Aviva

Aviva, a multinational insurer serving 25.2 million customers across the UK, Ireland, and Canada, faced mounting pressure to modernize core insurance workflows — particularly medical underwriting review, claims processing, and customer service. In Property & Casualty, underwriting speed directly affects competitiveness and loss ratios; slow manual review cycles create bottlenecks that erode margins and frustrate customers. Despite a decade of ML investment, Aviva needed to consolidate fragmented capabilities into a unified platform capable of deploying and reusing AI across business units at scale — or risk falling behind as the sector entered a full-scale AI arms race.

Aviva built an in-house AI platform designed to deploy use cases rapidly and reuse them across business units — avoiding the dependency on external vendors for core ML capabilities. The foundation was over 150 Machine Learning models trained on Aviva's own proprietary claims data, accumulated over more than a decade. Predictive ML was applied to price over 98% of retail business in UK Personal Lines. In medical underwriting, AI was integrated directly into case review workflows to accelerate assessments. For customer service, AI tools were deployed to reduce call wrap time for agents in Direct Wealth, with rollout extended to Insurance, Wealth & Retirement (IW&R). A partnership with OpenAI was also established to layer generative AI capabilities on top of this predictive ML foundation, with voice-enabled agentic claims handling in active development.

Aviva's AI programme has delivered measurable outcomes across three distinct operational areas:

• ~£100 million saved through claims transformation in UK General Insurance
• 50% reduction in medical underwriting case review time
• 20% reduction in customer service call wrap time for Direct Wealth agents, now rolling out to IW&R

Beyond the headline numbers, the in-house platform enables ongoing reuse of models across divisions, compounding returns from earlier ML investments. An AI-enabled claims agent — built in-house, voice-enabled, and capable of handling simple claims end-to-end without human involvement — is in testing and expected to launch later in 2026, targeting further savings from the claims function.

• Proprietary data is the real moat. Aviva's decade of training ML models on its own claims data gives it an advantage that newer AI adopters cannot shortcut.
• In-house platforms outperform point solutions at scale. Building internal capability to deploy and reuse AI use cases across business units compounds returns faster than procuring vendor-specific tools.
• Start with high-volume, structured workflows. Medical underwriting and call wrap reduction were achievable early wins because the workflows were well-defined and data-rich.
• Agentic AI will redefine claims handling. Voice-enabled agents handling simple claims end-to-end represent the next cost frontier — insurers should be piloting now, not planning.