Zurich taps Domino's and Nestlé veterans to hardwire AI into insurance
Insurers are closing their AI gap by hiring outside the sector. Zurich Insurance Group has brought in Terry Powell as Group CTO and Cristina Ghetti as Group Head of Digital Employee Experience to accelerate enterprise AI adoption.
Both start on 20 October and report to Ericson Chan, Group Chief Information & Digital Officer. As Chan puts it, "Their leadership in AI across retail and fintech is exactly the expertise we need to accelerate our digital transformation and drive adoption."
Why these hires matter
Terry led AI-led customer initiatives at Domino's Pizza and ran technology transformation at ANZ Bank. Running ML systems that predict delivery times and orchestrate kitchen workflows at national scale forces rigor in latency, reliability, and cost-all essential for production-grade underwriting and claims platforms.
Cristina managed large-scale AI rollouts at Nestlé, including training and change management across divisions for supply chain and demand forecasting. Her remit at Zurich: make AI usable, trusted, and embedded in everyday work so adoption actually sticks.
From prototypes to production AI
Zurich's internal "Agentic AI Hyper Challenge" produced 200+ prototypes-moving beyond brittle rules-based software. The next step is turning the best ideas into governed, measurable services customers and employees use daily.
LLMs are already sifting underwriting files to extract signals from medical records and property data. The same tooling screens claims for fraud patterns by comparing new submissions with years of historical data.
AI is also reshaping risk assessment. Zurich and peers are modeling liability for autonomous vehicles, cyber exposure in connected cars, and usage-based policies that price behavior with real-time telematics.
What this signals to insurance leaders
- Borrow proven playbooks: Retail and CPG cracked scaled AI operations first. Reuse their patterns for data pipelines, MLOps, and reliability.
- Adoption is the choke point: Tooling is easy. Changing workflows, incentives, and skills is hard. Give someone ownership, like Zurich did with digital employee experience.
- AI belongs in the core: Treat models as shared services in your tech platform, not side projects.
- Real-time matters: Pricing, fraud, and telematics demand low-latency inference and event-driven architectures.
Execution playbook: first 90 days
- Pick three high-ROI use cases: underwriting triage, claims fraud scoring, and contact center assist. Define success metrics and SLAs upfront.
- Stand up a governed data layer: quality, lineage, access controls, and PII handling. Production AI starts with production data.
- Operationalize MLOps: model registry, evaluation gates, canary deploys, automated rollback, and continuous monitoring for drift and bias.
- Instrument adoption: track feature usage, time saved, decision quality, and exception rates. Celebrate wins; fix friction quickly.
- Create human-in-the-loop checkpoints: underwriter and adjuster overrides, feedback capture, and periodic model review.
- Cost control: set per-request budgets, batch where possible, cache results, and right-size models to the task.
Guardrails that keep you safe
- Compliance by design: document data sources, model purpose, limitations, and evaluation results for audit readiness.
- Fairness and explainability: use interpretable features where required; provide rationale summaries to underwriters and customers.
- Security first: isolate model endpoints, redact sensitive data, and monitor for prompt injection or data exfiltration in LLM use.
People and skills: the adoption multiplier
Cristina's focus highlights the real unlock: training and workflow fit. AI sticks when teams know how and when to use it-and what happens if they don't.
- Role-based training: build targeted paths for underwriters, claims handlers, and frontline agents with live scenarios.
- Embedded enablement: in-product tips, office hours, and a fast feedback loop to improve prompts, policies, and UX.
- Change incentives: align performance metrics with AI-augmented outcomes (quality, speed, customer satisfaction).
If you need structured upskilling for insurance roles, explore curated learning paths by role at Complete AI Training.
The bottom line
Zurich is moving AI from experiments to the operating model by pairing deep tech leadership with enterprise-wide adoption. The play is clear: build reliable AI services on top of governed data, measure outcomes, and make it effortless for people to use them.
Do that, and AI stops being a cost center project. It becomes how you underwrite better risk, settle cleaner claims, and win customers faster.
Your membership also unlocks:
