Cheapest Quote vs Delay Risk: The Economics of Choosing a Contractor in Singapore

Every owner who has collected three quotations for a fit-out, an office reinstatement or an addition and alteration (A&A) job has faced the same question: the lowest price is in front of you, the difference is real money, and the risks of accepting it are hard to put a number on. The usual answer is intuition, or a vague sense that “you get what you pay for”. Neither helps when you have to justify the decision to a board, a landlord or a managing agent.

This guide replaces intuition with a method. It draws on the research literature on project overruns, which is large and surprisingly consistent: most projects run late, and the distribution of delays has a long right tail. It explains why the lowest bidder in a competitive tender is systematically more likely to have under-priced the work (the “winner’s curse”), and how under-priced work tends to come back to the owner as variations, claims or delay. It then works through an Illustrative comparison of two quotes using probability-weighted delay cost, which is the same logic behind our cheap quote vs delay risk simulator.

None of this is an argument for always paying more. A higher quote is not proof of quality, and a lower quote is not proof of risk. The point is to compare quotes on expected total cost to the owner, which includes the cost of time.

The question owners are really asking

When an owner compares a S$168,000 quote with a S$185,000 quote, the visible difference is S$17,000. The invisible question is: what is the total cost of each option once the project is finished, including the cost of any delay, any variation claims, and any disruption to the business?

That total cost has three parts:

The third line is where most comparisons fail. A delay of two or three weeks on a commercial fit-out can easily cost more than the gap between two quotes, but it does not appear on any quotation. Our guide on the true cost of a late handback and the overstay cost calculator cover the weekly cost side in detail. This guide focuses on the probability side: how likely is a delay, and how large could it be?

What the evidence says about overruns

Base rates: most projects overrun

One of the largest bodies of evidence on project outcomes comes from Bent Flyvbjerg and colleagues. Their 2002 study of 258 transport infrastructure projects in 20 countries found that actual costs were on average 28% higher than forecast, and the authors argued that the pattern was too consistent to be explained by honest error alone.¹ Other researchers have disputed that interpretation, and the cause of overruns remains debated; the existence of a large average overrun is the point that matters here.

The later database behind Flyvbjerg and Gardner’s book How Big Things Get Done (2023) covers more than 16,000 projects across more than 20 fields and 136 countries. According to a review of the book, 91.5% of those projects went over budget, over schedule, or both, and fewer than 1% were delivered on budget, on time and with the promised benefits.²

Closer to the scale of building works, a study of large construction projects in Saudi Arabia by Assaf and Al-Hejji reported that 70% of projects experienced time overrun, and that change orders were the most common cause of delay identified by owners, consultants and contractors alike.³ That finding matters for this guide: the delay problem and the variation problem are the same problem seen from two angles. See our guide to variation orders and provisional sums.

These are not Singapore fit-out statistics, and nobody publishes a clean base rate for small A&A jobs in Singapore. But they establish a prior that any owner should start from: on-time completion is the exception, not the default.

Reference class forecasting: the outside view

Why do plans keep underestimating? Kahneman and Lovallo described two ways of forecasting. The “inside view” builds a forecast from the details of the current plan: the programme, the sequence, the crew size. The “outside view” asks how long similar projects actually took.⁴ The inside view is systematically optimistic, because plans describe what should happen, not the many small things that do happen.

Flyvbjerg turned the outside view into a practical method called reference class forecasting: identify a class of comparable past projects, establish the distribution of their outcomes, and position the new project within that distribution.⁵ The 2006 paper describes the method’s first practical application, to cost forecasts for large transport projects.

For an owner comparing quotes, the reference class is not “fit-outs in general”. It is “jobs of this type, done by this contractor”. That is why a contractor’s delivery record is worth more than its programme. A programme is an inside-view forecast. A record of comparable completed jobs is an outside-view forecast.

Delays are fat-tailed, not bell-shaped

The second lesson from the research is about the shape of the distribution.

If delays followed a normal (bell-shaped) distribution, an owner could reason from the average: a job that is “usually two weeks late” would rarely be more than four or five weeks late. The data say otherwise.

