Richard Hartley: Digitising risk with agentic AI can expand the insurance industry

Richard Hartley, chief executive at risk digitisation platform Cytora, characterises his firm’s core objective as building a platform that “collects, understands and digitises everything that a broker, insurer or reinsurer would receive” in terms of information.

Using artificial intelligence (AI) to achieve this goal presents a path not purely to a leaner, more streamlined industry, but to a bigger one, Hartley tells Insurance Times.

He predicts that by eliminating manual tasks, speeding up data processing, automating claims handling and freeing up humans for more focused work, a subsequently revitalised and more efficient insurance industry could expand to cover around 70% of the world’s current uninsured economic losses.

“Our mission is to enable a far larger commercial insurance industry, where higher volumes of risk can be absorbed by brokers, insurers and reinsurers,” Hartley explains.

“[Around] 70% of economic losses today are uninsured and a lot of that risk is retained by businesses of all sizes.

”We envision in this higher volatility world of climate change, in the next 10, 50 or 100 years, you need to build a version of the insurance industry that can absorb that risk and enable that to be transferred.”

Large language models

Like an increasing number of insurtechs, Cytora uses large language models (LLMs) to process raw data – a method which Hartley says allows for both freedom and accuracy.

He explains: “We use LLMs and that means you only define your target schema – what are the data fields that I want to digitise?

“Each company is authoring their view of risk. Ultimately, you’re not in control of what you receive. A broker is going to send you a submission in whatever format they like, but you can control and author your view of risk and turn it into the view that you want.

“The way we use agentic AI is that the transformation of what you receive into your target view of risk is done by multiple agents that are all collaborating together.

“It means you can get to very high levels of performance because different agents can do different things and they can be supervised by a manager agent that makes sure the individual agentic tasks are done appropriately.

“For example, imagine you had a loss run that has 100,000 claims – you could have individual agents extracting each claim and then a manager agent looking at it in totality and saying ‘does this all look good?’”

According to guidance published by the UK government in June 2025, agentic AI ”refers to AI systems composed of agents that can behave and interact autonomously in order to achieve their objectives”.

An ideal industry

To ensure AI agents are providing a balanced view, Hartley describes how different models with different tunings work together.

“We ask every digitisation task to be done by seven different agents in parallel,” he says. “Each agent will have a different level of what’s called temperature, so they’ll take a slightly different path to solve that digitisation task.

“That means if all agents come back with the same field – for example, [the] total number of employees being 120 – we can be extremely confident that that is the correct value. If one agent comes back with a different value, that allows us to downgrade the confidence of that field and trigger, for example, a human review.”

Many commentators consider the insurance industry to be an ideal use case for artificial intelligence technology thanks to its large data sets, often delivered in unstructured formats, that are commonly used for day to day operations. Moreover, processes such as claims handling are widely considered to be plagued by slow response times and repetitive manual tasks.

This challenge is demonstrated by analysis of Financial Ombudsman Service (FOS) complaints data by market intelligence firm Insurance DataLab, conducted in January 2025 and published exclusively by Insurance Times. This found that almost 30,000 claims related complaints were referred to the ombudsman in the 12 months to the end of September 2024, equating to 71% of all insurance complaints over this period.

Hartley believes that while removing manual tasks will change the role of claims handlers, it will be an “improvement to the role” rather than having “extremely high skilled people making risk decisions [and] doing extremely low skilled admin jobs at the same time”.

Looking to the future, Hartley is excited for agentic AI to improve – and develop the ability to explain itself.

He notes: “We call it chain of thought reasoning. This means you can get to this amazingly comprehensive and rational explanation exactly as an agent is executing.

“When you compare that to how people are, people don’t often go around explaining what they’re doing. So, there’s something very beautiful – and it’s a huge opportunity for insurance – about having much more transparency about how tasks are being executed and completed.”