.png)
STRABAG
Construction · Austria / Europe
70% of flops caught at month 3
STRABAG's DARIA model detects 70% of financially failing construction projects within the first 3 months of build time
XGBoost model trained on 11,000 historical projects from 12 countries scores 1,000+ live road construction projects monthly, shifting controllers from reactive fire-fighting to proactive risk-based prioritisation.
Friction
Construction projects regularly run over time and over budget, but losses only become visible when it is too late to course-correct. Controllers were manually monitoring hundreds of projects simultaneously under time pressure, unable to spot warning signs early enough.
Breakthrough
DARIA compares each live project to patterns from 11,000 historical failing projects. Rather than a black-box score, it generates a visual comparison that makes the reasoning transparent to controllers and project managers, enabling intervention conversations, not just alarms.
Impact
70% of financial failing projects detected within the first 3 months of build time, well before intervention becomes impossible. 1,000+ active road construction projects scored automatically every month. Deployed across 12 countries on 11,000 historical training projects. Next step: extending the model to the pre-tender phase to identify risks before a project starts.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach below!
Problem
In large construction businesses, project portfolio risk management is structurally difficult: there are too many projects to monitor individually, the warning signs are often buried in project accounting data, and the people closest to the project have the most incentive to report optimistically. By the time a problem surfaces in a management review, the window for intervention has often closed.
What made it smart
The decision to use visual comparison with historical failing patterns rather than just outputting a risk score was crucial for adoption. Controllers do not just need to know a project is at risk; they need to understand why, in a way they can explain to a project manager. Explainability drove adoption.
Technical approach
XGBoost gradient-boosted decision tree model trained on commercial project data from 11,000 historical projects across 12 countries. Monthly batch analysis of all active road construction projects, with risk flags pushed directly into the controller's existing digital workspace. Visual comparison with historical patterns provides explainability. 2.5 years of development in collaboration with STRABAG's internal data science team.
Strategic lesson
Financial risk AI in project environments creates value only if controllers act on the flags. That requires explainability, not just a score but a reason. The model that shows its work gets used; the black box gets ignored.
Reflection question
In your portfolio of projects, initiatives, or accounts, where are the early warning signals that systematically get noticed too late, and what would change if AI flagged them at month 3 instead of month 9?