Building trust in AI, one claims decision at a time

A recent Insurance Post roundtable held in association with Shift Technology explored how insurers are approaching AI in claims with caution and collaboration.

Roundtable participants:

• Steve Parry, group claims director, Hiscox
• Waqar Ahmed, claims chief operating officer, Aviva
• Chris Walsh, commercial claims director, Axa
• Joseph O’Connell, claims director, The Acorn Group
• Elspeth Hackett, claims transformation director, Esure
• Lee Dainty, commercial claims director, RSA
• Nicola Chapman, global head of claims, Rokstone
• Richard Napoli, claims and legal services director, Markel UK
• Eibhlin Swan, director of claims strategy & customer experience, Allianz
• Mark Wood, Walbrook Advisors
• Jeremy Jawish, CEO, Shift Technology
• George Robbins, head of UK, Ire & Nordics, Shift Technology
• Eric Sibony, CSO, Shift Technology
   

This is a matter of identifying where humans add value – and using AI to take care of the rest.

AI as a collaborative partner

As AI finds its place in the insurance claims process, a new ethos is emerging – not of replacement, but of enhancement. The most resonant insights from the roundtable pointed towards a vision where AI acts not as a cold automation tool, but as an intelligent co-worker, helping to elevate the role of human adjusters.

Waqar Ahmed of Aviva argued that AI should enhance – rather than erode – the human strengths of claims handlers. “This technology should be there to enable people to be more effective,” he said. “How do we augment them so they can amplify their own capabilities, primarily around empathy for someone who’s had something awful happen to them?” For him, the role of automation is to support frontline teams by simplifying their work – freeing them to focus on the empathy that customers need most.

This isn’t about racing towards full automation but about building the right hybrid model. As Chris Walsh of Axa put it: “This is a matter of identifying where humans add value – and using AI to take care of the rest.”

Jeremy Jawish of Shift Technology explained that a key benefit of GenAI is supporting junior claims handlers by embedding experienced knowledge into their workflows. “The biggest impact we’re seeing is helping them to leverage this expertise,” he said. By feeding GenAI with example claims from different jurisdictions, firms can provide younger staff with relevant precedents and outcomes. This approach is already delivering strong feedback from clients.

Nicola Chapman at Rokstone said early automation efforts delivered “quick wins” by freeing up time on routine checks. That time, she explained, now goes into the “actual adjusting and knowledge part of it” – where human expertise adds the most value.

Eric Sibony of Shift Technology outlined a maturity model in which AI systems are treated like junior employees who mature over time through training and supervision. “It’s as if you were hiring someone very junior. It is like an apprentice at first. It needs to prove it can be of value,” he said. He laid out five maturity stages for AI in claims – from assistant to collaborator to expert – each with increasing autonomy and impact.
 

It’s as if you were hiring someone very junior. It is like an apprentice at first. It needs to prove it can be of value.

Customer-centric transformation

For all the talk of technology, it was the customer who kept coming back into the frame. Steve Parry of Hiscox described how his team keeps the customer front of mind throughout the design process. “We’ve almost got the customer in the room all the time,” he said, referring to their ongoing work on customer journeys.

This mindset shift requires intentional design choices, especially around when to introduce digital pathways and when to hand the reins back to a human. As Eibhlin Swan of Allianz warned, “At times we’ve been guilty of building with an internal mindset. She said that leads to inefficiencies and frustration. “We need to build up with our customers, not just for them,” she said.
 

It’s about making sure our customers can choose when they want to be on or off that digital journey.

Several leaders underscored the importance of giving customers true flexibility in how they interact. As Parry put it, “Most things I’ve seen in the industry are people forcing you down a path that you don’t want to go down.” Swan agreed, arguing for digital journeys that are opt-in, not mandatory: “It’s about making sure our customers can choose when they want to be on or off that digital journey.”

The pressure to meet customer expectations has never been greater, especially in digitally mature markets like the UK. George Robbins of Shift Technology noted: “Over 90% of traffic on a mobile phone is on an app. That kind of demand for interaction, for smooth processes, is something that you live under.” For him, customer expectations are shaped not by what insurers offer, but by broader digital habits – driving pressure to deliver intuitive, app-like experiences.

Walsh emphasised the strategic value of proactive communication – especially when powered by AI. He pointed to push notifications as a key part of claims engagement: “If we can crack the push notes and be really clear on the push information which is vital, the complex information which needs a discussion or interaction becomes far easier to manage.” For Walsh, these tools help reduce noise and make room for higher-quality interactions.

In this context, AI stops being a process tool and becomes a relational one. It reinforces attentiveness, transparency, and care – qualities every customer values, but especially in the high-stress environment of a claim.

Controlled and ethical deployment

If AI is a powerful tool, it’s also one that must be handled with care. The discussion around deployment made it clear that ethics, control, and governance aren’t optional – they’re essential.

Robbins cautioned that the insurance industry must move proactively on AI governance. “At some point you are going to get asked that question; regulators are going to start wanting to understand and regulate the models and the decisions you’re making off the back of this,” he warned. Rather than waiting for legislation to catch up, he encouraged firms to “start thinking about that in a structured way.”

That proactive stance was echoed by Ahmed, who explained that Aviva has established an internal ethics committee to evaluate the impact of its AI initiatives – on customers, brokers, and shareholders. “I started to ground myself on: if I had to explain this to my mother, would she agree with it?” he said. That is his way of ensuring decisions remain relatable and defensible.
 

At some point you are going to get asked that question; regulators are going to start wanting to understand and regulate the models and the decisions you're making off the back of this.

