Monte Carlo Cost Risk Analysis for Bridge Projects: A Plain-English Guide

Most bridge engineers were trained to produce a single-point cost estimate: add up the quantities, multiply by unit prices, apply contingency, and report a number. But a single number is a fiction — it implies a precision that preliminary bridge estimates simply do not have. Construction costs for a 200-ft bridge can swing 25–35% depending on market conditions, foundation surprises, and contractor competition.

Monte Carlo analysis replaces the fiction of a single number with an honest range. This guide explains what it is, how it works, why it matters for capital programming, and how to produce one without a statistics background.

Why a Single-Point Estimate Is the Wrong Tool for Preliminary Bridge Budgets

When you're doing a TS&L study or a corridor-level capital program estimate, you're working with Class C or Class D accuracy: ±20–30% at best. A $5M point estimate on a project with 25% uncertainty means the actual cost could reasonably land anywhere from $3.75M to $6.25M. Presenting "$5,000,000" implies you know more than you do.

The consequences are real. If a DOT programs $5M for a bridge that comes in at $6.5M, a funding gap appears at the worst possible time — at final design or even during construction. Capital programming based on systematically optimistic point estimates is one of the primary reasons bridge programs chronically run over budget.

What Monte Carlo Analysis Actually Does

Here is the conceptual process, stripped of jargon:

1. Identify the uncertain inputs. In a bridge cost estimate, the uncertain variables are unit prices (concrete, steel, labor), quantities (which depend on foundation depths not yet known), and contingency (for unknown unknowns).
2. Assign a range to each uncertain input, not a single value. For example: concrete unit price is "most likely $450/CY, could be as low as $380 or as high as $560." This is represented as a probability distribution — typically triangular or normal.
3. Run thousands of simulations. In each simulation, the model randomly draws a value for each uncertain input from its distribution, calculates the total project cost, and records the result. Bridge Copilot runs 10,000 iterations.
4. Read the output distribution. After 10,000 iterations, you have 10,000 simulated project costs. Sort them. The 10th percentile value is P10, the 50th percentile is P50, the 90th percentile is P90.

Reading the Three Key Values

What Drives the Width of the Cost Range?

The gap between P10 and P90 tells you how much cost uncertainty you're carrying. A tight range (P10 to P90 within ±15%) means your estimate is well-anchored. A wide range (±40% or more) signals that a key cost driver is highly uncertain — typically:

• Foundation type unknown: The difference between spread footings and driven piles can be $200K–$600K per substructure unit. Without a geotechnical report, this drives enormous range.
• Steel price volatility: Structural steel prices have ranged from $0.50/lb to over $1.00/lb in recent years. For a steel plate girder bridge, this is the single largest variable.
• Remote site with limited contractor competition: Rural or Alaska/Hawaii projects often see 25–40% higher bids due to reduced bidder competition and mobilization costs.
• Scope definition quality: A poorly defined preliminary scope produces a wider range than a well-scoped one. This is expected and honest — the range is communicating the state of your knowledge.

The Traditional Alternative: Contingency Percentages

The traditional approach to uncertainty in preliminary estimates is to apply a contingency percentage: "base estimate plus 20% contingency." This approach is faster but has two weaknesses:

1. It's symmetric. A 20% contingency implies costs are as likely to run below estimate as above — which is generally false. Preliminary bridge estimates tend to underestimate final bid prices more often than they overestimate, particularly when foundation conditions are unknown.
2. It doesn't communicate the risk profile. "Plus 20%" doesn't tell the owner whether you're 70% confident or 90% confident that the budget is adequate. P80 or P90 explicitly communicates confidence level.

For quick scoping at the program level, a flat contingency is still useful. For any project where a formal budget authorization will be made, a probabilistic range is more honest and more defensible.

How Bridge Copilot Runs Monte Carlo Analysis

Bridge Copilot generates a Monte Carlo cost risk analysis automatically for every bridge design it produces. The process:

1. The base cost estimate is computed from regional unit prices and derived quantities.
2. Uncertainty distributions are assigned to unit prices (±15–25% for major items), quantity line items (±10–20%), and a global contingency factor.
3. 10,000 iterations are run, each sampling from these distributions.
4. The output displays a histogram of simulated costs and reports P10, P50, and P90 as dollar values.
5. The PDF export includes the Monte Carlo chart and P-value table in a format ready for capital programming submittal.

The entire analysis runs in seconds, alongside the structural computations. Engineers reviewing the output should treat the range as an honest characterization of estimate uncertainty at the preliminary design stage — not as a refined cost model.

Practical Takeaways for Bridge Engineers

• Always present preliminary bridge estimates as a range (P50/P90) rather than a single number, even if you don't run a formal Monte Carlo analysis.
• For capital programming, budget at P80 unless agency policy specifies otherwise.
• The width of your P10–P90 range is a direct measure of how well your scope is defined. Narrow the range by resolving foundation unknowns early.
• A Monte Carlo analysis is not more accurate than the input distributions — garbage in, garbage out. The value is in making uncertainty explicit, not in producing a number that looks more precise than it is.
• When an owner asks "how much will this bridge cost?" the honest answer is a range with confidence levels, not a single figure.

© 2026 Bridge Copilot LLC — Home · How It Works · Blog · Privacy Policy