AXA Switzerland

In Switzerland's commoditised P&C insurance market, where pricing transparency severely limits product differentiation, the claims experience has become the primary determinant of customer loyalty and retention. AXA Switzerland's traditional workflow relied on manual triage, document review, and physical damage assessment across motor vehicle, property, and health lines — a labour-intensive model producing multi-day settlement timelines and rising administrative costs. Switzerland's four-language environment (German, French, Italian, English) further complicated document handling and consistency. As digital-native competitors raised customer expectations for instant service, the gap between AXA's operational throughput and market expectations created measurable churn risk and unsustainable unit economics per claim.

AXA Switzerland deployed an end-to-end AI claims automation platform enabling straight-through processing (STP) across motor, property, and health lines. Computer vision models — trained on hundreds of thousands of vehicle damage images — analyse policyholder-submitted photographs to classify damage types, estimate repair scope, and generate cost assessments that correlate closely with subsequent professional evaluations. Multilingual OCR and NLP extract structured data from German, French, Italian, and English documents including police reports, medical invoices, and repair estimates. The platform was built via a hybrid model: AXA Group's global AI infrastructure provided scale, while Switzerland-specific adaptations addressed local regulatory frameworks including TARMED and DRG for health claims. Critically, fraud scoring was embedded as an in-line pipeline step rather than a downstream gate, allowing high automation rates without sacrificing fraud controls.

Overall claims processing costs declined by an estimated 25%, with savings partially reinvested in expanded customer service capabilities. Motor vehicle claims achieved approximately 40% straight-through processing, with qualifying cases settled within 24 hours of notification — compared to multi-day timelines under the prior process. Key outcomes:

• ~40% motor claims STP rate with no human intervention required
• 25% reduction in overall claims processing costs
• <24-hour settlement for qualifying motor claims
• AI fraud detection captured a materially higher proportion of fraudulent claims than the prior rules-based system while simultaneously reducing false-positive investigations of legitimate claims
• Policyholders settled within 24 hours reported substantially higher satisfaction scores regardless of settlement amount

• Speed drives satisfaction more than settlement amount: Same-day settlement produced substantially higher customer satisfaction scores regardless of payout size — P&C insurers should treat cycle time as a primary CX metric, not just an operational one.
• Embed fraud scoring inside the pipeline, not after it: In-line fraud detection enables high STP rates while reducing both missed fraud and false-positive investigations — a downstream gate creates unnecessary friction.
• Hybrid global-local deployment accelerates time-to-value: Leveraging Group-level AI infrastructure with local regulatory adaptation delivered scale economies unavailable to standalone insurers.
• Multilingual NLP is a prerequisite in multi-language markets: Achieving consistent extraction quality across all four Swiss languages was a foundational requirement before full automation was viable.