Automation & Integration in Munich

Insurance / Property and Casualty Insurance

A Munich-based insurance group (EUR 18B GWP, 12 million policyholders, operating across motor, property, liability, and accident insurance lines — 8,500 employees in Germany, with smaller operations in Austria and Switzerland) needed to automate its claims processing to reduce cycle times, detect fraud more effectively, and improve customer satisfaction in an increasingly competitive German market. (1) Claims volume was overwhelming: the group processed 3.8 million claims per year — 2.4 million motor, 800,000 property, 350,000 liability, and 250,000 accident/health. Each claim followed a manual workflow: FNOL intake (phone, email, or letter), data entry into the claims system, coverage verification, reserve setting, investigation/adjustment, and settlement/payment. The claims operation employed 2,200 FTEs — and still averaged 18 days for motor claims and 42 days for property claims. (2) Fraud was costly: the group estimated EUR 280M in annual fraud losses (7.4% of claims costs) — but the existing fraud detection was rule-based and caught only the most obvious patterns. Organised fraud rings using sophisticated methods (staged accidents, inflated damage claims, phantom injuries) were difficult to detect with static rules. The fraud investigation team (45 FTEs) could only investigate 8% of flagged claims — the remaining 92% of flags were either false positives or investigated with insufficient depth. (3) Supplier network was manual: the group worked with 4,500 repair shops, 2,200 medical providers, and 800 property restoration companies. Referrals, authorisations, and invoice processing were managed through phone calls, faxes, and email — creating delays and data entry errors. Repair shop selection was not optimised — shops were selected based on proximity rather than quality, cost, and capacity data. (4) Customer experience was poor: customers filed claims through phone (65%), email (25%), and letter (10%). Phone FNOL took an average of 12 minutes and required the customer to provide information that the insurer already had (policy number, contact details, vehicle information). Status updates required calling the claims team — generating 4.2 million status enquiry calls per year. Customer satisfaction with claims handling was 3.2/5 — the lowest-rated aspect of the insurer's service. (5) Reserve accuracy was inconsistent: initial reserves were set by claims handlers based on experience and guidelines — but varied significantly by handler (coefficient of variation: 35% for similar claims). Inaccurate reserves affected financial reporting and reinsurance recovery calculations. The actuarial team spent 6 weeks per quarter adjusting reserves based on development patterns.

We built a comprehensive claims automation platform over 14 weeks. (1) Digital FNOL: multi-channel digital claims intake — mobile app (photo-based damage documentation for motor and property), web portal, and AI-powered phone intake (voice agent handling standard FNOL calls). Auto-population: customer and policy data pre-loaded from the policy system — the customer confirmed rather than re-provided their information. Damage documentation: guided photo capture through the mobile app — AI models assessed damage severity from photos, providing immediate preliminary estimates for motor claims. Digital FNOL reduced intake time from 12 minutes to 4 minutes (phone with voice agent) and 3 minutes (mobile app). (2) Automated claims processing: coverage verification (automated checking of policy terms, coverage limits, deductibles, and exclusions — instant coverage confirmation for standard claims), reserve setting (ML-based reserve estimation using claim characteristics, historical development patterns, and current repair cost indices — producing consistent, data-driven initial reserves), routing (automated assignment of claims to appropriate handlers based on complexity, line of business, and required expertise — simple claims routed to straight-through processing, complex claims to specialist handlers), and straight-through processing (standard motor claims under EUR 2,000 with photo documentation processed automatically — damage assessment, repair authorisation, and payment without handler intervention). (3) AI fraud detection: ML fraud models trained on 5 years of claims data (15 million claims, including known fraud cases). Network analysis: identifying connected claims (shared addresses, phone numbers, repair shops, medical providers, and legal representatives) that indicated organised fraud rings. Behavioural analysis: detecting anomalies in claim patterns — filing timing, damage descriptions, medical treatment patterns, and document characteristics that deviated from legitimate claim patterns. Fraud scoring: every claim scored for fraud probability — high-risk claims routed to the investigation team with specific risk indicators highlighted. Investigation prioritisation: fraud team focused on the highest-value, highest-probability cases — replacing the previous approach of investigating a random sample of flagged claims. (4) Supplier network automation: digital referral — automated matching of claims to optimal suppliers based on location, specialisation, quality ratings, cost performance, and current capacity. Electronic authorisation: suppliers received work authorisations digitally with scope of work, authorised amounts, and timeline expectations. Digital invoicing: suppliers submitted invoices electronically — automated three-way matching against authorisation, scope of work, and market-rate benchmarks. Quality tracking: automated capture of customer satisfaction data for each supplier interaction — feeding back into the supplier selection algorithm. (5) Customer communication: automated status updates — customers received proactive notifications at every stage (claim received, coverage confirmed, supplier assigned, repair scheduled, payment processed) through their preferred channel. Status portal: real-time claim status accessible through the mobile app and web portal — eliminating the need for status enquiry calls. Estimated timeline: AI-generated claim timeline showing expected milestones — customers knew when to expect resolution rather than calling to ask.

Claims cycle time: motor claims from 18 days to 6 days (67% reduction). Property claims from 42 days to 19 days (55% reduction). Straight-through processing: 35% of motor claims (under EUR 2,000 with photo documentation) processed without human intervention — from FNOL to payment in under 48 hours. Fraud detection: fraud detection rate increased from 4.2% to 8.8% of total fraud (estimated) — representing EUR 128M in additional fraud identified annually. Network analysis identified 12 previously undetected fraud rings in the first year. False positive rate reduced by 60% — fraud team investigated 25% of flagged claims (from 8%) with higher accuracy. Cost: annual fraud savings EUR 128M. Supplier efficiency: average repair time reduced by 22% through optimal supplier matching. Repair costs reduced by 8% through market-rate benchmarking and automated negotiation. Supplier satisfaction improved from 3.5/5 to 4.3/5 — faster payments and clear authorisations improved the relationship. Customer satisfaction: claims NPS from -8 to +32 (dramatic improvement). Status enquiry calls reduced by 75% (from 4.2M to 1.05M per year). Digital FNOL adoption: 55% within 12 months (35% mobile app, 20% web portal). Reserve accuracy: coefficient of variation for initial reserves from 35% to 12% — ML-based estimation produced more consistent, accurate reserves. Actuarial adjustment work reduced by 65% — initial reserves closer to ultimate outcomes. Staffing: claims team from 2,200 to 1,500 FTEs — 700 positions redeployed through natural attrition and retraining (no redundancies, per Betriebsrat agreement). Annual value: EUR 128M (fraud detection — additional fraud identified and prevented), EUR 85M (staffing efficiency — 700 FTE reduction through automation and straight-through processing at average EUR 65K fully loaded cost plus overhead savings), EUR 42M (supplier cost — repair cost reduction through optimal matching and market-rate benchmarking), EUR 28M (customer retention — improved NPS driving 2.1% retention improvement on EUR 18B GWP), EUR 15M (reserve accuracy — reduced actuarial work and improved financial reporting accuracy) = EUR 298M. Development cost: EUR 420,000. Payback: under 13 hours.