With the world currently roaring about the revolutionary potential of generative AIs such as ChatGPT and more recently DeepSeek, it can be easy to get caught up in the wave of enthusiasm without fully understanding what the implications of this exciting new technology are. Likewise, others may find themselves underestimating AI, understandably viewing the hype as overdone and lacking in substance.

The truth is most likely somewhere in the middle – generative AI may not be magic, but neither should it be underestimated, least of all in the insurance industry, where it has the potential to radically improve efficiency and enhance the abilities of human employees to process claims.

Enhanced productivity – with appropriate oversight

For Paul Budd, Partner at SCL Ltd, one major area for productivity gains in the insurance claims experience is document processing through optical character recognition (OCR) technology. 

“Most claims require a minimum of three to four documents that require manual review and submission or processing,” he said. “A well-trained OCR system, at full maturity, can scan and input 50–70% of claims documents with minimal manual intervention. It has the added advantages of greater fraud detection and the elimination of human errors in processing. Combined with better and more consistent customer outcomes, the business case should take care of itself.”

However, Budd noted that while AIs can have a significant positive effect on such processes, they should not be left to operate on their own – rather, they should be viewed “as a complementary tool to human capital”, especially “during these early stages of adoption”.

“After all,” he continued, “large language models (LLMs) rely on humans for training, fine-tuning, and enhancement before they can efficiently replicate historically manual processes across the value chain.”

LLMs are a subtype of generative AI that can generate and interpret text, as well as other tasks that require the ability to process language.

For most firms, the key to success is integrating AI into existing workflows, rather than treating them as standalone tools

James Prebble, CEO of Palladium Digital Group, agreed, noting that “for the most part, travel insurance firms who have successfully deployed AI technology pilots have focused on the repetitive, time-consuming parts of claims management, like sorting through documents, verifying claims, and spotting potential fraud”.

He reported that in a recent project he was involved with, AI tools were being deployed to “extract key details from claim forms, check them against policy terms, and start a claims process”, with the result allowing claims agents to focus on more complex cases “where their expertise really adds value”. 

“For most firms, the key to success is integrating AI into existing workflows, rather than treating them as standalone tools,” said Prebble. The result of this strategy, he argued, was that firms could boost their efficiency without compromising quality
of service.

John Martell, Head Of Digital Solutions at Charles Taylor Assistance, agreed, adding that, ultimately, “harnessing AI has to be about ensuring it is used in areas that will benefit customers most”. 

“We must remember, too, that technology needs to be balanced by niche human skills, to ensure that claims are managed effectively, interventions are made at the right time and customers’ individual needs are met. Human expertise and knowledge are critical in this context,” he said.

Freeing up time for more complex cases and vulnerable customers

Martell explained that if implemented correctly, AI could enable “straight-through claims processing and payments within minutes, with no human touch”.

The result, he said, was that claims surges “no longer need to translate to arduous experiences”, and could potentially allow customers to bypass queues for claims handlers at all hours of the day.

“In turn,” he said, “this is freeing up human expertise for complex case management, vulnerable customers, and those who simply prefer to talk to a human being.”

Martell noted that around 60% of Charles Taylor’s travel claims are now conducted online, with many customers accessing digital options via interactive voice response (IVR) technology. 

Claims surges no longer need to translate to arduous experiences

“AI-supported IVR will triage travel claims notified by phone – separating simple from complex cases, and transferring customers to human experts if claims aren’t suitable for automated processing,” he explained. “If they are suitable, IVR tools will either send an electronic notification of loss (ENL) link to the mobile number from which the customer is calling, or automatically request a mobile number to which this link can be sent if a call from a landline is detected.

“Once an ENL link has been submitted, and supporting evidence uploaded, the automated fraud detection and decision tools will trigger either immediate automated payment, automated repudiation or an automated request for more information within minutes, with zero human touch. Meanwhile, AI-enabled tools will send customers updates on claims progress in real time.”

Poor quality data – and managing expectations

Budd noted that two of the most significant hurdles for insurers looking to implement AI solutions are “legacy IT and poor quality data”, both of which inhibit wider adoption because they are not easily understood and utilised by the AI.

Beyond this, however, he suggested that the “expectation management of executives and the board” was a key consideration – particularly when it comes to how quickly
material improvements to process efficiencies, cost reduction
and customer experience can be delivered at scale.

“One workaround we see companies with limited maturity and resources adopting is to establish AI ‘sandboxes’, isolating data,    people and governance models to expedite adoption and reduce reliance on enterprise IT systems. Whilst this can limit scalability long term, data sandboxes present a viable lower-risk option for businesses behind the vital innovation curve,” said Budd.

