International Construction Law Review
FROM UNKNOWABLE TO KNOWABLE
Majd Nasrallah, FCIArb
Construction Delay and Disputes Expert
ABSTRACT
AI tools are changing how projects spot delay risk. Some systems can warn of problems weeks before delay shows on site. This article explains why that matters for notice duties under two of the most widely used standard form contracts in international construction: the NEC4 Engineering and Construction Contract and the FIDIC Red Book (both 1999 and 2017 Editions). The central argument is that AI does not change the law. It can change the facts. If risk becomes knowable earlier, the duty to notify may be engaged earlier too. The article proposes a practical “actionable awareness” test to help tribunals navigate probabilistic AI outputs. It also addresses a point that is often overlooked: the NEC early warning obligation runs to both parties, and the shared-data environments that AI tools create may affect employer conduct as much as contractor conduct. Governance, not just legal doctrine, will determine how these disputes are resolved.1
1. INTRODUCTION
Construction delay analysis has long been a discipline defined by hindsight. Programmes are updated after events occur. Claims are assembled from records created, at best, contemporaneously with site progress. The formal identification of a delaying event typically lags behind the event itself, sometimes by weeks, sometimes by months.
Predictive project controls tools are beginning to close that gap. Machine learning platforms compare historic productivity data, supply-chain lead times and live site progress to generate probabilistic forecasts of delay risk. Research confirms that such models can classify delay risk with statistical reliability before slippage appears in the programme.2 But that research
1 Dentons, “Predictive technologies and the evolving standard of care in construction”, (Lexology, 24 February 2026).
2 Gondia, A, Siam, A, El-Dakhakhni, W and Nassar, AH, “Machine Learning Algorithms for Construction Projects Delay Risk Prediction”, (2020) 146(1) Journal of Construction Engineering and Management. The study is based on 51 building construction projects in Egypt, using decision-tree and naïve Bayesian classification. The authors expressly state that “the resulting numerical/categorical values [are] not necessarily transferable to other cases/data sets”. Risk inputs were scored through expert meetings, introducing expert judgment into the classification. Testing accuracy under 10-fold cross-validation was 47.2 per cent (decision tree) and 51.2 per cent (naïve Bayesian).
Pt 4] From Unknowable to Knowable
397