This article was paid for by a contributing third party.
Designing the insurance policy of the future
As AI reshapes commercial insurance, industry experts discuss the role of human judgement, technology and governance in enabling effective underwriting, product development and decision-making.
The panel
- Raj Bosamia, chief information officer, Iprism
- Jon Whitear, speciality sales, EMEA and APAC, Hyland
- Matthew Priestley, director of solution architecture, Europe, Duck Creek Technologies
- Scott McGee, deputy editor, Insurance Post (moderator)
Artificial intelligence may be advancing rapidly across insurance, but the industry is still some distance from handing underwriting decisions wholesale to machines.
That was one of the central messages from a recent Insurance Post webinar, ‘Designing the policy of the future’, held in association with Hyland and Duck Creek Technologies, which examined how AI could reshape commercial insurance and the respective roles of technology, human judgement and governance.
Three levels of AI
For all the attention being paid to agentic AI, much of the industry remains focused on relatively straightforward applications.
Matthew Priestley, director of solution architecture, Europe, at Duck Creek Technologies identified three broad stages of adoption. The first is personal productivity, including summarising submissions, extracting information from documents and querying underwriting guidance.
The second sees AI used across more of the underwriting workflow, sifting submissions, identifying missing information and helping underwriters focus more quickly on the important elements of a risk.
The third is agentic underwriting, where AI takes data and, within defined parameters, reaches decisions and takes actions.
Most of the market is probably somewhere between the first and second stage [of AI adoption].
Matthew Priestley, Duck Creek Technologies
“Most of the market at the moment is probably somewhere between the first and second stage,” Priestley said, adding that most current value still comes from assisting underwriters rather than making decisions for them.
Jon Whitear, speciality sales, EMEA and APAC, at Hyland agreed that insurers are primarily using AI to remove wasted time and tackle tasks that have previously proved difficult to automate, particularly around documentation. The next challenge is to turn individual productivity gains into repeatable, consistent business processes.
Enthusiasm for the technology has also produced plenty of proofs of concept without always demonstrating sustainable business value.
“There’s still a lot of failed POCs out there and a lot of, I call them, nice-to-have projects,” Whitear said. “Does it generate a repeatable motion that’s going to get a benefit or value back for the business? It’s not always seen.”
The panel stressed the importance of identifying specific problems where AI can generate measurable value, rather than pursuing projects simply because the technology is attracting attention.
Building organisational capability
Moving beyond individual experimentation will require insurers to develop AI capability more widely across their organisations.
Priestley expects prompting to become a relatively basic digital skill. Of greater value, he said, will be employees capable of bridging insurance and technology: people who understand underwriting processes well enough to identify a problem and have sufficient knowledge of AI to see how it could be approached differently.
Leadership has a role in providing the tools and setting appropriate guardrails, while employees need the opportunity to experiment with the technology.
Some organisations may already have more AI expertise than they realise, Priestley suggested, with individuals developing their own prompts, automations and applications. Internal hackathons and similar initiatives could help uncover that expertise and turn “individual adoption into organisational capability”.
Raj Bosamia, chief information officer at Iprism said AI-focused roles are also beginning to emerge to translate technical capability into underwriting and business workflows. With AI increasingly reaching the boardroom, deployment is becoming a strategic decision requiring senior sponsorship, investment and a clear view of return on investment.
Looking further ahead, he envisaged underwriting moving from “processing risk” towards “curating it”, with automation supporting faster and better-quality decisions.
Human ‘on the loop’
Greater autonomy will also change the way human oversight works.
Bosamia described a potential agentic ecosystem comprising several specialised components: one generating or processing a quote, another checking outputs and errors, and another challenging the decision before a managing component determines whether human intervention is required.
At sufficient scale, he said, the familiar concept of a human “in the loop” may have to evolve.
“You’re going to have to have a human on the loop rather than in the loop because you’re not physically going to be able to look at everything,” he said.
Underwriting expertise would remain central to such a model. Bosamia identified judgement, accountability and the ability to recognise edge cases as areas where people continue to add particular value. AI can provide consistency at scale, but can struggle to recognise when a risk is genuinely novel or when something does not look right.
You’re going to have to have a human on the loop rather than in the loop because you’re not physically going to be able to look at everything.
