Ronny Reppe, co-founder of Noria Group, explains how AI can finally transform the insurance claims environment.

Slow claims handling is holding back customer-centricity in insurance, causing dissatisfaction and financial stress for policyholders, erosion of reputation and loyalty, increased internal costs and regulatory consequences. Some things will never change. Complex claims involving large losses, multiple parties, disputes, inadequate
underwriting or incomplete and missing documentation will always slow down processes. But, other hurdles are within our power to overcome. Manual, paper-based processes can be slow and error prone. Many insurers simply do not have the people or resources to handle a sudden influx of claims. As a whole, the industry is still moving too slowly towards a digital future.

But in the age of generative AI and large language models (LLMs), the wait for a technological solution that will
dramatically increase the velocity of claims handling is finally over.

Semantics of content

As Forbes Council Member Dr Jeremy Nunn writes: “[AI can improve] effectiveness and efficiency thanks to its ability to understand the semantics of content and automatically acquire knowledge…through intelligent document processing (IDP), the process of data extraction automation from unstructured and semi-structured documents and conversion into structured and usable data, AI has tangibly amplified extracting data with high accuracy.”

Working with a Norwegian client, the team at Noria has created a solution known as the AI-powered virtual assistant for claims handling that is generating impressive efficiency gains, even at the pilot stage. PwC’s research into digital intelligence found AI-based extraction techniques can save up to 40% of hours usually spent on manual processes such as claims handling. We have already seen efficiency gains of up to 50% in this area, but even a gain of 10% creates a very compelling business case and a no-brainer for adoption.

How does it work? Our people receive a lengthy document describing the case they are working with. While it would usually take 15 to 20 minutes to read five pages manually, this time is slashed to under a minute with the virtual assistant.

Dynamic questions

The user selects from a set of dynamic and configurable pre-defined questions, after which the AI generates a series of answers for the user to consider and choose between. We are training the AI to understand the type of data our claims handlers usually extract from documents,
drawing upon Noria’s decades of data history and tapping the experience of our handlers to help the model grow smarter, faster.

The result is speedier claims processing, more accurate decision-making through data-driven insights, satisfied
customers and reduced operational spend due to enhanced efficiency and effectiveness. Crucially, the tool will act as an invaluable extension of the human team to help insurers cope with situations where they would normally be overwhelmed by a high influx of claims.

As with any efficiency gain, the technology will also boost productivity, innovation and staff engagement by freeing up your team to tackle new challenges. This is a new assistant, which means it still needs a skilled and knowledgeable human-in-the-loop to assist with training, sense-checking and understanding if its outputs are correct. We are at the beginning of a training and UX journey that will take some time. But, early results are extremely encouraging and motivating.

Organisations looking to adopt this sort of technology should use caution, and ensure they take the virtual
assistant approach rather than simply handing over a task to untested AI and let it go unmonitored. Results may be hit and miss for some time, while the ratio of right-to-wrong answers will be the main factor in determining the efficiency.

Be curious

At Noria, we encourage our team to be curious and apply the assistant across different areas of the business. Using a similar approach to how we experiment with ChatGPT, this is a technology that anybody can play with and test with new use cases. Alongside claims processing, other applications in marine insurance could include enhanced customer support, risk assessment, policy recommendation, regulatory compliance and loss prevention.

We have also found use cases in programming, where an AI tool has strong potential to help our developers code faster and more accurately. Like the claims processing assistant, the tool would work by suggesting code for auto-completion or generating code based on prompts. McKinsey found that developers can complete coding tasks up to twice as fast with generative AI, particularly when “expediting manual and repetitive work, jump-starting the first draft of a new code, or accelerating updates to existing code”.

For us, a 20% efficiency improvement in programming would be enough to justify an investment, which is why we’ll be looking to build an AI-powered programming assistant pilot within the next few months.

Generalised model

But why not simply use ChatGPT? The issue lies with ChatGPT’s generalised model, which tends to become
weaker and more inaccurate when it is applied to domain- specific use cases.

I was interested to read a press release from SaaS company Simplifai that noted the importance of training LLMs specifically on information directly relevant to the insurance sector. The authors cite concerns around data security, and inaccurate presentation of data meaning public LLMs such as ChatGPT and Google Bard being labelled unsuitable to address the needs of the claims handling industry. In conclusion, my advice for marine insurers is to take the plunge in AI-powered claims handling, but use caution at first by keeping a knowledgeable human in the loop. We have been impressed at Noria by the value and efficiency gains already generated by working with this technology and, have a plan for unlocking further efficiencies across several parts of the business.