Pricing the Probable: How Outcome Simulation Is Rewriting Construction Legal Strategy

Every large construction dispute is, at heart, an argument about counterfactuals. Who delayed whom, by how many days, and at what cost, and what would the project's bottom line have looked like if the contested events had never happened? For most of the industry's history, lawyers answered those questions with experience, intuition, and a binder of precedents. Today, a growing share of the answer arrives as a distribution: a fan of thousands of simulated outcomes, each weighted by probability, that tells a general counsel not just whether a claim might succeed but how likely it is to clear a given threshold. That shift, from narrative judgment to quantified forecasting, is quietly reshaping how legal and risk teams in real estate and construction decide which battles to wage.

The stakes justify the rigor. Across more than 2,000 projects in 107 countries with a combined capital value of roughly US$2.25 trillion, the disputed sums totaled US$84.4 billion, with contested costs averaging nearly a third of each project's budget, according to HKA's CRUX Insight analysis. When the downside of a single claim runs to tens of millions, a tool that sharpens the odds even modestly pays for itself many times over.

$84.4B

Total sums in dispute, CRUX dataset

33%

Avg disputed cost as share of CapEx

26 mo

Avg ICC arbitration duration, 2024

79%

AI accuracy in a landmark outcome study

The Old Way: Judgment in a Binder

The traditional approach to evaluating a construction claim rested on a small number of experienced minds reading a large pile of documents. A delay analyst reconstructed the schedule by hand; a quantum expert tallied the costs; senior counsel weighed the credibility of witnesses and the temperament of a likely tribunal. The output was a memo and a range, "we think we recover somewhere between sixty and eighty cents on the dollar", anchored more in professional feel than in transferable method.

That feel was operating against a backdrop of genuinely punishing numbers. The most common single trigger of claims worldwide is change in scope, but the dominant force is a cluster of design failures, incorrect, incomplete, and late-issued design information, which together affected 44.8% of projects in one CRUX edition, more than scope change alone, as reported by Engineering News-Record. These are not exotic events; they recur on project after project, which is precisely why their financial consequences are, in principle, modelable.

The legacy method had three structural weaknesses. It produced single-point estimates that hid the real spread of outcomes. It buried the assumptions driving those estimates inside an expert's head, where opposing counsel and arbitrators could not interrogate them. And it scaled poorly: every claim required the same expensive, bespoke effort, so portfolios of disputes were rarely evaluated on a consistent basis. When proceedings can run, on average, 27 months to a final award, as the ICC's 2023 statistics show, a portfolio assessed claim-by-claim, gut-by-gut, is slow, inconsistent, and hard to defend.

The recurring causes of construction claims

Share of projects affected, global CRUX dataset

Source: HKA CRUX Insight global top-10 causes, as reported by Construction Briefing. Because the same handful of causes recur across projects, their cost and schedule impacts are well suited to probabilistic modeling.

The Shift: From Point Estimate to Probability Curve

The methodological breakthrough was not new, Monte Carlo simulation has been used in schedule risk analysis for years, but its migration into legal strategy is recent. Instead of forcing experts to commit to a single number for, say, the duration of an excusable delay, simulation lets them specify a range and a shape: most likely 40 days, but plausibly anywhere from 20 to 90. Run that uncertainty through ten thousand iterations across every contested variable, and the model returns a curve showing the probability of each net recovery. Academic work has demonstrated the technique's value in construction delay risk assessment, and quantum specialists increasingly apply the same logic to claims and disputes directly.

Running parallel to this is a second current: data-driven case analytics. The proof-of-concept that captured the legal world's attention came from a study in which a machine-learning model predicted the judgments of the European Court of Human Rights with 79% accuracy across 584 cases, simply by analyzing the language of the filings, as covered by the ABA Journal. Commercial outcome-prediction platforms now report accuracy in the 70 to 85% range depending on case type and data quality, according to industry reviews summarized by LeanLaw. Combine the quantum-side Monte Carlo with the analytics-side prediction, and a legal team can attach a probability not just to "how much" but to "will we win at all."

This matters most at the decision that dominates construction disputes: settle or proceed. Between 90% and 97% of civil cases in the United States resolve before a trial verdict, multiple analyses of court data confirm, as summarized by LegalClarity. If almost every dispute ends in a negotiated number, the strategic question is never really "can we win in court", it is "what is this claim worth as a settlement, given the distribution of outcomes if we don't settle?" Simulation answers exactly that question, and it does so in a currency both sides can argue over.

Why settlement modeling dominates: outcome paths for a disputed claim

Illustrative distribution of net recovery across 10,000 simulated scenarios

Illustrative simulation output using a triangular-style distribution; shape and settlement logic reflect Monte Carlo methods described by Long International. Values are schematic, not a forecast of any specific case.

