Accident Claim Extraction

A European motor insurance company processes over 120,000 claims per year. Each motor claim begins with a European Accident Statement (EAS) — a dual-page form capturing driver details, vehicle data, damage descriptions, and liability circumstances. These forms can be handwritten or digital, and Mysa achieves 93%+ extraction accuracy on handwritten submissions and near-100% on digital ones. Every extracted field is immediately leveraged to make decisions, and when information is missing, the system automatically follows up with users to collect it.

Motor claims intake at scale involves processing high volumes of accident statements with strict accuracy requirements. Three factors define the opportunity:

• •Document complexity at volume. Accident statements arrive in varied conditions — handwritten, photographed at angles, partially completed. At 120,000+ claims per year, even small per-claim inefficiencies compound into significant operational cost.
• •Accuracy drives downstream efficiency. Field-level precision on policy numbers, registration plates, dates, and liability assessments determines whether a claim flows through straight-through processing or requires manual intervention.
• •Scalability without proportional headcount. Absorbing volume increases traditionally requires hiring and training — a process that takes months. With Mysa, claim teams can scale up or down with volume without adding or removing staff.

Mysa's Document Intelligence pipeline is calibrated specifically for European Accident Statements. The system extracts and validates data from both pages — including handwritten fields, checkboxes, and vehicle diagrams — then generates a structured FNOL record ready for adjudication.

The key differentiator: Mysa retrained its extraction models on the insurer's own document corpus — mixed languages, variable scan quality, non-standard layouts. This corpus-specific training is what pushed accuracy from baseline to production-grade. The retrained annotations remain proprietary to the insurer, meaning they retain full ownership and can access or re-use the data at any time.

1. 1.Initial calibration (2 days): Mysa's extraction models were tested on a sample of live accident statements to establish baseline performance.
2. 2.Corpus-specific retraining (2 weeks): Models were retrained on the insurer's document corpus — proprietary data annotations that the insurer retains full ownership of.
3. 3.Full deployment (week 3): Integrated into the insurer's claims workflow with continuous accuracy monitoring and feedback loops.

More claims now flow through straight-through processing without manual intervention. When information is missing, Mysa automatically follows up with users to collect it — eliminating a major bottleneck. The time between claim submission and claim decision has dropped significantly, and claim teams spend their time on assessment and decision-making rather than transcription. Annual cost reduction exceeds EUR 1.2M, with full ROI achieved within the first quarter of deployment.

“Mysa didn't just digitise EAS — it fundamentally changed how quickly we can act on claims. We went from a 20-minute manual process to instant, structured data.”

— Head of Claims Operations