AI in insurance: a guide to maximising efficiency for a better customer experience

There’s a lot of chatter about AI in the insurance industry and its potential to have a huge impact on how insurance companies operate and the experience they provide for end customers.

But AI is set to be a years-long transition. So where do companies start right now? 

According to a recent EY survey, 99% of insurance players are already investing in Generative AI (GenAI) or are interested in investing due to expectations for productivity, cost and revenue benefits.

From deciding what problem you want to solve down to whether to build your own AI model or use a specialised tool, companies are now in uncharted territory when it comes to kick starting their AI strategies.

Here are my thoughts on how insurtechs and insurance companies can approach integrating AI into their strategies – and why improving operational efficiency through customer care and claims is a good place to start.

First thing’s first: vertical or horizontal AI in insurance?

With giants like OpenAI and Google building horizontal AI models like ChatGPT and Gemini – which are becoming exponentially more powerful – companies can easily use these solutions for a variety of AI use cases.

For example, an insurance company could use ChatGPT to train its own AI customer chatbot. 

However, I’m not convinced that starting out with horizontal AI is the smartest way to start out for complex cases like a chat agent. It will take lots of time, effort and internal expertise to not only train these models on your chat solution but to also connect it with other internal integrations. 

The best approach – particularly if you’re not a company dedicated to AI – is to buy or use external tools with a vertical AI approach for your specific needs.

A Forbes article on vertical AI highlights that ‘most companies need a model that will incorporate and convert its extensive amount of industry data and expertise into meaningful outputs that deliver targeted solutions addressing industry-specific requirements. These models require specialized algorithms specifically designed for the particular industry or use cases they serve.’

Taking the example of a customer chat bot, you could find a tool that specialises in customer support solutions or even solutions for the insurance industry, making the depth of knowledge much deeper than a horizontal AI model. Many of these vertical AI models also come with built-in connectors with other well-known tools to make implementation a whole lot easier. 

‍

Next: what problem do you want to solve with AI?

When it comes to insurance, we can classify AI initiatives based on how they impact the combined ratio, which is a combination of the loss ratio, expense ratio and acquisition ratio.

The name of the game is to have a combined ratio below one in order to be profitable. Let’s take a look at how AI could fit into this.

• The loss ratio: the percentage of claims paid related to the premium collected. This includes incurred losses – or the amount of money the insurer expects to pay out in claims – and the loss adjustment expenses, which are the costs related to investigating, adjusting and settling claims.

I strongly believe that insurers should be on a mission to maximise the loss ratio in order to create more value for our society, so I’m not a big fan of using AI to lower the loss ratio unless it’s about preventing claims. 

In reality, using AI for more efficient technical segmentation (i.e. splitting good risks from bad ones) will only lead to more discrimination and a bigger insurance gap.

• The expense ratio: the percentage of the total operational costs to administer an insurance program related to the premium collected. This includes commission paid to the distributor to acquire the insurance contract and all the costs associated with underwriting, policy servicing and admin.

One very efficient way to maximise the loss ratio while being profitable is minimising distribution commissions. That being said, I don’t see how AI could have a big immediate impact here. It could identify how to more efficiently distribute insurance, but I think embedded insurance is the ultimate way to optimise distribution cost.

‍However, GenAI can be hugely transformative when it comes to managing the administrative burden of an insurance program.

When we see that the expense ratio represents 25% to 40% of insurance premiums, admin costs become a burden that the end customer has to bear.

‍

Best approach: using GenAI to increase efficiency and optimise cost management

So where can GenAI have the biggest impact? By increasing the efficiency of the biggest department in any insurance business: customer support and claims management.

This is due in part to the fact that insurance involves a huge amount of customer admin, whether it’s customers bargaining the price of their policy; asking questions about their coverage or payments; requesting to adjust, renew or cancel; submitting claims; or the internal work of collecting all the necessary documents.

The department also expands once you have a diverse product portfolio and serve multiple countries and languages.

Insurance support and claims handling is an extremely challenging job: they have to be prepared for any question and deal with customers who are often in tough, emotional situations. 

This is where AI can come in – allowing our team to become more efficient at scale.

‍

Examples of AI in insurance claims management and customer support

A recent Accenture report called AI ‘transformative’ for insurance customers and carriers, particularly when it comes to claims, because ‘insurers can leverage the technology to improve customer relationships through enhanced interactions, while realizing gains in both efficiency and decision effectiveness.’

Here are three AI applications we’re implementing at Qover that increase efficiency and help optimise costs by having a fundamental impact on the customer care and claims departments:

• AI-powered document submission tool: allows customers to upload claim documents directly to their user portal. Using a combination of OCR and AI, we can extract key data, put it in a structured form and then ask customers to confirm whether it’s correct; that way we can automate the data moving forward.

• AI-powered workflow tool: allows us to better manage our workflow by sending the right customer requests to the right department, but also prioritise tickets by analysing the customer’s intent. Not only does this create more efficiency, but it can also lead to quicker replies and higher customer satisfaction.

• AI-powered agent: allows us to more efficiently respond to customer questions. The AI agent can prepare accurate answers based on publicly available documentation such as FAQs or internal template responses that we feed it. To ensure quality control, we go through an initial testing phase where a live agent checks the AI’s answers.

Risks and challenges of AI in insurance claims management and customer support‍

While GenAI can be truly transformative for insurance companies and insurtechs, one of the key risks – aside from potential labour impacts – is the potential financial consequences of allowing an algorithm to automatically answer customer questions.

Let’s recall one such case from Air Canada where an AI chatbot wrongly informed a customer that they would get a refund when they cancelled their flight. While a plane ticket might not cost that much for an airline in the grand scheme of things, it’s a totally different story for an insurance business.

While this may lead to many risk carriers being extra cautious or even reluctant to embrace the AI revolution, the benefits and appropriate mitigation measures significantly outweigh the disadvantages.

For example, here are some measures that could be put in place based on what we’ve implemented at Qover:

• We tell customers when an AI agent answers them without any human supervision. That way we always let customers know that they can choose to be put in touch with a live agent.

• When there’s a question about coverage or something that could impact the loss ratio or claims, we will initially enact a ‘four eyes principle’, meaning a live agent validates the AI’s answer. In the next phase, the AI tool will be able to copy-paste the publically available FAQ content word-for-word, so there is no risk of hallucination.

• Properly configuring and training the AI agent is critical, as is applying the human in the loop principle to continuously check its answers. A well-configured AI agent can answer as well as a live agent in terms of the level of personalisation and accuracy, including translating the answer when needed. 

• Our legal and compliance team makes sure that we act according to the latest regulation around AI. The European AI Act, for example, ensures ‘that AI systems respect fundamental rights, safety, and ethical principles’.

Conclusion: AI can maximise efficiency for insurance players and drive a better customer experience

It’s clear that AI has the potential to transform the way insurtechs and risk carriers do business, particularly when it comes to optimising customer care and claims to provide a better experience for end consumers.

I believe the best way to ensure success and efficiency for this use case is to work with a vertical AI model that specialises in solving the specific problem you have. 

And although there are certainly concerns as we navigate this new tool, the potential benefits far outweigh the risks.