Questions people ask
What causes most estimate variance?
Scope moving and unstated assumptions, more often than slow work. A brief that grew is the commonest explanation, and it is invisible if nobody recorded the original shape.
Is negative variance good?
Not automatically. Finishing well under estimate repeatedly usually means the estimates carry padding, and that padding held capacity nothing else could use.
How should variance be reviewed?
In a batch, across similar work, without attaching consequences to individual misses. The moment a miss costs somebody something, the estimates adjust instead of the estimating.