The Situation
This GTA general contractor, running $180M in annual revenue, was losing on both sides of the bid table. Projects it won were coming in with 23% average cost overruns, eroding margin on every job. Projects it should have won were going to competitors whose estimates were tighter and more defensible. The estimating team was caught between bidding high and losing work, or bidding low and losing money — and increasingly it was doing both.
The root cause sat in how estimates were built. The process depended on a handful of senior estimators working from spreadsheets and hard-won institutional knowledge — knowledge that lived in people's heads, not in any system. Modern projects had outgrown that approach: volatile material costs, tightening labour markets, and variable subcontractor performance introduced complexity that no spreadsheet model could hold, and every estimator priced that uncertainty differently.
The firm also faced a succession risk it had not fully priced. Two of its most experienced estimators were approaching retirement, and with them would go the pattern recognition the entire bid process relied on. Leadership needed that judgment captured in a system before it walked out the door.
Our Approach
We started with a Phase 1 AI Readiness Assessment focused on the firm's bid history — years of estimates, actual costs, change orders, and outcomes scattered across spreadsheets and project files. That archive turned out to be the firm's most undervalued asset: it contained exactly the signal needed to learn where estimates went wrong and why. Consolidating and structuring it became the foundation of the engagement.
Through Phase 2 Strategic AI Integration, we built an AI-powered estimation engine trained on that historical bid data, layered with live inputs the manual process could never systematically account for: material cost trends, labour market conditions, and subcontractor performance history. For each new bid, the engine generates cost projections with confidence ranges and flags the line items where the firm's past estimates had deviated most from actuals — turning every previous overrun into a calibration point.
The senior estimators were central to the build, not displaced by it. Their judgment validated model outputs and encoded rules of thumb the data alone could not surface, which meant the system captured the retiring estimators' expertise while giving the whole team a common, defensible basis for pricing. Estimators retained final authority on every bid; the engine made their judgment faster, more consistent, and grounded in evidence.
The Results
Bid accuracy climbed from 77% to 94%, collapsing the cost overruns that had been draining margin from won work. With tighter, more defensible numbers, the firm could bid confidently where it previously padded for uncertainty — and its win rate rose 31%. The firm recovered $4.2M in margin in the first year, and the system proved itself immediately: as the client put it, the tool paid for itself on the first major bid, winning a $12M project the firm would previously have lost.
"The AI estimation tool paid for itself on the first major bid. We won a $12M project we would have lost."
Lessons for Your Organization
Construction firms tend to see estimation as an art that resists systematization — this engagement showed the opposite. The firm's own bid history held the answer all along; it simply had never been structured into a form that could inform the next estimate. Contractors facing the same margin squeeze should treat their historical project data as a strategic asset, and treat the coming retirement of senior estimators as a deadline for capturing what those estimators know.