AI plus human brings trust at scale in claims decisioning
If you are still asking your clients to wait five, seven or 14 days before you respond to their claim or query, then you are doing something wrong. One of the craziest tensions that blights our operating environment as we push into the AI era is the mismatch between what digitalisation and technology allows us to do versus what we have become accustomed to.
The train trip to a cancellation surprise
The above tension was flagged during a fascinating presentation on the integration of AI into the non-life insurance claims process. Reuben John, co-founder and CEO at Swiss insurtech True Aim AG, who took to the virtual stage at the popular Insure Talk 63 event, started his talk with a personal account of a train trip between Zurich and Geneva. Well, a train trip that set off from Zurich for Geneva, but was cancelled shortly after departing due to issues on the track.
Realising that he had purchased travel insurance alongside the ticket, John grabbed his smartphone, logged into the insurance app and lodged a claim within seconds. And it took another minute or so before he heard his phone’s familiar notification tone. Looking down, he expected to see a push notification from his bank saying that the ticket price had been refunded; but to his surprise, it was one of those auto email responses that thanked him for his claim, and promised a decision within 14 days. “How is it that with all of the information, the technology and the ecosystem we have, we are not able to process claims in a live and dynamic manner,” John asked.
The presenter argued that in the context of AI, digitalisation and other technologies available today, insurers should be able to offer near real-time settlements for simple claims such as this. He said the future of claims settlement was here, and that insurers were already using solutions that can onboard a claim; read and extract all of the relevant data from the claim; contextualise the data; run a fraud scan and produce fraud intelligence if needed; proceed with complex decisioning on the claim; and make payment on it within minutes. For extra applause, AI-backed processes can produce audit trails or decision dossiers that stand up to regulatory scrutiny.
AI and complex decisioning engines
True Aim was self-described as “sitting at the cutting edge of what is possible when it comes to AI and complex decisioning engines … that are built for regulated environments.” In practice, a claims assessor at a broker or insurer can be ‘joined’ by a claim pilot that functions in much the same way as ChatGPT or Claude. This claim pilot “sits right next to your claims assessors, helping them to contextualise every single claim, recording every action that is taken, and acting as a sparring partner as they work through claims,” John said.
The audience was challenged to expand their thinking on practical use cases for emerging technologies within their administration, advising, claims and underwriting functions. And then the impact of technology in claims assessing was unpacked through a ‘then and now’ explainer. In the legacy system, a claim is submitted electronically and the various documents indexed and collated in a file for human review. The human would have to weigh this information against the insurance contract and come to a determination on how coverage is provided.
It sounds simple enough, but challenges creep in for complex cases, with additional thought applied to excesses, deductions and liability. “In complex insurance policies, there are sometimes more than 10 000 different routes to a simple yes or no,” John said. “The logic tree is incredibly hard to model, and even harder for a human to think about.” Technology was offered as a sensible way to ensure accuracy and consistency in claims decision making. Over time, the industry will migrate from today’s largely manual process, where some inconsistencies creep in, to one where all decisions are as intended.
Central intelligence agents for brokers, insurers
The broad question explored through the presentation was how to create a central intelligence for a brokerage or insurance firm that captures legacy knowledge, ensures greater consistency and speed in decision making, complies with data protection legislation, and provides an audit trail to keep the conduct authority happy. “We now have a shift towards automated decision-making and accountability,” John said, adding that it was possible to complete end-to-end claims processing, from first notification to payout, in under two minutes.
This efficiency depends on the overlap of countless established and emerging technologies. For example, automated document ingestion and extraction using a combination of vision models and large language models; real-time policy coverage checks using traditional automation; and more complex decisioning using generative AI. It is now possible “to fully reconstruct every single decision and show how it was made in a tamper-proof way.” To achieve this, users must understand the difference between ‘black box’ AI, where you cannot explain what has happened between prompt and output, and governed AI that introduces explainability.
Governed AI is core to enabling brokers or insurers to deploy modern technology solutions that, at the extreme, might ‘employ’ thousands of unique AI agents to solve a single complex problem. The idea, according to your writer’s hazy recollection of decision trees and other mathematical dark arts, is to test thousands of possible routes before settling on the most consistent and defensible answer. John explained it better, saying that these simulations allowed generated answers to be “more probabilistically correct.” Adding traditional deterministic trees and an audit trail to the mix ensures “a really solid foundation for making correct decisions when using AI.”
The challenge is to turn this technical knowledge into practical insurance solutions that give clients a seamless process from claim submission to payout. “AI is not always the best way to solve this,” John said. “Sometimes a click-through process … is more efficient than engaging with an AI.” This resonates with what your writer has heard elsewhere, in that you need to weigh up the trade-offs between traditional process automation and AI, in the context of cost and customer experience etc.
At the human-AI horizon
John spent some time on the intersection between AI systems and humans, which is as good a place as any to wrap this article. “The AI agents can go off and do their thing, but they have very specific guardrails within which they need to work to ensure that they are deployed safely and properly,” he said, coining the phrase ‘guarded autonomy’. AI does the repetitive admin and low-risk work, including audit trails and compliance whereas humans bring a soft touch for ambiguous situations, escalations and the subjective components of certain claims.
The presenter said that humans, and the insurer, had to handle the edge cases and exceptions in such a way that the customer believed the insurer was there for them. To conclude, he said “the concept of scalable trust is something which really inspires me; I hope that we can move towards this intersection between AI and humans in a way that actually improves the quality of life for many, many people.”
Writer’s thoughts:
The capabilities that AI, automation and machine learning offer brokers and insurers in claims, client onboarding and quoting are simply staggering. Have you explored off-the-shelf agentic AI solutions for your business, or is Copilot on the desktop enough for now? Please comment below, interact with us on X at @fanews_online or email us your thoughts [email protected].