Raj Bosamia, Iprism
“The human role will shift from processing to challenging” AI outputs, he said. Communication and customer relationships would remain important too, particularly where customers want to discuss their risks rather than deal solely with an automated process.
Priestley also pinpointed empathy, particularly during claims, as an area where insurers should be cautious about replacing human interaction. “That’s fundamentally when a customer makes their judgement on you as a company,” he said.
Automation raises a further question around how tomorrow’s underwriters acquire experience if AI handles much of the work through which previous generations learned their trade.
Underwriting knowledge has traditionally developed through reviewing large numbers of risks, understanding referrals and seeing what happens afterwards. If more of those tasks move to AI, insurers may need to rethink training. Priestley suggested greater use of mentoring and simulations, as well as asking junior underwriters to assess risks before seeing the machine recommendation.
Without that investment, he warned, automation could widen the industry’s skills gap.
Context is critical
The quality of AI decisions will depend heavily on the information and rules underpinning them.
A general-purpose language model may be highly capable, Priestley noted, but it does not inherently understand an individual insurer’s business. It needs access to the appropriate products, risk appetite, underwriting guidance, authority limits, policy history and organisation-specific rules.
An insurer does not necessarily have to cleanse its entire data estate before starting to use AI. For each decision, however, it needs confidence that the relevant data is authoritative, current and traceable.
Priestley expects this to lead to hybrid architectures combining probabilistic language models, which are well suited to interpreting documents and other unstructured information, with deterministic models for areas such as rating and underwriting authority.
ChatGPT is so simple and Claude is so simple; actually using AI at scale in business is less simple
Jon Whitear, Hyland
Whitear similarly emphasised the importance of giving AI controlled access to insurers’ own information and “tribal knowledge”, rather than relying indiscriminately on external information.
Data security is a key consideration. Whitear identified inappropriate use as one of the immediate risks, with employees potentially exposing proprietary information by submitting documents to uncontrolled AI tools without appreciating the implications.
Established technology disciplines remain essential, he added. Encryption, data residency and access controls are still required, alongside guardrails governing how AI tools can access and use information.
“ChatGPT is so simple and Claude is so simple; actually using AI at scale in business is less simple,” he said.
Governance cannot play catch-up
Governance will become increasingly important as insurers give AI greater autonomy.
Bosamia said insurers already operate within regulatory frameworks covering areas such as accountability, human oversight and fair outcomes. His concern was that agentic capabilities could outpace organisations’ ability to govern and explain them.
“The risk is scaling the agentic AI faster than the governance can justify it,” he said.
Insurers should therefore build regulatory discipline into AI deployments from the outset, rather than later having to unravel how models reached particular decisions. Doing so, Bosamia suggested, could mean future regulation becomes “much more of a confirmation, not a disruption afterwards”.
Priestley said accountability remains with the insurer regardless of whether a decision is made by a person, supported by AI or made by an AI system.
Firms also need to be able to reconstruct decisions. If challenged two years later about an AI-driven underwriting decision, an insurer should be able to establish the data used and its source, the model and version involved, the underwriting rules and authority applied, and the actions taken by the agent.
“If you can answer those questions and demonstrate that you remained in control of it,” Priestley said, insurers should be in a stronger position to demonstrate effective governance.
The discussion pointed towards a gradual expansion of AI across underwriting, with the pace of adoption determined as much by insurers’ ability to transfer expertise, establish controls and build trust in the technology as by the capabilities of AI itself.
For the foreseeable future, human judgement will remain a significant part of that equation.
View the on-demand webinar: ‘Designing the policy of the future’
Sponsored content
Copyright Infopro Digital Limited. All rights reserved.
As outlined in our terms and conditions, https://www.infopro-digital.com/terms-and-conditions/subscriptions/ (point 2.4), printing is limited to a single copy.
If you would like to purchase additional rights please email info@postonline.co.uk
Copyright Infopro Digital Limited. All rights reserved.
You may share this content using our article tools. As outlined in our terms and conditions, https://www.infopro-digital.com/terms-and-conditions/subscriptions/ (clause 2.4), an Authorised User may only make one copy of the materials for their own personal use. You must also comply with the restrictions in clause 2.5.
If you would like to purchase additional rights please email info@postonline.co.uk