Estimate Validation: How to Run a Cold-Eyes Review on Capital Projects
How estimate validation differs from estimate review, and how to run a genuine cold-eyes check that tests the number before sanction on capital projects.
A Class 3 estimate arrives at the funding gate with a 60-page basis document, 40,000 lines of detail, and eleven working days before the board paper locks. The review runs for a single day. The team that built the estimate walks the room through their own numbers, the questions stay at summary level, everyone agrees it looks reasonable, and the estimate goes forward unchanged. That is not an estimate review — it is a presentation with an audience. The distinction matters more than most organisations admit: IPA reports that only 44% of projects complete an owner estimate validation using detailed metrics before authorisation, and that project processes which fold review and validation into a single exercise deliver worse cost outcomes than those that keep them apart.
This article covers what separates an estimate review from an estimate validation, how to set up a genuine cold-eyes review with the independence and inputs it needs, what the qualitative review pass should test, how to validate the total using triangulation and ratio-to-driver metrics, and what the exercise must produce to change a number rather than merely bless it. The payoff is a review that surfaces the 5–15% error before sanction, when it is still a correction rather than a variance.
Estimate Review and Estimate Validation Are Not the Same Process
AACE International separates the two deliberately. RP 31R-03, Reviewing, Validating, and Documenting the Estimate, defines a three-step process — review, validate, document — and the first two steps answer different questions with different evidence.
Estimate review is qualitative. It tests whether the estimate technically conforms to scope and to estimating requirements: was it built using the required practices, tools and data; does it cover the entire scope; is it free of macro-level errors and omissions; is it structured and presented in the expected format; is it compatible with the schedule. Review is a conformance check.
Estimate validation is quantitative. RP 110R-20, Cost Estimate Validation, defines it as a quality assurance process that tests whether the estimate meets the project objectives and the estimate cost strategy — its appropriateness for purpose, including competitiveness. Validation compares the estimate to external metrics: gross unit costs, cost-capacity ratios, labour hours per unit of installed quantity. Critically, it uses metrics different from those used to build the estimate. An estimate built bottom-up from crew rates cannot be validated by re-adding the crew rates.
The practical failure is that most organisations run the first and call it the second. Reviewing the basis of estimate, confirming the WBS is complete and checking that indirects were applied is necessary work — but it will not tell you whether £355m is the right number for the scope. Only comparison against outturn data from completed projects will do that, and IPA’s position is explicit: simply conducting an estimate review, or an assurance check, is not a best practice on its own.
A review tells you the estimate was built correctly; only validation tells you whether it was built to the right number.
Setting Up a Cold-Eyes Estimate Review: Independence, Timing and Inputs
The ability to review an estimate effectively begins before the estimate is started. Three conditions have to be engineered in.
Independence with authority. The validator must be a cost engineer who sits outside the project team — internal but unattached, or a third party. Anyone on the team has too much stake in the outcome. Independence is only real if the findings report to the gatekeeper or the estimating function manager, outside the project manager’s line of authority. A validator who reports into the project they are validating is a reviewer with a job to protect.
Requirements defined up front. Validation is assurance against a stated standard, and without a stated standard it collapses into opinion. That standard lives in the owner’s estimate requirements document and the estimate plan, and is restated in the basis of estimate. RP 110R-20 goes further and requires an explicit estimate cost strategy — a statement, agreed by the sponsor, of whether the base estimate is targeted (aggressive, competitive), set to historical norms (conservative, predictable), ad-hoc, or drawn from a reference database. Most estimates carry one of these biases implicitly and document none of them. If the strategy is unstated, the validator has nothing to validate against, and the estimator carries personal risk for an overrun they were never told to price out.
Time and access. A comprehensive review of a Class 3 estimate is a demanding, time-consuming exercise, and it is routinely constrained by too little of both. The document set has to be agreed in advance and delivered on a date: basis of estimate, summary and detail reports in native files, risk analysis and escalation studies, the reconciliation to the previous estimate, priced equipment list with design parameters, WBS and CBS, plot plans and P&IDs. Native files matter — a locked PDF of the summary is not reviewable.
The Review Pass: Comprehensiveness and Appropriateness
Work the review in the order the errors compound. Larry Dysert and his AACE co-authors frame it as three sequential tests — comprehensiveness, then appropriateness, then reasonableness — and the sequence is not arbitrary. If the scope is not fully covered, the estimating method cannot be judged. If the method is wrong, the cost basis cannot be judged.
Comprehensiveness tests the design and technical basis. Does the estimate cover the full scope, and is everything outside it explicitly excluded rather than silently absent? The reliable failure modes are the same on most projects: temporary works, tie-ins and hot work on live plant, commissioning and start-up spares, owner’s costs, first fill, land and consents, and the interface scope that sits between two contractors and is priced by neither. Read the exclusions list against the scope description, not against the estimate.