• Love, Sing, Wang, Edwards and Odeyinka analysed schedule overruns on 276 Australian construction and engineering projects and found the data were non-Gaussian, with skewness and kurtosis inconsistent with a normal distribution. The best overall fit was a four-parameter Burr distribution, a family with a heavy right tail.⁶
• Flyvbjerg and co-authors found that cost overruns on IT projects follow a power-law distribution: many projects overrun a little, and a fat tail of projects overrun enormously. A later clarification restated the cost-overrun sample as 4,677 projects.⁷
• For the Olympic Games, Flyvbjerg, Budzier and Lunn estimated power-law tail exponents between roughly 1.2 and 1.7, a range in which the variance of outcomes is not finite and averages never settle down. They called the resulting pattern “regression to the tail”: eventually a new, larger extreme arrives.⁸

A small office fit-out is not an Olympic Games. Interdependence is far lower, so the tail is thinner. But the mechanism that produces fat tails is present on every construction job: one problem (a late approval, a wrong material delivery, a concealed defect behind a wall) delays the next trade, which delays the next. Delay compounds; it does not average out.

What this means for an owner

Two practical conclusions follow:

1. Start from the base rate, then adjust. Assume some delay is likely and ask what evidence would justify assuming otherwise for a specific contractor.
2. Price the tail, not just the average. The expected delay cost is dominated by the minority of bad outcomes. An owner who looks only at the “most likely” scenario will consistently under-estimate risk.

Why the lowest bidder is often the one who under-priced

The winner’s curse

In 1971, three researchers at the oil company ARCO, Capen, Clapp and Campbell, studied why oil companies earned poor returns on offshore lease auctions. Their explanation became known as the winner’s curse: when several bidders estimate the value of the same uncertain asset, the winner tends to be the bidder whose estimate erred the most in the optimistic direction.⁹

Construction tendering is the mirror image. Each contractor estimates the cost of the same uncertain scope. Estimates scatter around the true cost. The lowest bid is, statistically, more likely to come from the contractor who under-estimated the cost, not the one who is most efficient. The more bidders and the more uncertain the scope, the stronger the effect.

This is not a moral judgement. It is selection: picking the minimum of several noisy estimates selects for the most optimistic error.

How experienced contractors escape the curse (and who pays)

Dyer and Kagel tested this directly. Experienced construction executives fell prey to the winner’s curse in laboratory auctions, yet performed well in real tendering. The authors identified field mechanisms that let contractors escape the curse even after bidding too low, along with learned judgement about which jobs to bid and private cost advantages.¹⁰

Some of those escape mechanisms are benign: a contractor recognises a mistake and withdraws, or absorbs a small loss. Others shift the cost to the owner. Bajari, Houghton and Tadelis studied highway paving contracts in California and found that contracts are routinely adapted after award because the original specification was incomplete. They estimated that adaptation costs amounted to 7.5–14% of the winning bid, and found that bidders anticipate these changes and adjust their bids strategically. Those costs were much larger than the bidders’ mark-ups from market power.¹¹

The three recovery routes

When a quote is too low for the scope as actually built, the shortfall comes back in one of three ways:

The first route is visible and contractual; our guide to variation orders and provisional sums explains how to close it. The third route is the least visible at tender stage and frequently the most expensive. A contractor that has under-priced a job has an economic incentive to minimise resources on it, and fewer resources mean a longer duration.

Expected value vs tail risk

Expected value

The simplest correct comparison uses expected total cost:

Expected total cost = Quote + Expected scope growth + (Probability of delay × Expected delay length if late × Cost per week of delay)

Each term can be estimated, even roughly, and a rough estimate is better than an implicit zero.

Why the expected value is not enough

Expected value treats a 10% chance of losing S$100,000 the same as a certain loss of S$10,000. Most owners do not. A small business cannot absorb a three-month delay on its only outlet; a tenant whose lease expires on a fixed date faces a cliff, not a slope. Two additional measures help:

• Percentiles (P80, P90, P95): the cost that is not exceeded in 80%, 90% or 95% of scenarios. A P90 of S$254,000 means a one-in-ten chance of a total cost above S$254,000.
• Conditional value-at-risk (CVaR), also called expected shortfall: the average cost in the worst tail, for example the worst 5% of scenarios. Rockafellar and Uryasev’s formulation made CVaR a standard risk measure because it captures how bad the tail is, not just where it starts.¹²

An owner with a hard deadline should give heavy weight to P90 and CVaR. An owner with flexibility and deep pockets can lean more on the expected value.

