Construction Contingency: What a Budget Holds Back

What does a construction contingency hold back?#

A construction contingency is the sum a budget holds back for cost that the design does not yet show. It sits on top of the base estimate. It is not a profit margin and it is not a discount. For a villa or a small resort, the allocator reads it as the budget's own admission of how much is still unknown.

The size of that admission is set by how finished the design is, not by the builder's confidence. A budget priced from a concept sketch has to carry a wide reserve. A budget priced from a full set of tender drawings can carry a narrow one. Two public frameworks formalise this. One is the estimate classification of AACE International. The other is the New Rules of Measurement, NRM 1, published by RICS.

This article takes both frameworks and applies them to a hypothetical villa and a hypothetical small resort. The figures in those two examples are illustrative arithmetic, invented to show the mechanics, and none of them comes from a Victaura project or from a market price. The class ranges themselves are read from a public document, the 2025 summary of the ASTM E2516 standard published by Alberta Infrastructure (industry estimate, dated 16 July 2025).

The questions for a principal who finances or buys off plan are three. At which estimate class was the budget made, how much reserve sits above the base cost, and who controls its release. A budget that cannot answer all three is a price, not an estimate.

Contingency is also not the same as the allowance for inflation, and owner changes need a line of their own. NRM 1 treats risk allowances and inflation as separate lines of the cost plan, and inside the risk allowance it gives employer change risks their own line. A budget that folds all three into one percentage hides which assumption is carrying the weight.

What is an AACE estimate class?#

An estimate class is a label for how much of the design exists when the estimate was made. AACE International, the cost engineering association, publishes the system as Recommended Practice 17R-97, with industry versions for process plants (18R-97) and for building and general construction (56R-08). ASTM International publishes a standard derived from them, E2516-11, which Alberta Infrastructure says is mainly based on the AACE practices.

Class 5 is the least defined estimate and Class 1 the most. Alberta's summary, dated 16 July 2025, gives project definition of 0–2% for Class 5, 1–15% for Class 4, 10–40% for Class 3, 30–70% for Class 2 and 70–100% for Class 1 (industry estimate, read from the document's Table 1). The bands overlap, because definition is a judgement and not a count.

Each class is tied to a purpose. Class 5 serves feasibility and screening. Class 4 serves concept study. Class 3 serves budget control. Classes 2 and 1 serve control, tender and bid check. Alberta's summary ties Class 3 to funding requests, and a budget used to raise capital before that stage is likely a Class 4 (Victaura Research assumption). The principal should not accept it as a Class 1 because it is written down to the last unit.

Precision on the page is not accuracy. A budget quoted to the nearest thousand dollars can still sit at Class 4. The class is a statement about the inputs, and the number of digits says nothing about them. Alberta's document adds that ranges can overlap, so a Class 4 estimate built on reliable repeat data may be as accurate as a Class 3 estimate built on new technology.

The class is assigned by the estimator, and a reader can test it. Ask for the drawing list behind the budget. A budget that cites a site plan and a room schedule is a Class 4 at best. A budget that cites coordinated architectural, structural and services drawings with quantity take-offs (measured quantities of each item) can be argued upward. If the estimator cannot name the class, the principal should assume the lowest one.

How wide are the expected accuracy ranges for building estimates?#

For building and general construction, the outer bound of the expected range narrows from -30% and +50% at Class 5 to -5% and +10% at Class 1. In Alberta's Table 1, adapted from ASTM E2516-11, the low and high ends read as follows. Class 5, -20% to -30% and +30% to +50%. Class 4, -10% to -20% and +20% to +30%. Class 3, -5% to -15% and +10% to +20%. Class 2, -5% to -10% and +5% to +15%. Class 1, -3% to -5% and +3% to +10% (industry estimate).

The range is asymmetric, and the asymmetry is the point. In every class the outer high end is larger than the outer low end. A budget can be wrong upward by more than it can be wrong downward, and the reserve exists for that reason. The principal reads the high side of the table as the one that matters.

