The estimate-to-actual gap: why GC bids miss final cost and how to close the loop

The estimate-to-actual gap (cost variance) is the difference between the budgeted cost (the estimator's bid or internal takeoff) and final cost after construction. In practice, most projects come in over the original estimate. Industry analyses commonly report that roughly 85-90% of projects overrun, with average growth often in the high teens to ~20-30%. A large share of that gap traces to early estimating and scope assumptions, then to change orders and rework that land during construction.

General contractors narrow the gap with two habits: disciplined estimating before award, and a hard feedback loop that reconciles actuals back into the next estimate. That is the same institutional-memory problem described in why every construction estimate still starts too close to zero.

How to measure the gap

Compute overall and by CSI trade (or your house trade map) and by phase (conceptual, schematic, bid, buyout, closeout). One study of hundreds of U.S. street projects found final costs averaging only a few percent above the low bid, while larger, more complex building work often sees much higher variances. Delivery method also matters (hard bid vs CMAR vs design-build).

Normalize older projects to constant dollars (BLS PPIs or similar) before you compare years. Align definitions: engineer's estimate vs GC bid vs GMP vs final approved cost are not interchangeable.

Illustrative patterns

Mock project-level rows (shape only, build your own library from closed jobs):

Mock trade rollup, finishes and sitework often run hotter than structure:

If finishes routinely run 15-20% over, that is a scope and allowance problem to fix in scope generation and coverage, not a mystery for the next school job.

Distribution shape (directional). Most projects cluster in roughly +0-10% variance, with a long tail beyond +20%. That right skew is why averages look worse than "typical" jobs and why a few disasters warp company memory.

Common causes and when they hit

Industry research consistently points to:

Most overrun dollars crystallize during construction and closeout, not on bid morning. The original bid can be directionally right while pending COs and omissions still add up. Studies that find final cost highly correlated with the low bid are consistent with that: the base is real; the gap is mostly measurable adds.

Mitigation: tighten estimate → actual

• Thorough scope definition: complete trade scopes, allowances, alternates, exclusions; use coverage matrices (bid coverage)
• Consistent estimating standards: master assumptions, carry rules by delivery method and stage (carry methodology)
• Contingency with eyes open: higher early; thin coverage or incomplete design gets an explicit risk factor, not silence
• Estimate QA before bid: peer review; quantities and rates vs history; scope-completeness checks (final bid review)
• Mandatory closeout reconciliation: final costs, CO log by cause, and actual unit costs written back to the estimate library
• Change-order discipline: every extra cost documented and approved; unauthorized scope is how overruns hide
• Connect systems: when budget, buyout, and actuals live in disconnected spreadsheets, variance arrives as a surprise at closeout

The contractors who improve are the ones who compare the bid to early job-cost and buyout results while the project is still alive, not only in a postmortem nobody reads.

KPIs worth tracking

Useful dashboard views: share of projects within ±5% / ±10% / ±20% of estimate; average variance by trade; CO mix by cause; one sample project's estimate trail from schematic → bid → final.

Estimate → actual process

1. Estimator's bid / internal estimate: documented assumptions and carry
2. Award: base contract locked
3. Buyout: trade prices vs estimate; first hard signal of gaps
4. Construction: job cost and COs accumulate
5. Closeout: final actuals
6. Analysis: variance by trade and cause written into the next pursuit's standards

Skip step 6 and every new bid starts near zero again.

Building your own dataset

You do not need a national study to run this loop. Start with closed projects you control:

Supplement with public capital-project reports where agencies publish budget vs actual, and with bid tabs for award-price context. Adjust older dollars with PPIs. Publish or share only aggregates if confidentiality matters, the learning is in the pattern, not the named job.