Client expanding to Denmark & Norway · Phase 3: AI-powered dynamic pricing for commercial insurance (in progress)
Who we worked with
A mid-sized insurance company based in Northern Europe, offering property, casualty, and commercial insurance products across Finland and Sweden. With approximately 250 employees and 40,000+ claims processed annually, the company was poised to expand into Denmark and Norway.
Their core claims management platform — a 14-year-old Java EE (J2EE) monolith running on an on-premise WebLogic server — handled everything from policy administration to claims payment, but manual workflows and fragmented data were holding back both speed and accuracy.
18-day claims cycles in a 5-day world
Competitors were settling routine claims in under a week with automated workflows. A 14-year-old monolith, manual research across five browser tabs, and paper-based field reporting were leaving the company structurally unable to compete.
Slow Claims Processing
Average time from claim submission to resolution was 18 days. Competitors with automated workflows were achieving 5–7 days, creating a visible competitive disadvantage.
18-day average cycleUnderwriters Buried in Routine Work
60% of underwriter time was spent on low-complexity claims requiring no real judgment — each reviewing approximately 15 claims per day manually.
60% routine claimsNo External Data Integration
Underwriters manually researched each claim across 4–5 browser tabs — national property registries, weather services, police reports, credit bureaus — all outside the system.
5 disconnected sourcesPolicies Locked in Scanned PDFs
Policy documents were stored as unindexed scans. Finding specific coverage terms required reading entire documents manually, slowing every adjudication decision.
No Mobile Access for Field Adjusters
Field adjusters had to return to the office to file reports — adding 1–2 days to every claim cycle that required an on-site visit.
+1–2 days per site visit3-Week Regulatory Reporting
Quarterly compliance reports for the Nordic Financial Supervisory Authority required 3 weeks of manual data compilation — an enormous recurring overhead.
3-week manual processTwo overlapping phases, parallel processing during cutover
A 28-week phased delivery — modernizing the claims platform while building the AI layer in parallel, with a 4-week dual-run period where old and new systems processed claims simultaneously to validate accuracy before full cutover.
- Migrated from J2EE/WebLogic to Java Spring Boot microservices on AWS — organized by insurance domain: policy admin, claims intake, adjudication, payments, reporting, document management
- Built a responsive Angular front-end — underwriters on any device, field adjusters filing reports from mobile on-site
- Document management system with structured storage — policies indexed and searchable instead of opaque scanned PDFs
- APIs integrating 6 external data sources directly into the claims workflow: property registries, weather services, police databases, credit bureaus, claims history & geolocation
- Automated regulatory reporting — quarterly FSA reports generated in under 4 hours, down from 3 weeks
- GDPR-compliant data handling with automated retention policies and right-to-erasure workflows for Nordic operations
- AI claims triage system — incoming claims automatically classified as routine, moderate, or complex using a gradient boosting model trained on 5 years of historical outcomes
- Straight-through automation for routine claims (≈40% of volume) — processed end-to-end with no human review for high-confidence known patterns
- AI document extraction engine (NLP + OCR) — policy documents and claim evidence (photos, invoices, police reports) parsed automatically; key fields populated in the claims record
- Fraud detection model scoring each claim on 38 features: timing, amount vs. policy, claimant history, geographic patterns, document analysis
- Underwriter assistance panel for complex claims: relevant policy terms, similar past claims & outcomes, external data lookups, and AI-recommended settlement range
85% auto-extraction accuracy
XGBoost + SHAP explainability
Labeled outcomes + fraud cases
Explainable to regulators
National Property Registries
Instant property ownership, valuation, and encumbrance data for property claims
Weather Services
Historical weather data correlated with claim date and location for storm/flood claims
Police Report Databases
Automated retrieval and cross-reference of police reports for theft and accident claims
Credit Bureaus
Claimant financial history as a fraud risk signal for high-value claims
Claims History Databases
Industry-wide prior claims for repeat claimant pattern detection
Geolocation Services
Location verification and geographic fraud pattern analysis
Built for compliance, scale & intelligent automation
Results that redefined their claims operation
Across processing speed, underwriter productivity, fraud detection, and regulatory compliance — all delivered in 28 weeks with a validated parallel-run cutover.
Processing
Productivity Gain
Auto-Extracted
Cases Flagged
7 engineers, one cohesive delivery
A specialist team combining insurance domain knowledge with Java/cloud expertise and ML engineering — structured for end-to-end ownership across the full claims stack.
Solution Architect
Insurance domain expertise, microservices design, AWS architecture, GDPR/FSA compliance
Senior Java Developer × 2
Spring Boot microservices, claims adjudication logic, external data API integrations
Senior Angular Developer
Responsive web & mobile UI, underwriter dashboard, field adjuster mobile workflow
ML Engineer
Claims triage model, fraud detection (XGBoost), NLP document extraction, SHAP explainability
DevOps Engineer
AWS ECS, Terraform, GitLab CI, automated FSA reporting pipelines
QA Engineer
End-to-end claims workflow testing, parallel-run validation, GDPR compliance verification