The text of the same document states single outer bounds for each class, and they are the outer ends of the table. It gives -30% to +50% for Class 5, -20% to +30% for Class 4, -15% to +20% for Class 3, -10% to +15% for Class 2 and -5% to +10% for Class 1. The desk uses the table for the ranges and the outer bound where a worst case is needed.

The document warns that a target range should never be pre-determined. It says the accuracy range should always be set through risk analysis of the specific project, and it lists complexity, the quality of reference cost data, the assumptions used, the estimating technique, market and pricing conditions and the experience of the estimators as drivers. The allocator takes the table as a prior, not as a verdict.

What does the process-industry practice AACE 18R-97 add?#

AACE 18R-97 is the process-industry version of the classification, and its ranges are wider. A secondary summary on epcland.com gives, for Class 5, -20% to -50% low and +30% to +100% high, and for Class 1, -3% to -10% low and +3% to +15% high. A 2005 copy of 18R-97, filed in a 2019 public board packet on mpwmd.net, gives the same endpoints (industry estimate; AACE sells the current edition, which the desk did not open).

A process plant is not a villa, so the 18R-97 numbers are a ceiling and not a benchmark. Process plants carry flow diagrams, equipment lists and commissioning risk that a resort does not. The reason to cite 18R-97 is that it shows how far the range can open when the design is thin. A Class 5 high end of +100% means the budget can double.

The ranges are stated at a confidence level and not as hard limits. The epcland summary describes them as an 80th percentile band, while the 2005 copy of 18R-97 says the ranges, after contingency set at a 50% level of confidence, typically give a 90% confidence that the actual cost falls between the low and high ends. Either way, some projects land beyond the high end of their class. The reserve sized to the stated range is therefore not a cap.

A consultant guide gives different figures for the same class, which is a reason for caution. A Vista Projects guide gives a Class 3 headline range of -20% to +30%, and -10% to -20% low and +10% to +30% high in its own text, against -5% to -15% and +10% to +20% in the Alberta table (directional, consultant guide). Ranges are conventions that vary by edition, by industry and by author, and the principal should ask which table the estimator used.

How does RICS NRM 1 treat the risk allowance?#

NRM 1 is the RICS rulebook for order of cost estimating and cost planning of capital building works, and it gives the risk allowance a line of its own. The third edition was published in October 2021, effective 1 December 2021, and reissued in October 2022 as RICS practice information. Risk allowances sit beside preliminaries, overheads and profit, fees and inflation as separate components of the estimate.

NRM 1 defines the allowance as the amount added to the base cost for items that cannot be precisely predicted, to reach the cost limit. That wording is quoted from NRM 1 on the Designing Buildings wiki. The definition matters because it says what the line is for. It is a provision for the unpredictable, and the base cost is expected to be complete without it.

The method moves from a percentage to a risk register as the design matures. For order of cost estimates, NRM 1 calculates each of the four allowances as a percentage of the base cost estimate, but it recommends that the percentages are not standard and rest on a formal risk analysis, and that a project-specific risk register is kept and reassessed at regular intervals. The sections of NRM 1 the desk read do not prescribe how a register is priced. One approach, used in the resort example below, gives each risk a likely cost and a probability, multiplies the two to an expected value and totals the results (Victaura Research assumption, not an NRM 1 rule).

NRM 1 splits the allowance into four parts, and each part has a different owner. Design development risks cover the unknowns while the design is still moving, including changes in estimating data and third-party matters such as planning requirements. Construction risks cover site access, existing buildings, ground conditions and existing services. Employer change risks cover scope changes the client makes. Employer other risks cover items such as acceleration, early handover and liquidated damages, the sums fixed in a contract for late completion.

The split shows what each party can reduce. The design team can shrink the design development allowance by finishing drawings. A contractor can take some construction risk under a fixed price. Only the principal can limit employer change. A single undifferentiated percentage hides who can bring which part down, and an allocator should ask for the four lines.

How does a contingency read on a hypothetical villa budget?#

Take a hypothetical villa with a base estimate of USD 2,000,000. This is an invented figure, declared illustrative. It is not a price, not a project of Victaura's and not a market observation. It serves only to show how an estimate class turns into a dollar reserve.

