Calibrate project estimates with completed jobs

Save the forecast before the work starts

A completed project is useful evidence only if you can compare it with the estimate you actually approved. Before starting, save the project export, the quoted scope, the cost basis, and the reported P50 and P80 cost and duration. Record the date and the assumptions that would materially change the result.

When the project closes, add the realized cost and elapsed duration using the same definitions. If the forecast includes allocated overhead but the actual cost excludes it, the comparison will flatter the estimate. If you forecast working days and record calendar days, you will diagnose a schedule problem that may simply be a measurement mismatch.

Count coverage, not just average error

Consider eight hypothetical projects with similar cost definitions. The numbers below are illustrative, in thousands of dollars. “Covered” means actual cost was at or below the saved P80 estimate.

Five of eight outcomes are covered: 62.5%. The target coverage associated with P80 is 80% over repeated comparable forecasts. Eight observations are too few to treat 62.5% as a precise long-run rate or proof that the model is broken. It is a reason to investigate, particularly if the same pattern persists as more jobs close.

Track P50 coverage as well. If actuals exceed both P50 and P80 unusually often, the estimates may be shifted too low. If P50 coverage looks reasonable but P80 misses recur, the upper tail may need attention. These are diagnostic possibilities, not conclusions you can establish from this table alone.

Review the cause of each miss

• Missing work: was a known activity absent from the approved estimate?
• Narrow ranges: did the pessimistic duration omit a plausible outcome?
• Event risk: did an uncertain event occur, and was its impact represented?
• Shared causes: did one supplier delay or review cycle affect several tasks together?
• Scope change: did the delivered work differ from the forecast scope?
• Measurement: were time, rates, and overhead recorded consistently?

A risk event occurring is not automatically evidence that its probability was wrong. An event assigned 10% probability can occur on the next project. Look across comparable opportunities for that event, and distinguish frequency errors from impact errors.

Keep original and revised scope separate

Suppose job B's $54,000 actual includes an approved $8,000 scope addition. The original forecast still belongs in the record. Add a revised forecast at the time the extra work is approved, and record actual costs by scope where possible. Do not subtract $8,000 mechanically unless that is the measured cost of the addition; a change-order price is not its cost.

Review both questions: was the original forecast useful for the original scope, and was the updated forecast useful after the change? Replacing the original estimate after seeing the outcome hides the information you need to improve.

Make one explainable adjustment

If repeated misses come from client review cycles, revise the relevant task ranges or add a clearly defined review risk. If the error is missing subcontractor expense, fix the cost basis. Avoid applying an arbitrary blanket multiplier that obscures the cause or counts a risk twice.

Keep the changed assumption in your next project notes, save the new forecast, and continue the log. Group similar work where practical, while acknowledging that a narrower group leaves fewer observations. Calibration is an ongoing check on assumptions; it cannot certify a future quote.

Start with the project workflow and backup guide, revisit how to represent risk without duplicate reserves, and use P50 versus P80 to interpret the percentiles you record.