What It Looks Like Now

In practice, an outcome-simulation workflow for a construction dispute moves through a recognizable sequence. The legal and delay teams break the claim into its contested components, entitlement, causation, quantum, and recoverable time. Each component is expressed as a range with a probability weight rather than a fixed value. Comparable past awards and tribunal tendencies, drawn from case-analytics datasets, inform the win-probability inputs. The engine then runs the scenarios and returns an expected value, a confidence band, and the probability of clearing key thresholds, break-even on legal spend, recovery of the full claimed amount, or a target settlement figure.

Crucially, the model also runs the forum question. Litigation, arbitration, and negotiated settlement carry different cost structures, timelines, and variance. Arbitration's reputation for speed is only partly deserved: ICC proceedings concluding in 2024 averaged 26 months with a median of 22, an improvement on 2023 but still over two years, per the analysis of the ICC Court's 2024 statistics. Modeling each forum side by side lets a client see the time-value and risk-adjusted cost of each path rather than choosing on reputation alone.

Resolution pathways compared for a high-value construction dispute
Pathway	Typical duration	Cost & control profile	Outcome variance
Negotiated settlement	Weeks to months	Lowest cost; parties retain control	Low, agreed figure
Mediation / DAB	1 to 6 months	Moderate; non-binding leverage	Low to moderate
Arbitration (ICC)	~26 months avg (2024)	Confidential; high fees; limited appeal	Moderate to high
Litigation	Often 2 to 4+ years	Public; appealable; cost-shifting risk	High, point outcome

The construction sector is an unusually fertile place for this approach because its disputes are concentrated and high-value. Construction, engineering, and energy together accounted for roughly 44 to 45% of all new ICC arbitration filings in recent years, per Pinsent Masons, and in some Gulf institutions construction and real estate made up 58% of cases, according to Reed Smith's review of 2024 institutional data. A practice area this large, this repetitive, and this expensive is exactly where probabilistic tooling delivers the most leverage.

Construction commands the arbitration docket

Construction, engineering & energy share of new ICC arbitration filings

Sources: ICC caseload composition via Pinsent Masons and Reed Smith. The sector's concentration makes it a natural proving ground for outcome analytics.

The Cost of Delay, Quantified

Outcome simulation does not exist in a vacuum; it is responding to a measurable worsening of the underlying problem. North American dispute values have climbed sharply, with Arcadis reporting an average U.S. dispute value of roughly US$60 million in its 2025 edition, as cited in Engineering News-Record, up from the low-$40-millions only a year earlier per the 2024 report. Schedule overruns, meanwhile, have hovered near two-thirds of planned project duration across CRUX editions. When the cost of being wrong about a claim rises this fast, the value of quantifying uncertainty rises with it.

North American dispute values are climbing

Average dispute value, U.S. / North America (US$ millions)

Sources: Arcadis Global Construction Disputes Reports via Giatec (2024) and ENR (2025); earlier years via Arcadis 2021 report. Figures mix U.S. and North America regional bases as published.

Reported outcome-prediction accuracy by matter type
Domain	Reported model accuracy	Notes
ECHR judgment study	~79%	584 cases, language-based model
Commercial litigation	~70 to 87%	Varies by dataset & platform
Contract disputes	~80 to 90%	Higher where data is structured
Case dismissals	~85%	Procedural prediction tasks

The Next Few Years

Three developments look likely over the next three to seven years. First, simulation will move upstream, from dispute resolution into contract drafting and project governance, pricing the litigation risk of a given clause or delay-notification regime before a shovel hits the ground. Second, outcome models will increasingly feed the fast-growing market for dispute finance: third-party litigation funding was valued at roughly US$15.2 billion globally in 2024, per Strategic Market Research, and funders are natural consumers of quantified win probabilities. Third, the analytics will become a negotiation language in their own right, with both sides arriving at mediation holding distributions rather than anchors.

None of this is risk-free, and the most serious risks are generic to predictive analytics rather than specific to any tool. A model's confidence band is only as honest as its inputs; garbage ranges produce a precise-looking but worthless curve. Historical award data encodes the biases and idiosyncrasies of past tribunals, which may not hold for novel project types or jurisdictions. And there is a behavioral hazard: a crisp number invites over-reliance, tempting teams to treat a 75%-confidence output as certainty and to under-weight the residual 25% where careers and balance sheets are made or broken. The legal-technology field's own surveys repeatedly stress human oversight; the Thomson Reuters Institute has framed AI's role as augmenting professional judgment, not replacing it. The discipline that experienced counsel bring, knowing which assumptions are fragile, does not disappear; it moves to the front of the process, where the ranges are set.

Conclusion

The arc here runs from intuition to instrumentation. Construction legal strategy began with experts who held the whole dispute in their heads and produced a number they could defend only by their reputation. It is moving toward a world in which that number is a distribution, explicit about its assumptions, comparable across a portfolio, and legible to opponents, funders, and tribunals alike. Given a sector where contested sums run into the tens of billions and resolutions take years, the appeal is obvious. The enduring task for legal teams is not to surrender judgment to the model but to aim it: to decide which uncertainties matter, to set the ranges honestly, and to read the resulting curve with the humility that any forecast deserves.