Every infrastructure budget carries uncertainty. This introduction explains the anatomy of a cost estimate — base, contingency and escalation — and why probabilistic methods, not a flat percentage, produce a defensible number you can take to a funding gate.
A cost estimate is a forecast made under uncertainty. Scope is still evolving, quantities are approximate, rates move with the market, and discrete events — a geotechnical surprise, a planning condition, a supply-chain shock — may or may not occur. A single number presented without an explicit treatment of that uncertainty is a guess dressed as a fact. The discipline of cost-risk and contingency is what turns a point estimate into a defensible budget: a figure with a stated confidence level that a funding body, an assurance reviewer, or a board can rely on.
The reason this discipline exists is well documented. Infrastructure projects systematically come in over budget, and the dominant cause is not bad luck but optimism bias — the consistent tendency of estimators and proponents to assume the most favourable scope, productivity and pricing. Left unchecked, optimism bias produces a base estimate that is closer to a best case than an expected case. Contingency, sized to genuine residual risk, is the structured correction for that bias.
A number tied to a confidence level (P50, P90) and backed by a risk model holds up under independent review and at a funding gate.
Probabilistic modelling and reference-class cross-checks expose the gap between a best-case base estimate and the expected outcome.
Pricing the right contingency up front means the budget already accounts for risks that would otherwise surface as cost overruns mid-delivery.
Across the major Australian frameworks — Queensland's TMR manual, the national RES Contingency Guideline, and the Commonwealth's DITRDCA cost-estimation guidance — the message is the same: above the relevant value threshold, model risk probabilistically and report contingency at P50 and P90. A flat percentage is explicitly discouraged.
Every framework decomposes a project budget the same way. Three components stack to produce the figure that is actually funded — the out-turn cost.
Base estimate + Contingency + Escalation = Out-turn cost
Keeping these three components separate is not a formality. It lets a reviewer see what is known scope, what is risk provision, and what is the time-value of money — and challenge each independently.
A waterfall view of how the components stack. The base estimate is the expected cost of defined scope; contingency is added for residual risk; escalation converts the result from real to out-turn (nominal) dollars.
The expected cost of the defined scope, excluding contingency and escalation — built bottom-up (first principles) from quantities and rates.
The provision above the base for residual (inherent + contingent) risk, sized to a chosen confidence level. It is expected to be expended, not a slush fund for scope change.
Cost growth over time (inflation / market) converts real cost to out-turn (nominal) cost. Base + contingency + escalation = the out-turn total that is actually funded.
The expected cost of the defined scope, excluding contingency and escalation. Built bottom-up from measured quantities and unit rates.
The provision above the base estimate for residual (inherent + contingent) risk, sized to a chosen confidence level; expected to be expended.
Cost growth over time due to inflation and market conditions; converts real cost to out-turn (nominal) cost and is kept separate from contingency.
Base + contingency + escalation — the nominal total. This is the figure that funding decisions are made against.
Contingency covers two distinct kinds of uncertainty. A complete risk model captures both — missing either understates the provision required.
Variability in items that are already measured in the base estimate — the quantities and rates themselves. The base price of bulk earthworks or structural concrete is never exact; it sits within a range. Inherent risk is modelled by ranging those base line items rather than adding new ones.
Modelled as: probability distributions (triangular, PERT, lognormal) applied to base quantities and rates.
Discrete risk events not in the base estimate — they may or may not occur. A latent-condition claim, a heritage discovery, a regulatory change. Each is modelled as a probability of occurrence multiplied by a cost impact, so it contributes to the budget only in proportion to its likelihood.
Modelled as: probability × impact, captured from a facilitated risk workshop and held in a dollarised risk register.
Contingency provisions for identified residual risk on the agreed scope — it is expected to be drawn down as those risks materialise. New scope is a budget change, managed separately. Conflating the two is one of the most common ways a contingency loses credibility under review.
A flat percentage is quick, but it cannot tell you the confidence level of the resulting budget. Monte Carlo simulation can — which is why every major Australian framework prefers it above its value threshold.
Adds a single percentage — say "+15%" — on top of the base. It is fast and transparent, but the figure is a judgement call with no stated confidence level: it cannot say whether the budget covers a 50%, 70% or 90% chance of not being exceeded. It also ignores how individual risks interact.
Appropriate for: early strategic estimates and smaller projects below the probabilistic threshold (broadly under $10M–$25M depending on framework).
