Insurance firms chasing the financial upside of agentic AI are likely to come up empty-handed if they approach it as a software rollout rather than a rebuild of how institutional expertise gets captured and scaled, according to Dr. Magdalena "Magda" Ramada Sarasola (pictured), senior director and global insurtech innovation leader at WTW's Insurance Consulting and Technology division.
Ramada said the most persistent misconception among insurance executives is mistaking agentic AI for a souped-up chatbot. The bigger issue, she added, is where companies think the value is actually coming from.
She also argued that the design choices around a model matter more than the model itself.
"In practice, orchestration and contextualization beat raw model intelligence every time. The way you structure the agent's access to tools, the order in which it accesses them, the deterministic guardrails you put around it, the way you embed your institutional expertise into the system… that is where the competitive advantage lives,” Ramada said. “Two insurers using the same foundational model can get radically different results depending on how they architect and deploy a solution around it.”
As the insurance industry pushes past the AI pilot stage, Ramada has found measuring return on investment becoming a sharper pain point.
General-purpose AI tools can boost individual productivity, she noted, but those efficiency gains rarely show up in a clean-cut way. "The financial impact comes from vertical, insurance-specific AI systems that are deeply embedded in core workflows, and from the ability to scale and iterate those systems continuously," Ramada said.
Many insurers have proven the technology works in isolated tests. However, converting that into systems that generate compounding returns is "a different and much harder challenge."
Rather than full autonomy, the clearest wins today sit in automation and augmentation. In pricing and underwriting, that means agentic systems handling automated model monitoring. She pointed to portfolio management as another strong fit, particularly for insurers wrestling with segmentation, latent risk clustering, and real-time exposure tracking. Reserving is benefiting too, she said, through agent-assisted documentation and cross-referencing of assumptions, with actuarial judgment still driving the process.
Claims, however, is the rare function edging toward genuine autonomy, though only for standardized, low-complexity risks with clear-cut decision logic, like certain property or motor claims.
"In underwriting, AI agents are already doing powerful things in the intake, assessment and suggestion layers, from supporting the ingestion and structuring of submissions, to enriching with internal and external data, benchmarking risks against portfolio peers, flagging exposure-to-coverage misalignment, and recommending pricing and terms," Ramada said.
The results are quantifiable: Ramada cited one insurer's work with WTW that produced "over $13 million in annual savings on claims costs and 95% First Notice of Loss (FNOL) decision accuracy within ten months."
Ultimately, the toughest obstacle to effective agentic AI deployment remains converting human skill into something a system can use. "An experienced underwriter or claims handler carries decades of pattern recognition, market intuition and contextual judgment that has never been written down, that they themselves struggle to articulate, and that manifests differently across every non-standard case they encounter," Ramada said.
Insurers also need to overhaul how they judge whether these systems are working. Model accuracy alone doesn't cut it, she said, since outcome quality hinges on context and how humans interact with the system.
This is also why the WTW leader is sceptical about AI being used as a reason for workforce reductions. "Regulated decisions need human accountability,” she stressed. “And critically, agents are only going to be as good as the expertise of the people working with them. If you hollow out your expert base, you hollow out your AI.”
Looking ahead, Ramada said "the successful agentic insurer will have made a genuine cultural shift from consuming technology to shaping it."
