Why insurance remains a human business despite the growth of AI

AI in insurance

The insurance industry is no stranger to AI, with machine learning models having been quietly operating behind the scenes in risk assessment, claims processing and operational efficiency for many years.

But the recent emergence of agentic AI – where so-called ‘AI agents’ operate independently, seeking to mimic human decision-making but without constant human oversight – represents a significant shift for our traditionally conservative sector, and one that is creating both excitement and uncertainty for its senior leaders. What are the opportunities to do more and better? Where are the potential pitfalls to avoid? What challenges might be faced along the way?

From theory to practice

As the voice of the UK insurtech industry, and together with our partners NashTech, we set out to find the answers to these questions, and to come up with some practical guidance for businesses considering implementing agentic AI initiatives.

Together, we convened a group of senior leaders from across the insurance value chain – including insurers, MGAs, brokers, insurtechs and technology partners – for a series of discussions that would later feed into a whitepaper on the topic, entitled: ‘Agentic AI in insurance: From theory to practice’. Our working group had a single aim: to explore realistic practical applications of agentic AI in insurance businesses.

Agentic AI will be our best friend if we have the appropriate understanding and controls that don’t allow it to become our worst enemy

Senior leader comment

We wanted to move beyond conceptual discussions and develop safe, effective and value-driven use cases as a result of providing a collaborative space in which these industry experts could validate assumptions, exchange insights and share evidence about what was working – and what was not.

We found a huge amount of interest in agentic AI, but some reticence in scaling solutions within what we know to be a highly regulated environment driven by customer trust.

Most initiatives to date had been confined to controlled pilots with an internal focus, and executives cited fears about accountability, governance, explainability and operational risk as creating a barrier to scaling emerging technologies across key, customer-facing operations.

Executives wanted to use agentic AI, believing in its value to drive operational and process efficiency, but were wary of the reputational risks it could bring.

As one senior leader put it: “Agentic AI will be our best friend if we have the appropriate understanding and controls that don’t allow it to become our worst enemy.”

Insurance is still a human business

Participants discussed the importance of insurance continuing to be ‘a human business’ and were clear that, just because AI could do something, didn’t necessarily mean it should.

They agreed the business case for implementing AI initiatives should demonstrate value far beyond simply a cost saving, citing the need to demonstrate alignment with broader strategic goals and detail compelling business reasons for change.

They talked about the need for good quality, consistent data to ‘feed the machine’, and for clarification on regulatory expectations, accountability and explainability requirements.

Overall, there was a feeling that humans would remain accountable for AI agent decisions, and that these decisions should be open to audit and being overridden, with responsibility for cases able to be escalated to human handlers.

As Charlotte Gregory, Partner at Capital Law, said: “Ultimate responsibility must remain with a human. This is not optional; it is a foundational principle that regulators expect.”

Adopting agentic AI

It was clear from those NashTech and Insurtech UK spoke to that agentic AI is the future, and that the appetite for wider adoption is there.

But its successful implementation in our industry requires much more than just following the hype.

Instead, it requires a pragmatic, collaborative and governance-first commitment from the industry to getting the basics right, accompanied by each individual organisation following a methodical process of organisational learning, strategic alignment, and disciplined execution.

Only by doing these things will agentic AI truly move beyond distant, futuristic concept to achieve that ‘best friend’ potential.