At Class 5 the high end of the range sits at USD 2.6 million to USD 3.0 million, and the low end at USD 1.4 million to USD 1.6 million. That is the +30% to +50% and the -20% to -30% of the Alberta table applied to the base. A budget at this class is a screening number. It can rank sites against each other, but it cannot support a financing decision on its own.

The same villa at Class 4 has a high end of USD 2.4 million to USD 2.6 million, at Class 3 USD 2.2 million to USD 2.4 million, at Class 2 USD 2.1 million to USD 2.3 million and at Class 1 USD 2.06 million to USD 2.20 million. The ground has not changed and the building has not changed. Only the drawings have. Between Class 5 and Class 1 the high end moves from as much as USD 3.0 million to as little as USD 2.06 million.

A flat reserve of 10% on this villa matches the lower edge of the Class 3 high end and covers only a third to a half of the Class 4 high end. Ten per cent of the base is USD 200,000, against a Class 4 high-end excess of USD 400,000 to USD 600,000. This is a statement about ranges and not a forecast. A reserve is normally sized to an expected overrun and not to the outer edge, but the allocator should know how far apart the two are.

How does a risk register size a reserve on a hypothetical small resort?#

Take a hypothetical resort of 24 keys (guest rooms) with a base hard cost of USD 9,600,000. The figure is invented and declared illustrative. The register below is invented too. It uses an expected-value method (Victaura Research assumption, not an NRM 1 rule), and it shows what the method yields and what it does not.

A register of five risks, each with a likely cost and a probability, gives an expected value of USD 550,000. Ground conditions, 30% on USD 600,000, is USD 180,000. Imported finishes and currency, 25% on USD 480,000, is USD 120,000. A permitting delay, 20% on USD 350,000, is USD 70,000. An owner specification upgrade, 40% on USD 300,000, is USD 120,000. A marine logistics delay, 15% on USD 400,000, is USD 60,000. The total is 5.7% of the base.

The expected value is a mean, and a reserve set at the mean is easy to break. Each risk either arrives or it does not. If the ground conditions risk lands at USD 600,000, that one item alone exceeds the whole USD 550,000 reserve. The register total is what many similar projects would spend on average, while a single project gets one draw from the distribution.

The register total should be read against the class range, and the gap between them is information. At Class 4, the +20% to +30% high end on USD 9.6 million is an excess of USD 1.92 million to USD 2.88 million. At Class 3, the +10% to +20% high end is an excess of USD 0.96 million to USD 1.92 million. A register of USD 550,000 covers a fraction of either. The gap is not an error, because a register lists risks someone has identified and a class range also carries the ones nobody has.

Who controls the contingency once construction starts?#

A reserve that anyone can spend is not a reserve. The principal should know who may release it, against what evidence and at what threshold. This is a governance question. The desk found no published statistic on how reserves are released on villa or small resort projects, so it states the question and not an answer.

Contingency inside a contractor's lump sum belongs to the contractor, and contingency held outside it belongs to the owner. A fixed-price contract moves some construction risk to the contractor in exchange for a price. Employer change risk stays with the owner under any contract. The NRM 1 split between construction risk and employer risk is a reminder that some risks cannot be priced away by a contract form.

Draw evidence matters more than the percentage. A reserve ledger shows each draw against a named risk, the balance at each interim payment valuation and which risks have retired. The design development allowance should fall as drawings are completed. A balance that stays flat while the design matures suggests the allowance was a number and not a calculation. For how payment tracks physical progress, see stage-of-completion pricing.

A reserve consumed by scope creep is not insurance. If it pays for upgrades the owner requested, the specification has changed and the budget has not been reset. Island and remote sites add weight to construction risk through logistics and imported materials, a pattern covered in island construction costs. The principal should ask for the unspent balance to be reported, not absorbed.

What should a principal ask about a project budget?#

First, at which class was the budget made, and which drawings stand behind it? The answer should be a class number and a drawing list. If the budget says only 'detailed', the principal asks for the percentage of design complete. An answer of Class 4 or Class 5 fits screening or concept study, and the reserve should be sized to match.