Worked example: Quote A vs Quote B (Illustrative)

The following example uses Illustrative numbers only. They are chosen to be plausible for a small commercial fit-out in Singapore and to show the method, not to describe any real contractor. The same defaults are loaded in the delay-risk simulator, where you can replace them with your own figures.

The inputs

Why these numbers? The probabilities reflect the base rates above, adjusted for what the owner learned from references: Contractor B’s last several comparable jobs finished within a week or two of programme; Contractor A’s references were mixed. The wider spread for A reflects less evidence and more ways for things to go wrong. The weekly cost of S$9,000 is an Illustrative sum of rent on the new premises, holding-over on the old premises, and staff disruption.

Delays are modelled with a lognormal distribution: if the job is late, the length of the delay is the median multiplied by a random factor whose logarithm is normally distributed. The lognormal is right-skewed and has no negative values, which suits delays, and its tail is heavier than a normal distribution but lighter than a power law. It is a deliberately moderate choice for small projects, given the fat-tailed evidence above.

Expected values (calculated)

For a lognormal, the mean equals the median multiplied by exp(sigma²/2).

• Quote A: mean delay if late = 4 × exp(0.405) ≈ 6.0 weeks. Expected delay = 0.60 × 6.0 ≈ 3.6 weeks. Expected delay cost ≈ 3.6 × S$9,000 ≈ S$32,400. Expected total ≈ S$200,400.
• Quote B: mean delay if late = 1.5 × exp(0.18) ≈ 1.8 weeks. Expected delay = 0.25 × 1.8 ≈ 0.45 weeks. Expected delay cost ≈ S$4,000. Expected total ≈ S$189,000.

On expected value, the “expensive” quote is about S$11,400 cheaper.

The simulation (1,000,000 Illustrative runs)

The last rows are the most instructive. Quote A is cheaper in the majority of scenarios (about 56% of runs). Its median total cost is lower than B’s. An owner who reasons from the “most likely” outcome will pick A, and more often than not will feel vindicated.

But A’s losses, when they come, are large, and the expected value and every upper percentile favour B. The break-even is simple: the S$17,000 price gap buys about 1.9 weeks of delay at S$9,000 a week. Quote A has roughly a 48% chance of running more than 1.9 weeks late in this example; Quote B has very little chance of doing so.

Adding liquidated damages (Illustrative)

Suppose both contracts carry liquidated damages (LDs) of S$6,000 per week, capped at 10% of the contract sum. In the simulation, LDs recover on average about S$8,700 from A and S$2,500 from B. The expected net cost becomes about S$191,600 for A and S$186,600 for B, so LDs narrow the gap. But A’s P95 net cost is still about S$275,800 against S$192,500 for B, because in about 39% of A’s runs the delay is long enough to exhaust the LD cap. LDs also assume the contractor has no valid extension of time and that the deduction is actually collected. See liquidated damages and extensions of time.

The cost of delay to owners

The weekly cost of delay is the single most important number in the comparison, and it is usually larger than owners expect. Typical components:

Our overstay cost calculator helps build this number for a commercial handback. For a broader view of where fit-out time goes, see realistic fit-out programmes and where delays come from.

The Singapore context in 2026

A busy market

Singapore’s Building and Construction Authority (BCA) projected in January 2026 that total construction demand would remain steady at S$47–53 billion in nominal terms in 2026, after preliminary demand of S$50.5 billion in 2025. Output was forecast at S$43–46 billion, and demand was projected to average S$39–46 billion a year from 2027 to 2030, with large projects such as Changi Airport Terminal 5 in the pipeline.¹³ BCA’s release did not forecast tender prices.

For an owner of a small job, a sustained pipeline of large projects has two practical effects. Good subcontractors and supervisors are in demand, so a contractor that has bid low may struggle to hold on to them. And contractors have alternatives: a job that turns out to be under-priced competes for resources with better-paying work.

