Why AI adoption is not a technical rollout, but a journey of trust

AI is often seen as the next step in automation for the insurance industry. But that perspective misses the most important point. The biggest challenge is not the technology itself. It is redesigning processes, decision-making, and trust.

That was the central theme of EY's session at Composable Futures 2026. The conclusion was clear. Successful AI implementation does not start with models or agents. It starts with defining what an insurer wants to become over the next five years, and how people, processes, and technology will work together to achieve that vision.

EY demonstrated that insurers with successful AI initiatives begin their transformation differently. They first define long-term business objectives, redesign their processes, and invest heavily in preparing end users. This allows confidence in new ways of working to develop gradually.

From tool to trust: why AI adoption does not happen automatically

Research into consumer attitudes toward AI shows that acceptance does not increase in line with technological progress. At the same time, trust changes significantly once AI moves beyond providing information and begins taking actions autonomously.

“People often say they would rather speak to a person than a chatbot. But when AI proves to be faster, better, or more efficient, many still choose AI.” This is the experience of Anouk Mouthaan, researcher at EY, who studied how consumers respond to agentic AI. This technology uses multiple AI agents that independently execute tasks.

Consumers generally accept AI when it helps them find information. However, trust declines once AI begins making autonomous decisions or performing actions on their behalf.

"Interestingly, it is often Gen Z who expresses the greatest concern about AI making mistakes. Baby boomers, on the other hand, grew up in a world where personal relationships were central, and are primarily concerned about the lack of empathy in AI agents."

The conclusion is not that people resist AI. Rather, trust is highly dependent on context. That context ultimately determines how far AI can be deployed in critical insurance processes such as claims handling.

From research to implementation

What do these findings mean in practice? Loes Andringa, Partner at EY, is currently supporting several insurers in implementing the Novulo platform.

"It is not about asking how AI can automate claims processing. The first question is what you want to achieve. Only then should you determine where AI adds value and, equally important, where it should not be used."

Don't skip strategic questions

Menno Bonninga, AI Lead at EY, believes many organizations start in the wrong place. "Organizations often think about AI strategy far too late. What should our insurer look like in five years? What will our customers expect? What will our business model look like?"

According to Bonninga, these are the questions that must be answered first. From there, organizations can work backwards to determine the technology they need and the capabilities their people must develop.

Only then can organizations create an effective collaboration between people and AI while maintaining trust, transparency, and control.

Designing the future claims process

Andringa explained how EY and Novulo jointly explored what a future claims process could look like. "We mapped the entire process. At the center sits Novulo. Around it is an orchestration layer coordinating a series of AI agents."

These specialized agents can each perform dedicated tasks, including:

• Claims intake
• Registration
• Policy verification
• Data validation
• Fraud detection
• Customer communication
• Claims settlement.

"The objective is not to use as much AI as possible," Andringa explains. "The real question is which AI-supported interactions customers actually feel comfortable with."

Why architecture is becoming the deciding factor

This shift places new demands on the underlying IT architecture. Novulo's central position in the architecture is intentional. Multiple AI agents can only collaborate reliably when underlying processes and data share the same language, definitions, and business context.

A composable platform provides exactly that shared foundation. It allows organizations to introduce, replace, or build AI capabilities without rebuilding their core systems every time. At the same time, every action performed by an AI agent remains fully traceable.

Human and AI: collaboration instead of replacement

According to Bonninga, AI enables organizations to rethink business processes entirely. "How do I coordinate five agents working together? How do I orchestrate them? And where do I keep humans in the loop?" People must always remain in control, while the system needs to be transparent about every step taken by its AI agents.

Andringa agrees.

"You only build trust in an application by demonstrating its value. That takes time. Trust cannot be claimed. It has to be earned."

Start with strategy development

The biggest misconception about AI implementation is that adoption is primarily a technical rollout. In reality, it is a process of building trust. It begins with defining a long-term strategy. Only then should organizations focus on change management, process design, governance, and user experience. Customers are open to interacting with AI, provided they retain control, clearly understand its added value, and know when a human remains involved in the process.

As Andringa summarized: "Trust is not built through promises. It is built through transparency in execution."

‍

Conclusion: AI requires process design, not more tools

The real transformation in insurance does not come from deploying AI itself. It comes from redesigning business processes as integrated systems where people and AI agents work together. Technology only becomes relevant after strategic choices have been made. AI is therefore not the starting point. It is the outcome of a well-designed strategy.