Nationwide Insurance

Nationwide Insurance, a Fortune 100 property and casualty carrier, faced a data fragmentation problem common to large insurers: vast volumes of policyholder, claims, and behavioral data spread across siloed systems with no unified platform to extract predictive value from it. P&C insurers operate in a highly competitive, margin-sensitive environment where the ability to detect fraud early, price risk accurately, and retain profitable policyholders directly affects combined ratios. Without a scalable AI infrastructure, Nationwide's data scientists were spending more time on manual feature engineering and infrastructure work than on modeling — limiting how many business problems they could tackle simultaneously and slowing the path from insight to production.

Nationwide built a centralized data science function using H2O-3 open source and H2O Driverless AI as its core AutoML platform. The team developed a proprietary, patented model factory — an internal system for managing the full lifecycle of AI and ML models across the enterprise, from prototyping through production monitoring. Driverless AI's automated feature engineering and model search capabilities allowed analysts to rapidly iterate across a wide range of use cases: customer churn prediction, intelligent call routing, risk segmentation, fraud detection, underwriting optimization, and customer 360 profiles. By centralizing tooling and governance under one platform, the team could run multiple workstreams in parallel while maintaining consistent standards for model quality, interpretability, and bias auditing across each business unit.

The model factory delivered measurable impact at significant scale:

• 25 billion models scored across all production use cases, demonstrating the platform's throughput at enterprise scale
• Millions of dollars in financial savings, driven by improvements in fraud detection, underwriting precision, and customer retention
• Materially reduced model prototyping time, allowing data scientists to move from problem definition to a working prototype faster than under the previous manual workflow

Beyond the headline numbers, Nationwide gained a more granular understanding of household-level changes, enabling more timely and personalized member outreach. The model factory's monitoring layer ensured that production models remained statistically unbiased and stable over time — a critical requirement in regulated insurance markets.

• Centralize tooling before scaling use cases — a shared AutoML platform lets small teams cover far more ground than siloed, bespoke model pipelines.
• Model governance at scale requires automation — a patented monitoring layer that tracks all production models is not optional at enterprise volumes; it is the infrastructure that makes trust possible.
• AutoML accelerates the prototyping cycle, not just deployment — the largest time savings often come earlier, in feature engineering and experiment iteration, not just final model selection.
• Cross-functional use cases compound value — applying the same platform to fraud, underwriting, churn, and retention creates shared learnings that improve each model over time.