How Much Contingency Should a Construction Project Carry?

How accurate is a construction estimate at each design stage?

AACE Recommended Practice 56R-08 covers building and general construction. It classifies estimates by how mature the project definition is and gives a typical accuracy range for each class at an 80% confidence interval.

Source: AACE International RP 56R-08. The ranges describe variation after appropriate contingency is applied, typically set so overrun and underrun are equally likely.

Two caveats from AACE matter in practice. The ranges are typical, not targets, and the real range for a given job comes from a risk analysis. For weak project systems and complex or risky projects, AACE notes the high side can run two to three times the table value. The RP also excludes one-of-a-kind work like health facilities, labs, and stadiums, which usually need wider ranges.

Why is the high side of every range wider than the low side?

Estimates miss scope more often than they double-count it. Something not yet drawn can't be priced, and it usually shows up later as added cost. The same skew shows up in outcomes: in Flyvbjerg's database of 16,000+ large projects, only 47.9% finished on budget. Our roundup of construction cost overrun statistics covers that data in more detail.

What are the three kinds of contingency, and who holds each one?

• Design contingency covers scope the documents don't show yet. The estimator holds it, and it should be close to zero by the IFC set.
• Construction contingency covers execution risk after buyout: coordination misses found in the field, gaps between trade scopes, and conditions nobody could see. On GMP work, the contractor usually carries it with owner approval to spend.
• Owner's reserve covers program changes the owner may still make. It sits outside the contract.

Keep escalation out of all three. AACE practice estimates escalation separately because it prices time, not uncertainty in the scope.

What burns through construction contingency fastest?

Construction contingency goes fastest when documents that were priced as a Class 2 set behave like a Class 3 set in the field. Helonic's preconstruction risk findings show what that looks like on issued drawings:

• Coordination conflicts were about 34% of flagged issues, and about 8 in 10 packages had at least one.
• Broken cross-references were about 21%. A missing detail becomes an RFI, then often a field change.
• Spec-to-drawing mismatches were about 16%. About 6 in 10 packages with specs had at least one.
• Three or more addenda came with about 35% higher issue density than zero or one addendum.

Every one of those found after buyout gets paid out of contingency. Found before bid, it's a revision. The change order cost benchmark shows what the field version costs.

What should an estimator check before setting contingency?

1. Which AACE class the set actually supports, judged by the deliverables, not the percent-complete label on the title block.
2. Whether every sheet in the index is in the set. A set completeness check answers that in minutes.
3. How many open cross-discipline conflicts are left, especially above ceilings and at structural penetrations.
4. Whether the specs and drawings agree on hardware, finishes, and equipment. Spec analysis flags the contradictions.
5. How many addenda are expected before bid day, and how late the last one will land.

If three or more of those come back weak, price the set as the class below the one on the cover. That's a more honest number than a flat percentage.

How does Helonic help an estimator defend a tighter contingency?

Helonic reviews the PDF set before pricing and returns a located list of coordination conflicts, missing references, spec mismatches, and code gaps. The estimator can clarify them in the bid, push them back to the design team as questions, or carry a specific allowance instead of a blanket percentage. Our preconstruction ROI analysis shows why fixing them on paper costs a fraction of fixing them in the field.

Construction Contingency FAQ