Magda Ramada: Why AI Pilots Stall, and What Scales

From Scouting for Growth · 21:56

Show by Sabine VanderLinden

More AI pilots will not fix an operating model that cannot scale them. In this episode of Scouting for Growth, I sit down with Magda Ramada, Senior Managing Director and Global InsurTech Innovation Leader at WTW, for an impromptu conversation recorded on the floor of ITC Vegas. Coming directly after our AI Horizon 2030: The Readiness Problem summit, we explore why so many promising AI experiments in insurance still struggle to move into production—and why the industry's familiar problem of pilotitis is becoming increasingly difficult to ignore. The uncomfortable truth is that AI has not created the pilot problem. It has made the transformation required to overcome it more demanding. Horizontal productivity tools may help people write emails or manage general tasks, but insurance-specific AI must understand the complexities of underwriting, claims, pricing, and capital modelling. Agents need to interact with specialized insurance systems, and organizations need to redesign how people and technology work together. Another proof of concept will not solve an operating model that remains unchanged. Magda identifies three constraints that leaders consistently underestimate: people, integration, and data. Skilled professionals must help shape the systems they use, rather than simply being asked to adopt them. Agents need to connect with the insurance software that supports real business decisions. And the data required for agentic workflows reaches beyond structured databases into what Magda calls golden data sets: evolving, machine-readable representations of expert knowledge, reasoning, decisions, and exceptions. These must be continuously updated as experts encounter new cases, identify patterns, and refine their conclusions. This becomes especially important as experienced underwriters and actuaries retire. Scaling expertise is not about replacing the people who possess it. It is about capturing how they think, what they question, and how they make decisions so that intelligence can be shared more effectively across the organization. Done well, AI becomes a partner that challenges assumptions, exposes blind spots, and helps experts identify mistakes. The objective is not to eliminate human judgement, but to build the trust required for people and agents to collaborate—and to increase autonomy only where it has been earned. We also explore how agentic distribution could create new opportunities for growth, as personal AI assistants increasingly become the first point of contact between customers and insurers. But a faster quotation is not necessarily a better one. Magda offers a simple example: an insurer might require answers to five specific questions before providing a quotation, while a customer's personal AI assistant may not know which information matters. Insurers will need systems that validate incoming information, ask the right questions, and protect underwriting appetite and pricing strategies from being reverse-engineered by automated agents operating at scale. Most importantly, Magda leaves insurance leaders with a practical agenda for Q1 2027: value existing experts and develop future talent, improve connectivity across insurance systems, and understand the architecture, controls, harnesses, and evaluation layers required to scale agentic AI responsibly. Because when intelligence becomes increasingly accessible, the advantage will not come from having more pilots. It will come from building the organizational capability to turn expertise, technology, and trust into repeatable execution—and sustainable growth. KEY TAKEAWAYS I want to challenge the assumption that launching another AI pilot brings an organization meaningfully closer to becoming AI-ready. Magda reminds us that insurance has been collecting shiny technology experiments for years. Agentic AI makes this familiar problem more complex because deploying an intelligent system is not the same as integrating it into a business process that people understand, trust, and can improve. Horizontal productivity gains have their place, but the more important question is whether AI can perform insurance-specific work within underwriting, claims, pricing, and distribution. Scaling AI is fundamentally an operating-model challenge, one that requires organizations to rethink workflows, responsibilities, and human–agent collaboration rather than simply add another technology layer. This is why our conversation focuses on people, integration, and data as an interconnected challenge. An insurer needs the knowledge of its best professionals, agents that can interact with specialized applications, and a sustainable way to extract and update the expertise that makes its decisions valuable. The golden data set is not a static repository of documents or historical information. It is a living representation of human judgement, capturing the questions experts ask, the reasoning behind their decisions, the exceptions they recognize, and the order in which work…

More episodes of Scouting for Growth

More episodes

PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.