Configurable agentic AI marks a 'paradigm shift' for insurers
Insurers are moving past scattered pilots. The push now is domain-based AI that scales across pricing, underwriting and claims-integrated, measured and governed end to end.
At Earnix's Excelerate 2025, product manager Barak Rozenwax outlined why the company is betting on configurable agentic AI after acquiring Zelros in April 2025. His message was clear: isolated tools won't deliver enterprise value.
From predictive AI to integrated agents
"Predictive AI is the heart. It's everything in our system, it's the core asset, it's essential, but it's only the basis," said Rozenwax.
Early chatbot and LLM experiments showed promise, but they sat outside core workflows. Enterprise proofs of concept helped, yet gaps remained-especially the handoff from building a pricing version to deploying it in production.
What agentic AI means in practice
Earnix is building a platform "factory" for AI agents-configurable, rules-based assistants you can deploy in live environments. These agents coordinate, exchange context and support decisions across the lifecycle.
Crucially, they don't replace human judgment. "It's not making the decisions for you, it's making them with you," Rozenwax noted. The aim: unify the customer experience and scale impact with generative AI as a service.
Why domain-based implementation beats scattered use cases
- Scale: Shared data, models and governance create compounding value across pricing, underwriting and claims.
- Speed: Common tooling reduces rework and accelerates versioning, testing and deployment.
- Control: Central guardrails and auditability keep models compliant and traceable.
- Consistency: One experience for teams and customers, rather than disconnected point solutions.
Practical insurance use cases
- Pricing and rating: Suggest rate plan tweaks, run scenario tests, and guide approvals-then push versions to production with audit trails.
- Underwriting: Pre-fill and validate data, highlight risk drivers, and recommend referral paths with clear rationale.
- Claims: Triage severity, flag potential leakage, and propose next best actions while keeping humans in the loop.
- Distribution: Equip agents/brokers with contextual product insights and compliant cross-sell prompts.
- Portfolio management: Monitor drift, spot segments underperforming target loss ratios, and trigger corrective actions.
- Compliance: Enforce rules, document decisions, and surface explanations for regulators and internal audit.
Implementation checklist for carriers
- Pick the domain first: Start where data and business ownership are strong (e.g., personal auto pricing).
- Define agent roles: What decisions they support, required inputs, allowed actions, and non-negotiable guardrails.
- Integrate deeply: Connect to policy admin, rating engines, data lakes and monitoring from day one.
- Keep humans in control: Approval flows, overrides and clear accountability for outcomes.
- Measure impact: Lead time from model to production, hit rate of recommendations, loss ratio movement, and customer NPS.
- Governance and risk: Bias checks, privacy controls, and post-deployment monitoring aligned to recognized frameworks like the NIST AI Risk Management Framework.
The road ahead for Earnix + Zelros
Earnix is positioning agentic AI as an engine that "will grow faster and smarter, and be much more impactful." The acquisition of Zelros extends its generative capabilities and aims to unify the user experience across the insurance value chain.
The shift isn't about shiny features-it's about integrated, configurable agents working together with your teams in production. For insurers, the winners will be those who scale AI at the domain level with strong governance and clear business ownership.
If you're building internal capability, explore curated AI learning paths by role at Complete AI Training.
Your membership also unlocks:
