Build a variance database
When it may fit: Results are scattered.
Evidence needed: Completed job costs and original estimates.
Trade-off: Takes data cleanup.
Construction Intelligence Brief
How to Investigate Historical Estimating Accuracy
Estimating is a forecast, and forecasts can be scored. Reviewing how past estimates compared with results shows the bias to correct before the next bid.
1. The business problem
When we estimate a job, how close do we usually come?
Many contractors learn from individual surprises but never score their estimates across the whole portfolio. Without a record, bias persists: bids that are consistently too low in some trades, too optimistic on productivity or too slow to reflect price increases.
Scoring a few years of estimates against actual results is usually the highest-value analysis a contractor can do on its own data.
2. What the evidence can reveal
An estimating accuracy review includes:
- Calculated Variance between estimate and actual, by project, cost type and trade.
- Calculated Average variance and its spread: a small average can hide large misses in both directions.
- Calculated Whether misses repeat by project type, size, owner or estimator.
- Information gap Reasons for each major miss, which show whether it was avoidable.
The aim is to separate bias, which can be corrected in the next estimate, from noise, which needs a contingency.
3. Investigate the problem
In the synthetic contractor, Calculated the estimated final margins for the six in-progress projects moved between the two snapshots by 1.5 pts down at Riverside, 7.0 pts down at Harbor, 6.5 pts down at Maple Street, 2.0 pts down at City Hall, 1.0 pts down at Eastgate, and 2.0 pts up at Valley School. The average change is about 2.7 points down.
Calculated Five of six projects moved the same way. That pattern is more like bias than random noise, and it matches a portfolio whose gross margin fell from 8.5% to 5.6%. The spread is also wide, from a gain of 2.0 points to a loss of 7.0, so a contingency would be needed even after the bias is corrected.
Hypothesis Since cost per labor hour rose 7.7% over the period, part of the bias may be labor inflation after the bids were priced. Information gap The demo does not show each project’s variance by cost type. A review that does so for completed jobs would show whether labor, materials, subcontractors or scope drove the misses.
4. Explore the interactive experience
Investigate it yourself.
Project Performance Dashboard: a project portfolio dashboard with estimated final margins, a scenario simulator for change orders and costs, evidence-linked answers from APIS, and a sample brief, all on synthetic data.
The experience below is built on a synthetic contractor, “Demo Builders Inc.” It has no bid history or backlog data, so its in-progress projects are used to show the method. Nothing in it is real.
- In the project table, compare Est. final margin across the two periods for all six projects.
- Which project moved most, and which moved the other way?
- Ask APIS which projects need management attention.
5. Evaluate possible solutions
Accuracy improves with a feedback loop between results and estimates.
6. What leadership should do next
Questions to answer
- What is the average and spread of variance?
- Is there a bias?
- Which cost types drive misses?
- What contingency do we need?
Data to gather
- Estimate and actual by cost code
- Project type and owner
- Estimator
- Reasons for major misses
Steps to take
- Score the last 20 projects.
- Correct the bias.
- Set contingency.
- Re-score yearly.
7. Engage Two Pillars
Improve Project Visibility and Financial Control
Two Pillars can help contractors investigate project portfolio, margin, estimating, cash-flow and operational performance using your own project data. A good first step is an estimating accuracy scorecard built from your completed projects.
Sources and data notes
- The organization, its figures, and every number in this article and its interactive experience are synthetic and illustrative. They are not real data, a benchmark, or a forecast.
- Reviewed October 2026. This article describes an approach to investigation. It does not replace review by qualified professionals.