For a major construction group, we deployed a claims management process using GPT-4 Vision and llama3.3 70b on-premise, reducing claims processing time from several weeks to a few hours while respecting regulatory constraints and data confidentiality.
ABOUT THE CLIENT
The construction group is a major player in infrastructure and large projects in Europe. With over 10,000 employees and projects in more than 20 countries, the company manages complex construction sites requiring precise coordination and rigorous incident and claims management.
CHALLENGE
Claims management represented a major challenge for operational and legal teams. The manual process required an average of 3 to 4 weeks to:
Collect photographic evidence
Analyze contractual documents
Draft claims
Coordinate different stakeholders
The complexity was amplified by:
The diversity of claim types
Multilingual documents
Regulatory constraints
The need to quickly handle urgent situations
SOLUTION
CAPTURE AND CERTIFICATION:
A mobile application enables geolocated capture of site incidents with strong user authentication. Data is automatically timestamped and securely stored, ensuring its evidential value.
ADVANCED AI ANALYSIS:
The system combines two AI models:
GPT-4 Vision analyzes construction site photos, identifies defects, and assesses their severity
Llama3.3 70b processes contractual and technical documents in multiple languages, extracts relevant clauses, and links them to incidents
DOCUMENT AUTOMATION:
The platform automatically generates claims by adapting to the company's historical writing style. It integrates photographic evidence, contractual references, and maintains consistency across all communications. The multilingual interface enables smooth collaboration between international teams.
Integration with existing systems (ERP, EDM) ensures complete traceability and real-time monitoring of claims, from creation to resolution.
METHODOLOGY
The project was completed in 6 months by a team consisting of:
1 AI Engineer
1 Lead Fullstack Developer
1 Product Owner
Key Phases
Analysis of needs and existing processes
Prototype development
Testing on a limited scope
Progressive deployment
User training
Continuous optimization
RESULTS
The solution's impact on operations is significant:
Operational Performance
Processing time reduction: from 3-4 weeks to 24-48 hours
Automation rate: 80% of standard claims
Volume processed: 500 claims per month
Quality and Accuracy
95% accuracy in document analysis
90% user satisfaction
60% reduction in processing errors
PERSPECTIVES
The project paves the way for predictive intelligence, IoT integration, and expansion into litigation management and predictive maintenance, demonstrating AI's transformative potential in the sector