Second, how much reserve sits above the base cost, and how is it split? The principal asks for the four NRM 1 lines, design development, construction, employer change and employer other, each in currency and as a percentage of base. The principal also asks which method sized them, a benchmark percentage or a register, and for the register if there is one.

Third, who releases the reserve, on what evidence, and what happens to the unspent balance? The answer should name a person or a body, a threshold and a reporting rhythm, and point to the clause of the construction contract that sets them. A reserve that returns to the owner when unused behaves differently from one that stays with the contractor, and the principal should know which one the budget assumes.

The wider evidence on overruns sits in an earlier piece, and this one does not repeat it. The resort conversion underwriting piece covers capex, contingency and the 2026 HVS survey. HVS states that its survey is meant to support preliminary cost estimates and comparisons and should not be relied on to determine the cost of actual projects. The desk reads that as a screening-grade use, which matches the Class 5 purpose in the Alberta table.

A residual land value model is only as good as its cost input. If the construction budget is a Class 5 number, the land price derived from it inherits the same range. The mechanics are set out in residual land value, and the allocator should check which class feeds the land bid.

Where is this analysis thin? Points honestly disclosed#

The desk did not open the current edition of AACE 18R-97, and read NRM 1 only in its risk allowance sections. AACE sells its practices; the desk read a 2005 copy of 18R-97 filed in a public board packet, and sections 2.15, 3.18 and group element 13 of the October 2021 NRM 1. The class ranges for buildings come from a public government summary of ASTM E2516-11. The process-industry ranges come from a secondary summary and that 2005 copy. The NRM 1 risk categories and percentage method come from the RICS text, while the expected-value method in the resort example is a Victaura Research assumption. Each of these is graded accordingly.

Sources disagree on ranges, and the desk has not resolved the disagreement. Alberta's bullet text gives the outer ends of its own table, while a consultant guide gives different figures for the same class. The desk used the Alberta table and flagged the others. A reader who needs an exact range should buy the AACE or ASTM document and use the table the estimator used.

The standards were built on large, institutional projects. The desk did not find, as of 27 October 2026, a published accuracy range calibrated to villas or small resorts in island or emerging markets. The ranges in this piece may understate the spread on such sites. That is a judgement, not a measurement, and it should be read as directional.

Both worked examples are invented. The villa base of USD 2,000,000, the resort base of USD 9,600,000 and every risk probability and amount in the register are illustrative. Probabilities in a register are subjective by construction. The examples show mechanics and do not predict any project's outcome, and the 5.7% figure is not a benchmark for resorts.

A class range is a statistical convention and not a guarantee. The epcland summary calls the ranges an 80th percentile band, and the 2005 copy of 18R-97 speaks of a typical 90% confidence that the actual cost falls inside them, so some projects fall outside the high end. A budget at Class 1 can still overrun by more than 10%. The class lowers the odds. It does not remove the risk.

What does this mean for the allocator?#

The reserve is a function of design maturity, and the allocator can test it with four questions. Ask for the class, the drawing list, the four risk lines and the release rule. The AACE and ASTM tables give the range to expect at each class, and NRM 1 gives the vocabulary to split the reserve by owner. A budget that survives those questions is a better document than one with a low headline number.

A thin reserve at an early class is a flag and a thick one at a late class is also a flag. The first suggests the budget has not priced the unknown. The second suggests the numbers have been padded or the class has been overstated. The allocator reads the reserve and the class together, and neither alone.

This article is research and not a recommendation on any budget or project. It states how the public frameworks define classes and allowances, and it shows with invented numbers how they work. It makes no claim about the cost of any Victaura project and publishes no figure for one.

Skin in the game disclosure. Victaura, through its parent Greystone B.V. (Netherlands), develops projects on Lake Como (Italy), in Nungwi, Zanzibar (Tanzania), on Gili Air and in Uluwatu, Bali (Indonesia), and holds an off-plan capital position on Al Marjan Island, Ras Al Khaimah (UAE). Readers should assume commentary may be influenced by, or benefit, these positions. This document is classified as marketing material under MiFID II Article 24(3). It is not investment advice.