Samples the modelled distributions over thousands of iterations (typically 5,000–10,000) to produce the full probability distribution of cost outcomes — the S-curve. Contingency is then read off at a chosen confidence level rather than guessed, and the difference between P50 and P90 is the risk-based contingency.
Required for: projects above the framework threshold; reported at P50 and P90 with S-curve, histogram and tornado outputs.
On the left, a deterministic estimate bolts a flat 15% onto the base — no confidence level attached. On the right, the same base modelled probabilistically: contingency emerges as the distance between the P50 and P90 of the cost distribution.
A "+15%" tells you nothing about the chance of overrun, ignores correlations between risks, and is easily challenged as arbitrary. Two estimators can defend wildly different percentages with equal confidence.
The contingency falls out of the analysis — it is the gap between the P50 and the P90 of the simulated distribution, traceable to specific modelled risks and re-runnable for a reviewer.
P-values are simply points read off the cumulative probability curve — the S-curve — that the Monte Carlo simulation produces.
The cost with a 50% probability of not being exceeded — the median. A coin-flip: the actual cost is equally likely to land above or below it.
The cost with a 90% probability of not being exceeded — the high-confidence budget. Only a 10% chance the project costs more than this.
Cost distributions are right-skewed, so the average (mean) sits above the P50 median. The 50/50 point is the P50, not the mean — a distinction worth holding onto.
The S-curve plots the probability (vertical) that the project cost will not exceed a given value (horizontal). Read up from a cost to find its confidence; read across from a confidence to find the budget. The band between P50 and P90 is the risk-based contingency.
Pick a confidence on the vertical axis (say 90%), trace across to the curve, then down to the cost — that is your P90 budget. The curve is the single most-quoted Monte Carlo output.
The horizontal distance from P50 to P90 is the risk-based contingency. In this example, a $100M P50 and a $118M P90 imply an $18M (18%) contingency — derived, not assumed.
Three documents, written by different bodies for different purposes, prescribe the same discipline: a first-principles base estimate, risk modelled with Monte Carlo, contingency reported at P50 and P90, and escalation kept separate.
The TMR Project Risk Management and Contingency Development Process Manual (1st Edition, March 2023) is mandatory for QTRIP / state-funded transport projects. It splits risk into inherent (planned) and contingent (unplanned), mandates Monte Carlo above $10M (state) / $25M (federal), and sets the Approved Project Delivery Value at P75 at tender award, with P90−P50 held at portfolio level.
TMR risk & contingency →The RES Contingency Guideline (3rd Edition, 2025) is guidance, not a mandatory standard, but it is the cross-jurisdiction authority. It recommends simulation above $10M, frames contingency around a Performance Measurement Baseline at P50 with a Management Reserve to P90, and treats reference-class forecasting as a benchmark check rather than a method.
RES contingency guideline →The Commonwealth's Cost Estimation Guidance (Guidance Notes, v2.0, Nov 2023) is mandatory for Infrastructure Investment Program funding. Monte Carlo is required for out-turn cost above $25M; P50 is the approval basis and P90 is notionally held and released on demonstrated need. It prefers the risk-factor method and requires correlation to be defined.
DITRDCA cost estimation →The thresholds look like they conflict; they don't. Whether a project crosses TMR's $10M state trigger or the Commonwealth's $25M out-turn trigger, the answer is the same — model it probabilistically. See how all three line up side by side in Frameworks Compared.
Most estimators can produce a contingency number. Far fewer can produce one that satisfies the specific framework a project is funded under and defends it against an independent review. That is the work Cenex does as standard.
Bottom-up PLMS build-ups — direct, indirect and margin — auditable to a four-level WBS. The base is the foundation everything else is challenged against.
@RISK models at 5,000–10,000 iterations, with inherent risk ranged on line items, discrete contingent risks priced as probability × impact, and correlation modelled — reported as S-curve, histogram and tornado.
Contingency reported at P50 and P90 (plus P75 for TMR award, P60 for VIC where relevant), with escalation to out-turn handled separately using framework-appropriate indices.
With no downstream delivery interest, Cenex challenges rather than inflates the contingency — sized to genuine residual risk, cross-checked with reference-class forecasting, and signed off by a Chartered Engineer.
Read the TMR, RES and DITRDCA framework pages, see them compared side by side, or return to the hub overview. For the engagement view, see our Risk Modelling & Management service.
Tell us the framework your project is funded under — TMR, Commonwealth / DITRDCA, or a state mandate — and Cenex will build the probabilistic estimate and contingency it requires.