Artificial intelligence is the technology game-changer of the moment. Predications of its impact range from a cautious integration into our day-to-day lives to the cataclysmic. The truth is that no-one really knows.
What we do know is that the insurance industry will not be immune from its potentially transformative impacts. The big questions are: Where will it hit hardest? What impact will it have? Where will human input play a part?
Generative AI undoubtedly has huge potential to improve the insurance sector
There is no hiding from the AI revolution. Back in March, as ChatGPT hit the headlines, the Future of Life Institute rounded up an impressive list of experts and tech sector leaders, including Elon Musk, to call for a pause in the development of AI. This call predictably fell on deaf ears. By early July Musk was announcing the development of his own AI package. Hard on the heels of Musk’s conversion, Google launched its Bard AI tool into Europe, having addressed arrange of concerns about data use and privacy.
The insurance industry knows it must embrace AI, says Rob Stewart, Head of Sales at the UK’s Claims Consortium Group. “Generative AI undoubtedly has huge potential to improve the insurance sector, which is in essence a service industry. Insurers will certainly be able to use it to enhance risk assessment and sharpen their pricing models with natural language processing enabling the analysis of large volumes of data and providing the opportunity to identify patterns that may indicate potential risks.
“Outside of underwriting, sales and customer support, however, perhaps the most exciting opportunities are in the claims area.”
Seizing these opportunities will not be an easy process for many insurers, warns Michał Trochimczuk, Managing Partner of Sollers Consulting. “One in three companies is already using artificial intelligence. But insurers working with an IT landscape not designed for this purpose will struggle to get quick results from AI projects. This is the reason why insurers are currently using AI in very specific areas.”
Traditionally, AI has been implemented in claims. But there are use cases of AI in many other areas of property and casualty insurance.
Like Stewart, he says claims is where most of the focus is currently. “British insurer Ageas and US insurer Kingstone use AI in claims reporting. The Polish regulator KNF has implemented AI in reserve verification. Several insurers are still exploring their options. HDI Global Specialty Norway is developing an AI solution for complex claims together with the broker Söderberg & Partner. Zurich Group has announced its intention to apply Chat GPT in the claims area.”
While claims is the main focus of attention, it is by no means the only area where insurers are deploying AI, says Trochimczuk. “Traditionally, AI has been implemented in claims. But there are use cases of AI in many other areas of property and casualty insurance. They include fraud detection, where the application of AI is already quite common, pricing, prevention, marketing, customer service, legal departments, cyber defence and underwriting.”
Fraud
The constant thorn in the side of all personal and commercial lines insurers, fraud, is where AI is already proving its worth, says Nika Lee, Chief Underwriting Officer for Aioi Nissay Dowa UK. “Like many insurers, Aioi Nissay Dowa Europe was dedicating significant time and resource to the detection of fraud, relying heavily on market intelligence and human skills. We needed a solution that would automatically flag if a claim could be fraudulent so that these cases could be prioritised for further investigation. This would then free the human skills of our team to support genuine customers at the point they need us the most – at claim.”
Investigations are now closed faster, accompanied by reports, supporting evidence, and investigator notes
The insurer teamed up with AI specialists Mind Foundry to develop a continuously evolving, meta-learning solution which can detect opportunistic and organised fraud.
Lee speaks enthusiastically about its success. “Investigations are now closed faster, accompanied by reports, supporting evidence, and investigator notes. In addition, there is enhanced ownership by claims handlers because they trust the solution due to its explainable nature and want to contribute to its improvement.”
She says the results are impressive. “Since introducing the new solution, fraud detection rates have more than doubled, referrals rose to 18% from 2% and claims handling costs have been cut. All this means that handlers work on much fewer false positive cases, resulting in better outcomes for customers as we are able to process claims more quickly and deal with genuine claims far more efficiently.
“This collaboration between humans and AI is providing benefits for the insurance industry and consumers by helping to keep the cost of fraud down.”
Humans
It keeps coming back to where we fit in. “Human in the loop” is a term taken from the military world that is frequently quoted when people talk about applying AI in a wide range of contexts. There have been plenty of examples of what happens when you take the human out of the loop and perhaps trust AI too much, especially when using it to source information from a data pool that might be polluted with incorrect facts, figures and analysis. Just as AI can be used for good, it can be used for evil too, with a range of threat actors ready to exploit its darker side.
Stewart says insurers must be on their guard against this and the lack of “emotional intelligence” most AI models currently exhibit. Using the skills and knowledge of their people will be key to seeing off the potential for serious reputational harm. “There is a risk of misinterpretation of customer queries which could lead to incorrect responses and drive increased customer complaints. This is partly because tools like ChatGPT are only as good as the data they have been trained on, and responses could be biased, causing ethical and legal and regulatory challenges.
“The bottom line is that despite all the claims being made for AI it cannot replace experience – particularly around fraud and claims – and any investment must be supported by in-depth training and access to experienced, knowledgeable humans.”
Where the power of AI to gather and process large amounts of unstructured data very quickly has the greatest potential is in underwriting, although this is where the insurance industry will run into the concerns about the use of personal data that are exercising the minds of governments and regulators.
Several major insurers, including Chubb and Zurich, are already running major projects to explore how far AI can be deployed in this area. Australian insurer IAG has invested in the Israeli insurtech Planck, which specialises in real-time analysis of data for underwriters, Japanese insurer Sompo International already uses Planck's technology.
Aioi Nissay Dowa has focussed on using AI to process data captured by black box telematics systems including location, speed, acceleration or braking, fuel consumption, idling time and vehicle faults, so it can be analysed to identify vehicles which may be used for deliveries and price them accordingly.
“Of course, the volume of data generated by telematics insurance is too great for manual review, which is why machine learning blends in analysis of GPS data to provide a prioritised ordering of policies for human investigation”, says Lee, again stressing the importance of applying the more nuanced and subtle skills of its people.
Of course, there are those who see more potential harm than benefit in artificial intelligence who will cast doubt of the continued presence of humans as AI grows in power and sophistication but, for now, the human is staying in the loop as far as insurers are concerned.
