Tim Staebler
NTT DATA
Global Data and AI Practice Leader for Insurance
In this written interview with Tim Staebler, Global Data and AI Leader for Insurance at NTT DATA, we explore how Agentic AI is moving beyond experimentation to deliver tangible impact across the insurance value chain.
How would you define the concept of Agentic AI within the insurance sector, and why do you believe it represents a strategic opportunity for insurers today?
I define Agentic AI in insurance as AI systems that can plan, decide (as needed), and act within defined guardrails, orchestrating workflows across people, data, and systems rather than just generating insights or content.
What makes this strategically important now is the convergence of three realities insurers face:
• Operational strain (expense ratios, talent shortages, claims inflation)
• Rising customer expectations for speed and personalization
• System complexity that makes traditional transformation slow and expensive
Agentic AI represents a shift from assistive intelligence to executional intelligence. It allows insurers to move from “AI that recommends” to “AI that resolves,” while still keeping humans in control. That creates measurable outcomes like cycle time reduction, leakage prevention, and scalability that boards and regulators can support.
Quantum readiness: An interview with Romi Sumaria on the future of insurance
Where do you see agentic AI delivering the greatest impact across the insurance value chain—underwriting, claims, customer service, or distribution—and how is Agentic AI designed to unlock that potential?
While Agentic AI can touch every function, I see the greatest near-term impact in claims and operations and underwriting.
• Claims benefit immediately because workflows are rules-heavy, data-rich, and latency-sensitive. Agentic AI can triage claims, request missing information, coordinate vendors, detect fraud signals, and escalate exceptions—end to end.
• Underwriting is next, where agents can dynamically gather third-party data, apply appetite rules, run scenario analysis, and prepare underwriter-ready decisions rather than raw outputs.
• Customer service improves through resolution-based agents—not chatbots that answer questions, but agents that complete tasks like policy changes or FNOL initiation.
• Distribution benefits more selectively, especially in agent-assisted selling and renewals.
Agentic AI unlocks this by combining reasoning, workflow orchestration, and system-level action, rather than operating as a standalone model. The value comes from coordination, not intelligence in isolation.
Many insurers face challenges with legacy IT infrastructure. How does Agentic AI approach integration with existing systems, and what lessons have you learned about balancing innovation with operational continuity?
Legacy infrastructure is not a barrier, it’s a constraint that has to be respected.
Agentic AI works best when treated as an orchestration layer, not a replacement layer. Instead of ripping and replacing core systems, agents:
• Interact through APIs, RPA, event streams, and service layers
• Sit above systems of record, not inside them
• Operate incrementally, starting with low-risk processes
The key lesson is sequencing:
1. First stabilize the process
2. Introduce agentic automation in bounded domains
3. Expand autonomy only after trust is earned
Innovation succeeds when it’s invisible to operations. If frontline teams feel disruption before they feel benefit, adoption stalls. The lesson learned is that alignment matters more than intent. An agent that isn’t aligned with business objectives, compliance constraints, or operational realities will ultimately fail, no matter how capable it is.
Data quality, security, and ethical AI decision-making are critical for insurers. What principles guide Agentic AI in ensuring responsible, compliant, and trustworthy AI deployment?
For insurers, responsible AI isn’t a philosophy it’s a license to operate.
Agentic AI must be guided by five core principles:
1. Human accountability – Agents act autonomously, but accountability always rests with named human owners.
2. Explainability by design – Every decision must be traceable, auditable, and regulator-ready.
3. Least-privilege autonomy – Agents are given only the authority required for their task, nothing more.
4. Data lineage and quality control – Inputs, transformations, and outputs are continuously monitored.
5. Policy-encoded ethics – Fairness, bias thresholds, and compliance rules are embedded directly into agent behavior, not bolted on afterward.
Trust is earned when agents behave consistently under stress, not when they perform well in ideal conditions. Especially when agents interact with systems tied to real financial exposure—or the physical world—predictability becomes more important than sophistication.
AI adoption is still in early stages for many insurers. What strategies do you recommend for scaling Agentic AI across an organization, and how can insurers manage change, culture, and talent development to maximize adoption?
Most insurers don’t fail at AI because of technology they fail because they don’t go deep enough into the the why, what, and how.
Why:
Economic leverage. If there’s no clear value thesis i.e, cost, speed, or revenue, the effort won’t scale.
What:
Outcomes that materially change how work gets done, by an order of magnitude, not just how insights are generated.
How:
Reliable, governed systems embedded into real operations, not side experiments.
Scaling agentic AI requires:
• Starting with a small number of lighthouse use cases with undeniable economics
• Establishing an agent operating model with ownership, governance, and controls
• Developing talent that understands both insurance risk and AI system behavior
Everything else, models, tools, vendors, buzzwords, is temporary. What endures is alignment, reliability, and economic impact.