Most estimate failures are predictable. Here are the seven variables that account for 80% of variance, and how we control each one.
An estimate is a prediction. Like all predictions, it can be wrong. But unlike weather forecasts, cost estimates tend to fail for the same reasons over and over again — and most of those reasons are visible before work starts, if you know what to look for.
The seven variables
After 17 years and over 1,200 projects, we have identified seven variables that account for roughly 80% of estimate variance. They are, in order of impact: (1) drawing completeness, (2) market timing, (3) contingency discipline, (4) scope creep, (5) subcontractor quality, (6) site conditions, and (7) owner indecision. The first three are in the estimator's control. The last four are not — but they can be anticipated and priced.
Drawing completeness is everything
A schematic-design estimate is not wrong — it is appropriately uncertain. The problem arises when a schematic estimate is presented as a construction budget and then defended as costs rise through design development. An honest estimator tells you the accuracy range alongside the number. If yours doesn't, ask.
Market timing matters more than people admit
The same building estimated in Q1 2022 and Q1 2024 might differ by 18% — not because the design changed, but because the material and labour market did. We apply escalation factors to every estimate over 6 months, and flag them explicitly. Most estimators don't.
Contingency discipline
Contingency is not padding — it is a specific allowance for a specific type of uncertainty. Design contingency (for incomplete drawings) is different from construction contingency (for field conditions) and owner contingency (for scope changes). They should be tracked separately and reported separately. Lumping them together makes it impossible to understand what's driving cost growth.
What to ask your estimator
Before you accept an estimate, ask three questions: What is the accuracy range? What is the contingency, and what is it for? What assumptions would materially change this number if they turn out to be wrong? If your estimator cannot answer all three clearly, get a second opinion.