Appropriateness tests the estimating process. Does the method match the estimate class and the maturity of the deliverables? A Class 3 estimate carrying Class 5 factored allowances in the piping and electrical bulks is a Class 4 estimate with a Class 3 label — and its accuracy range and contingency are being understated accordingly. Check that the allowances are quantified and defensible rather than round-number placeholders, that escalation has a stated base date and index, and that quoted procurement has been analysed rather than simply entered.
Both passes are qualitative, and both produce findings that are cheap to fix. Neither of them touches the question the sponsor actually cares about.
The Validation Pass: Testing the Number Against Outside Data
Validation asks whether the number is reasonable, and it works by successive disaggregation. RP 110R-20 sets out three methods that should be used together.
Successive detail. Start at the total and step down through progressively finer metrics, comparing each level to your historical band. Total installed cost, then cost per unit of capacity, then cost by major account (£/tonne of installed steel, £/m of pipe, £/m² of building), then the drivers underneath each — labour hours per tonne, £ per labour hour, material rate per tonne. Each step down localises the variance. A total that benchmarks well can conceal two large errors of opposite sign, and successive detail is how you find them.
Triangulation. Test the same quantity by more than one route. If the mechanical erection allowance benchmarks acceptably on £/tonne but the underlying assumption is 18 labour hours per tonne against a historical band of 22–28, the £/tonne agreement is coincidental — the estimate is carrying an optimistic productivity assumption offset by a high labour rate. Two metrics agreeing on the same figure is evidence; one metric agreeing is a coincidence you have not yet investigated.
Ratio-to-driver. Test the soft costs as ratios against the hard costs they should scale with — construction indirects as a percentage of direct field labour, engineering hours against equipment count, project management cost against total installed cost, contractor overhead and profit against the contract value. Indirect ratios are where estimates most often break: they are calculated last, checked least, and on a heavy industrial project can represent 30–45% of the field cost.
The comparison data has to come from more than one place. IPA’s research is unambiguous that using company and industry data improves cost performance, while relying on a single source does not — an organisation with strong in-house data can validate itself straight into repeating its own historical errors. Where the internal database is the only source, normalise it properly for location, base date, scope inclusions and capacity before any comparison is drawn.
One thing validation must never do is accept the total as a point value. The total cost is a base estimate plus contingency, escalation and reserves, and it should be expressed as a distribution, not a number. AACE’s position on this is direct: do not accept a number with an indeterminate basis. Where a total has been announced by the business before any estimate exists — the “tyranny of the number” — that figure is a target carrying an optimism bias, and validating the estimate against it is circular.
What the Estimate Validation Must Produce
A validation that ends in a memo has done half its job. The evidence points the other way: IPA found that discussing validation results in an interactive meeting including the owner project team, the gatekeeper and the validator improves cost effectiveness over issuing a report alone — and that only half of respondents held such a meeting.
The output set should be three things. A findings register with each item quantified, ranked by cost impact, and assigned an owner and a due date — a finding without a number attached will not survive contact with a schedule-pressed team. A validation report built on a standing set of KPIs, the same metrics checked on every estimate, so that results are comparable across the portfolio and the reviewer’s judgement is not the variable. And a recommendation on estimate adequacy stated plainly: whether the estimate is fit for the sanction decision, and whether the contingency matches the accuracy the review actually observed rather than the accuracy the class label implies.
The economics of doing this properly are well established. IPA’s research associates good-quality estimates with average cost savings of around 6% on large projects and 4–5% on site and sustaining projects, and reports that following a defined validation work process reduces cost by roughly 9% compared with projects that validate without one. Against a £420m estimate, a defined process with checklists and a standing KPI set is a rounding error in cost and a material movement in outcome.
The final discipline is the closeout review — capturing actual cost, scope and practice data at completion and feeding it back into the database that future validations draw on. Only about 30% of companies do it, and the reason is mundane rather than strategic: the data is not collected in the weeks after handover, the team disperses, and it is gone permanently. An organisation that never closes the loop will be validating its next estimate against nothing but its own optimism.
Key Takeaways
- Estimate review is a qualitative conformance check and estimate validation is a quantitative test against outside data; combining them into one exercise is associated with worse cost outcomes than running them separately.
- A cold-eyes review only works with genuine independence — a validator outside the project team reporting to the gatekeeper — plus a documented estimate cost strategy to validate against and enough time to work the detail.
- Run the qualitative pass in sequence: comprehensiveness of scope first, then appropriateness of method to estimate class, because errors at each level invalidate judgement at the next.
- Validate the number by successive detail, triangulation and ratio-to-driver metrics drawn from both company and industry data, and never accept a total cost expressed as a point value rather than a distribution.
- The validation must end in a quantified findings register, a standing KPI report and an interactive review meeting with the gatekeeper — and the loop is only closed when actual outturn data is captured at project completion.