Manpower constraints

Construction in Singapore relies heavily on foreign workers, and their numbers are capped. The Ministry of Manpower lists the construction sector’s dependency ratio ceiling (the maximum share of foreign workers in a company’s total workforce) at 83.3%.¹⁴ A contractor’s ability to add a second crew when a job slips is therefore bounded by its local headcount and quota, not simply by willingness to pay overtime. When comparing quotes, ask each contractor how many workers and supervisors it will deploy and whether they are its own employees or subcontracted.

How the public sector weighs time

Singapore’s public sector does not award construction contracts on price alone. Under BCA’s Price-Quality Method, building tenders with an estimated construction cost of S$3 million and above are evaluated with price weighted at 40–60% and quality at the remainder, and the quality attributes include past performance on timeliness and quality, safety record and the experience of key personnel.¹⁵ Private owners are free to adopt the same logic on a small job: score delivery record explicitly, alongside price.

Contractual levers that make on-time real

A contractor’s record is the best predictor of delivery, but the contract determines what happens when things slip. These levers turn a promise into an obligation.

1. Liquidated damages at a genuine pre-estimate

LDs give the owner a pre-agreed sum for each day or week of culpable delay without having to prove the actual loss. Under Singapore law, an LD clause is enforceable if it is a genuine pre-estimate of loss and not a penalty. The rate should be grounded in the weekly delay cost calculated above, and any cap should be understood as the point beyond which the owner bears the risk. Details in liquidated damages and EOT in Singapore.

2. A real programme, submitted early

The public sector form PSSCOC (Public Sector Standard Conditions of Contract) requires the contractor to submit a programme within 30 days of the letter of acceptance, allows the Superintending Officer to require a revised programme when progress departs from it, and allows 10% of interim payments to be retained until an adequate programme is provided.¹⁶ A private A&A contract can adopt a lighter version: a bar-chart programme with key milestones before work starts, updated fortnightly.

3. Notice and extension-of-time discipline

An EOT clause with clear grounds and notice periods protects both sides: the contractor gets more time for owner-caused delay, and the owner keeps its right to LDs. Without one, owner-caused delay can set time “at large” and leave the owner with only a claim for general damages.

4. Progress reporting

Weekly photographs, a short progress note against the programme, a look-ahead of the next two weeks’ activities, and a list of information the contractor needs from the owner. Reporting surfaces slippage when it is a few days, not a few weeks.

5. Payment linked to progress

Progress payments tied to completed and verified work (not to elapsed time) align the contractor’s cash flow with progress. In Singapore, payment claims and responses must also comply with the Security of Payment Act; see SOPA for owners and contractors.

6. Named key people and resources

Name the site supervisor and key subcontractors in the contract, and require the owner’s consent to change them. This targets the resource-thinning route by which under-priced jobs drift.

Common objections, answered

“Our job is small; these statistics are about megaprojects.” The overrun magnitudes in the megaproject literature are larger than on a small fit-out, but the direction and shape (most projects late, right-skewed outcomes) have been found across many project types and sizes.⁶ Use smaller numbers, not zero.

“The contractor promised the date in writing.” A programme is an inside-view forecast. The question is what happened on the contractor’s last five comparable jobs.

“We’ll just impose heavy LDs.” An LD rate that is extravagant compared with the greatest loss that could conceivably follow is at risk of being struck down as a penalty, and an unrealistically high rate may simply deter better contractors from bidding. Set LDs at a defensible estimate of your loss.

“The price gap is certain; the delay is only a possibility.” Correct, which is why the comparison uses probabilities. But a possibility with a 50% chance and a heavy tail is not a rounding error.

Summary

• On-time completion is the exception in the research literature. Start from that base rate.
• Delays are skewed and fat-tailed: a minority of bad outcomes dominates the expected cost.
• In competitive tendering, the lowest bid is statistically more likely to contain an under-estimate. Under-pricing is recovered through variations, quality drift or time.
• Compare options on expected total cost, and check the P90 and the worst-tail average, not just the most likely outcome.
• In the Illustrative example, the cheaper quote wins slightly more than half the time, yet costs about S$11,400 more in expectation and far more in the tail.
• Contract levers (LDs, programme, notices, reporting, named resources) make on-time delivery enforceable, but they do not replace choosing a contractor with a good delivery record.