Prebble said that one of the biggest challenges when implementing AI systems into insurance is making sure the data used to train AI systems is clean, accurate, and representative. 

“Poor-quality data will lead to mistakes, like rejecting valid claims or approving fraudulent ones. So before any sort of technology is deployed into the business, a thorough data audit is required first,” he explained. Even after this, he added, there needs to be ongoing management and review of the underlying data to ensure that it remains compliant with regulations around data privacy and fairness.

“And of course, there is also bias to consider,” he warned. “If the data AI learns from is biased, the decisions it makes could be too.”

While these risks are worth being aware of, Prebble said that they can be managed by investing in good data practices and regularly auditing the AI systems to catch any issues early. He emphasised that transparency is also extremely important when using AI solutions, and that using AI tools that can explain their decisions can help insurers understand what processes are going on under the surface. This in turn builds trust with system users, regulators, and ultimately customers.

“If there is a degree of proactivity in the management and refinement of the AI tooling being deployed, firms can avoid the majority of these pitfalls and make AI a reliable part of their operations,” he concluded.

Martell added that when looking to avoid potential pitfalls during their AI journey, insurers need to evaluate what their priorities are. He recommended that, when implementing the systems, “working with a provider who can demonstrate a good track record of deploying AI in the travel claims sector can help ensure progress is made in line with an insurer’s needs, and that compliant data use is underpinned by rigorous checks and governance”.

Hesitancy towards adoption

When considering why some insurance firms may be hesitant to deploy AI-driven solutions, Budd said that a lack of experience is often a cause, as well as a lack of funds to acquire it. 

“Great talent comes at a premium, and outsourcing can be prohibitively expensive if businesses are not simultaneously adapting the operating model to build their own capability in house,” he said. “Cost pressures are another: AI at scale can shake the enterprise technology systems to the core, [and] demand significant operating model and IT architectural upgrades, which can render recent investments as virtually sunk cost.”

Prebble agreed, adding that a lack of experience can often accentuate a “fear of the unknown” for many firms. 

“There is already enough complexity in the existing quote, bind and claims processes for insurers – adding additional technology into the mix often creates hesitancy, especially with something that is still relatively new and reasonably complex to understand,” he said. 

“I think most firms worry about what happens if the AI makes a mistake, especially in customer-facing areas like claims.
There’s also the cost and complexity of implementing AI – many insurers are working with legacy systems that aren’t exactly AI-friendly, and often it is difficult enough just to change a question in the quote process, let alone add a new technology to take care of an entire process.”

Working with a provider who can demonstrate a good track record of deploying AI in the travel claims sector can help ensure progress is made in line with an insurer’s needs

On top of this, firms may face resistance from employees if they feel that new AI tools may threaten their jobs or add to their workload – something Prebble said he had “seen first-hand” with some recent AI implementations he has been involved in. 

Budd added that firms looking to supply AI solutions to insurance firms should recognise these reservations, and should be pragmatic in what they offer, not “overreaching with promises of game-changing results in unrealistic time frames”.

Additionally, they should recognise that “the impact on the customer should be the primary reason to adopt AI, not exclusively cost-cutting,” he said, adding that cost-cutting for its own sake has never been a viable long-term strategy in insurance.

Prebble added that alongside the provision of training and ongoing support to insurers, being able to demonstrate the impact of the technology is also key: “AI providers need to show, not just tell, insurers how their tools can make a difference.”

He continued: “The usual case studies and real-world examples are incredibly powerful, but in my opinion there is no substitute for real deployments. Building robust business cases and ensuring AI deployments are low-risk, rapidly executed and designed to prove or disprove the perceived benefit as quickly as possible is the best way for firms to experiment with AI. 

“If a pilot project can provide real data points to support something like a reduction in claims processing times or saved money by catching fraud, it becomes much easier to get buy-in and then scale the solution.” 

Conclusion

AI has the potential to be hugely useful in accelerating the insurance claims process – but, like any new technology, it has to be fully understood and implemented with care to reach its full potential. Insurance firms should ensure that the data they produce is usable, and that they do not view AI as a way to replace humans. Instead, it should be treated as a tool to enhance human capabilities. 

Caution is admirable – but firms should translate that into a considered approach to implementing AI into work processes, rather than allowing the potential benefits to pass them by.
Pilot projects and taking a realistic attitude based on developing expertise should be key, while AI developers should also do their part in ensuring that clients are reassured, but not over-promised.