Digitising Risk with Agentic AI Could Expand the Commercial Insurance Industry
Richard Hartley, CEO of the risk digitisation platform Cytora, outlines a clear vision for how artificial intelligence (AI) can transform commercial insurance. His goal is to build a platform that fully digitises and understands all the information that brokers, insurers, and reinsurers receive in their day-to-day operations.
Rather than just making processes leaner, Hartley believes AI will help the insurance industry grow substantially. By cutting out manual tasks, accelerating data processing, automating claims, and allowing staff to focus on higher-value work, the industry could cover up to 70% of currently uninsured economic losses worldwide.
“Today, around 70% of economic losses are uninsured,” Hartley says. “Many businesses retain significant risk themselves. With increasing volatility from climate change, the insurance industry needs to evolve to absorb and transfer that risk effectively over the next decades.”
How Large Language Models Help Digitise Risk
Cytora employs large language models (LLMs) to process raw data, offering both flexibility and accuracy. Hartley explains that companies define their target schema — the specific data fields they want digitised — and the AI transforms incoming submissions into that structured view.
“Brokers send submissions in various formats, but you can author your own view of risk,” he notes. “Agentic AI uses multiple collaborative agents to perform different tasks, overseen by a manager agent ensuring quality and consistency.”
For example, if a loss run contains 100,000 claims, individual agents can extract each claim while a manager agent reviews the overall dataset for accuracy. This approach aligns with UK government guidance on agentic AI, which describes such systems as composed of autonomous agents that interact to meet objectives.
Ensuring Accuracy Through Multiple AI Agents
Hartley highlights that several AI agents work on the same digitisation task simultaneously, each with slightly different settings. This parallel approach allows the system to compare results and measure confidence.
- If all agents agree on a data point, such as the total number of employees, confidence in that data is very high.
- If one agent's output differs, confidence is lowered and a human review can be triggered.
This multi-agent setup provides a balance between automation and quality control, reducing errors while maintaining trust in AI outputs.
Why Insurance Is Well Suited for AI Adoption
The insurance sector handles vast amounts of data, often in unstructured formats like emails, documents, and submissions. Tasks such as claims handling are repetitive and prone to delays, creating clear opportunities for AI-driven improvements.
Data from the Financial Ombudsman Service shows that nearly 30,000 claims-related complaints were made in a 12-month period ending September 2024, representing 71% of all insurance complaints. This highlights inefficiencies that AI could help address.
Hartley believes AI will improve claims handling by removing low-skill administrative work from highly skilled professionals. “Claims handlers can focus on decision-making rather than paperwork,” he says.
Looking Ahead: Transparency and Explainability
Agentic AI is advancing toward better self-explanation through “chain of thought reasoning.” This means AI agents can provide detailed, rational explanations for their decisions as they work.
“People often don’t explain their thinking process, but AI can offer much more transparency in how tasks are completed,” Hartley notes. This feature could bring significant benefits to insurance, fostering trust and clearer communication.
For insurance professionals interested in leveraging AI tools and techniques, exploring AI courses tailored to insurance roles can provide practical skills and insights.
Your membership also unlocks:
