Budget variance and forecast accuracy model

Variance checked at the stage that produced it rather than at closeout, and every cost-to-complete carrying the error history behind it. Know which forecasts to trust before you price against them.

$1.5M

Reference-scale value

Data sets in the model

7 weeks

Build and deploy

Derived

Evidence grade

The decision this model makes

Variance reports compare actual to budget after the money is spent. That is an explanation, not a decision. By the time a code shows red, the concrete is poured and the invoice is approved.

Forty-one per cent of the variance pool surfaces only at closeout. Most of that was always going to happen — the share a stage-level check could have caught before the commitment is the claim, and it is deliberately small.

The second cost is not in the variance at all. Cost-to-complete on $52.9M of work in process runs 2.4% off, and every decision taken on that number is taken on a figure wrong by more than the margin it is protecting.

The model fires a variance check at the stage that produced it, and attaches every forecast's own error history to the forecast, so a number arrives with its confidence rather than without one.

Cause 01

Surfacing at closeout

Visible once the job is finished. Mostly real, and a quarter of it preventable at the stage.

Cause 02

Wrong cost-to-complete

Decisions on a forecast off by more than the margin. The cost is in the decision, not in the error.

Cause 03

Budgets not re-based

Still carrying last year's awarded rate. Priced by the rate-drift model and credited there.

Cause 04

Idle contingency

Held against codes with no overrun history. Small, immediate, and nobody asks for it.

Want more detail?

The rest of this model — the data sets it runs on, the cause split, the working behind the reference figure and a sample output — goes out by email. Two fields, nothing else.