Creating a standard governance framework for the use of artificial intelligence (AI) systems in the protection market would save money and improve regulatory compliance.
It would also help to address issues of technology working together for the better of consumers and firms, according to Quietroom director Rhys Williams.
Furthermore, he added that underwriting teams were embracing AI but consumers expected to get the same results from an AI process as from a human one.
‘Questions a shared framework would answer’
Williams was speaking at the Income Protection Task Force (IPTF) mid-year meeting about Quietroom’s research into the use of AI and new technologies within the sector.
One of the key findings was the potential benefits of a centralised framework for the use of AI.
“There isn’t a standard governance framework for AI in protection, everyone’s building their own – that’s a lot of duplication for an industry where the use cases are probably very, very similar,” Williams said.
“The cost of each business building its own framework will eventually exceed the cost of agreeing one between us and there are questions being asked by the regulator that a shared framework would answer.”
“There’s also the question of what happens if AI agents begin transacting on consumers’ behalf?” he continued.
“If firms’ AI systems don’t plug in, don’t work together in recognising each other’s outputs, exchanging information and acting on each other’s decisions, that fragmented governance could become a barrier to the services we can offer.”
Underlying this approach, Williams added that clarifying exactly what AI meant for the market would be beneficial.
“Perhaps what we need is a taxonomy, a framework that captures exactly what kind of AI we mean that helps a lay audience or a regulatory audience understand what we mean,” he added.
‘You want a pilot in the cockpit’
In terms of use cases, protection appears to be following a similar path to other industries in how new technologies are being applied.
Underwriting has typically been one of the faster adopters of the new technologies given its processing of large amounts of data and the aim to reduce manual processes.
Indeed, Williams agreed this was one area flagged as the “biggest opportunity for innovation”, although humans are set to remain key parts of the process.
“No-one we spoke to was too up for outsourcing the whole process to AI; the biggest obstacle there is the need to have faith in the decisions made,” he said.
“They compared it to a pilot landing a plane in zero visibility – they may have to trust their instruments, but you want a pilot in the cockpit.”
Williams emphasised that consumers “have every right to expect the same outcome from an online tool as a human underwriter”.
“Trying to get those two things to be equivalent will take time and will be a challenge but it’s a challenge that one underwriter we spoke to really welcomed.
“The reaction from their team had not been fear or suspicion but rather relief that they can go on and do the job they meant to,” he added.
‘Free, fast and good enough to use’
Expanding on how consumers were using these new technologies, Williams said the adoption of AI assistance was at an “extraordinarily rapid rate”.
“It’s faster than mobile phone, social media, broadband internet,” he said.
“They are using AI assistance to find information, to draft complaints, even to challenge underwriting decisions, and they may soon be asking AI to transact on their behalf.
“It’s not that consumers have suddenly become more sophisticated, it’s that the tools have become free, fast and good enough to use.”
However, there was one significant cautionary note highlighted.
“One word of warning, our wider research not in protection but more widely, has found AI tools like Chat GPT and Google’s AI overview routinely give answers that are slightly or majorly misleading; most of the fault lies not with the tools but with the underlying content it is drawing from,” Williams said.
But he emphasised that those intermediaries and insurers who get it right will enjoy greater visibility.