Controlled deployment is also about being honest about limitations. Chapman shared her experience of early testing of GenAI on claims documents. The system confidently cited an email as the source of a date of loss – only for her team to discover that the email never existed. “It just made it up,” she recalled. That moment prompted a reassessment: “I would need a 99.9% test rate on that to even consider using it on a daily basis.”

Sibony highlighted the tension between speed and supervision. “It’s actually very easy to automate claims, but it’s very hard to automate them well,” he cautioned. He said systems should evolve gradually, and under close supervision.

Walsh was equally cautious: “It’s not just a jump straight to it,” he said, describing how his team began with low-value, high-volume claims to build confidence before expanding further.

Caution was the common thread – whether it was about model reliability, the need for human oversight, regulatory pressure, or personal accountability.

The role of staff

Technology transformation is often seen as a top-down initiative, but participants strongly believed that involving staff in the evolution of claims systems was the way forward.

Joseph O’Connell summed it up succinctly: “The biggest fear is replacement of roles,” he said, but also noted that, when done right, AI is about elevating people – not displacing them. “We want long-tenured technical staff to be able to focus on the more interesting work, while repetitive tasks are handled in the background.”

Chapman shared how her team wasn’t just testing the system – they were co-designing it. “They’ve helped to actually design some of the fields, and they know what they want to get from the system for reporting and management information (MI).” Her staff demonstrated their day-to-day workflows, highlighting inefficiencies and delays. As she put it, “They don’t think, ‘This is going to take over my job’ – because they’ve been involved from the start.”

Swan highlighted the power of engagement through simplification. “Who actually likes sitting there inputting data into four different systems? Absolutely nobody,” she said. With AI relieving that burden, staff can “spend time on what matters: looking after our customers, or developing a technical skill set.”

Ahmed emphasised the importance of starting with frontline realities when deploying new tools. “We spent about three months going across all of our various sites to genuinely understand what the operating conditions of a handler are – and actually they’re pretty bad,” he said, citing multiple systems, logins, and double keying. Understanding those pain points shaped how and where new tools were introduced. 

Steve Parry also spoke about a cultural shift that can build during the transformation itself. Some team members were sceptical at first, “arms crossed, saying, ‘it’s not for me’.” But once involved in projects that used data for proactive risk management, their outlook changed: “You could feel the energy in the room.”

In this transformation, engagement isn’t just about acceptance, it’s about ownership.
 

You have to replace legacy, you have to be able to plug it in.

Overcoming barriers

The technical hurdles to deploying AI across the insurance sector are persistent, and often frustrating.

Richard Napoli was direct: “We haven’t got an API for our systems – and guess what? They can’t build an API for our core system.” That one line speaks volumes about the mismatch between modern tech aspirations and the plumbing many insurers are still working with.

Lee Dainty took a pragmatic stance. “You have to replace legacy, you have to be able to plug it in,” he said, emphasising the need for interoperability. For carriers with diverse product lines, a common platform still has value – but the key, he explained, is to “build into it” with point solutions, not build everything from scratch.

Robbins described how his team runs regular assessments of nearly every large language model on the market – recognising that performance is constantly evolving. “Rather than building our capabilities to be dependent on a single technology, we build a harness that is able to swap out components over time,” he said.

Walsh emphasised the need for seamless coordination across the claims journey. “It has to be end to end,” he said. “We are trying to map that through with suppliers, partners.” Even a great experience, he warned, can fall apart if one part of the journey isn’t joined up.

Elspeth Hackett of Esure added that when systems and people aren’t joined up, it’s the customer who feels it. “There’s nothing worse than a customer being passed on and not knowing who they’re speaking to – or worse, the handler not knowing what’s already been said,” she emphasised. “That connection is so important.”

In commercial lines, integration becomes even more complex. Dainty pointed to the broker-client dynamic, where AI needs to be flexible enough to respect nuanced workflows: “There is that conversation going on with brokers around how we enable that for them as well.”

Brokers play a vital role in this ecosystem, and as Napoli noted, that can complicate AI strategy: “Sometimes the broker sends to the policyholder direct and sometimes you’re not allowed to speak to the policyholder at all.” Mark Wood, of Walbrook Advisors noted that these variations mean that AI systems must be designed to accommodate a range of communication models, “especially in complex commercial environments”.

It’s clear that claims transformation can’t happen in isolation – it must work across the value chain.

A modular path to AI maturity

Most participants seem to be building AI maturity step by step. Sibony offered a simple but powerful message: this won’t just go from zero to one. It’s more like a continuous evolution.

Parry also positioned this maturity as part of a broader evolution, not just in tech, but in mindset. “We’re trying to make claims move from being reactionary to proactive,” he said, pointing to data initiatives that use insights to inform customers before problems occur.

Ahmed stressed that vision and practicality must align. “Transformation doesn’t quite work in annual planning cycles,” he observed, noting that success requires some “rewiring in the organisation.”

His point was echoed by Robbins, who observed that while individual use cases create “small progress”, bringing them together at scale is a major challenge, especially given the complexity of claims.

In the end, the consensus wasn’t about a single roadmap, but about staying responsive – building capabilities step by step, aligning technology with organisational change, and being ready to scale what works.

That adaptability also demands a broader lens. As the roundtable made clear, real transformation in claims isn’t about speed for its own sake. It’s about designing systems that work for the people using them – customers, handlers, partners – and adapting those systems with care. That means listening to frontline teams, keeping governance tight, and rolling out AI in ways that feel transparent, practical and trustworthy.

Sponsored content