Insurers are unlocking advanced technologies and machine learning at scale, as talk at last gives way to action

By Miranda Maxwell

As ChatGPT and Google’s Bard dominate water cooler gossip, insurers have been quietly yet busily building their own artificial intelligence and machine learning war chests.

Large language models, known as Generative AI, offer a multitude of use cases for the insurance industry to take advantage of the ability to create and analyse text, images, videos and audio.

The newest versions stand apart due to their use of transformers – neural networks that use an “attention mechanism” to process data. 

“For the first time, we can create models that are trained on immense data sets – trillions and trillions of tokens or words sourced from the Internet and other sources,” Alex Taylor, QBE Ventures’ Global Head of Emerging Technology, said. “These models can approach what some would describe as…applying intelligence to that knowledge.”

QBE Ventures says we should expect a lot of manual operation in all stages throughout the insurance value chain to disappear.

“We’re going to see all of these things that insurance has traditionally put in the bucket of either ‘too hard’, or ‘manually solvable but very expensive’, down to ‘fast, cheap and easy’. That’s where we’re going to see transformation.

“There’s this holy grail that’s been in insurance for a very long time now in unstructured data processing, particularly in complex product lines where you might get hundreds of pages of submitted data. These are the things that can take days at a time to understand exactly what you need to know to underwrite a particular policy.

“The promise of what that can transform – knowing that this is the last bastion of an entirely manual human operation in insurance – could be highly dramatic. Where a human might take days, [a machine] might take minutes or seconds.”

The application potential of AI is extraordinarily broad. As well as automating and speeding-up tedious tasks such as document review and data entry, it can create personalised insurance policies based on a customers’ specific risk profiles, improving the accuracy of pricing and reducing claims, and can manage marketing campaigns based on demographics, purchasing history and online behaviour. 

At policy inception or claim time, it can perform basic interactions with customers such as generating documents and answering questions.

Telematics devices used in usage-based insurance policies rely on AI, as do the chatbots found on most insurance company websites. 

Underwriters can use AI to scour online reviews, public records and social media to build profiles on applicants. This AI “risk scoring” can aid development of customised products based on models that can monitor emerging loss, pricing trends and shifts in the portfolio risk mix.

Generative AI can classify, edit, summarise and draft new content. Some key benefits are data insight, as well as speed, accuracy and cost improvements.

And AI can help insurers detect fraud by analysing patterns in data, flagging suspicious claims or behaviour.

During claims, AI can compare images and video of damage and compare it with policy documents, returning coverage decisions and settlement offers in a fraction of the current time. 

High-profile US insurtech Lemonade already handles close to half of its claims using AI, and recently announced it settled one within a record two seconds. Lemonade Chief Executive Daniel Schreiber says “this is what 21st century insurance feels like”.

AI is also being trialled by Chubb, which has purpose-built technology centres in Greece, the US, India and Mexico to wrangle masses of data. Chubb has been experimenting with AI use in its underwriting and risk teams, as well as in claims. It’s also being used to help with analytics for marketing and customer service.

A variety of cases have proven themselves, according to Chubb Chief Executive Evan Greenberg, who recently told analysts he believes insurers are “in the dawn of the period where we use these tools at scale”.

Google, Netflix and Uber are setting the benchmarks for AI maturity and capacity, consulting group McKinsey says, and illustrate potential gains that could be realised by insurers – namely augmenting complex decision-making, creating predictive models and aiding targeted customer engagement.

McKinsey estimates comprehensive AI adoption could add up to $US1.1 trillion in annual insurance sector value, driven by more efficient pricing, underwriting, promotion, service and personalisation.

“Investment in AI can be game-changing, and it will increasingly become a source of competitive advantage.”

As Chubb and Lemonade show, insurers are getting on board at pace. Locally, IAG is using an army of automation “bots” to cut tedious tasks, with estimates that by 2026 it will be saving 500,000 work hours each year.

In February, when IAG was hit with 16,000 claims in a week after extreme weather in New Zealand, it used the technology to cope, automating back-end work by recording and capturing information and – as Chief Operations Officer Neil Morgan puts it – “ingesting that content and acting on it really quickly”.

“We sort of think of the bots as part of our workforce,” he tells Insurance News

IAG has also invested in Ravin AI, which inspects the condition of vehicles by using AI to examine mobile phone and CCTV footage. It’s a solution IAG plans to roll out across its brands, and it has embedded AI to identify total loss motor claims at the point of lodgment.

“We can use AI to predict [total loss] at a very high confidence of over 90%, and off the back of that we can reduce end-to-end time for a customer…from four weeks down to a couple of days,” Mr Morgan says.

Applications are growing across the board, with the prevention of workplace injuries being made possible by smart wearable technology, and better real-time flood management with satellite imagery.

“In terms of risk knowledge and presentation, AI is potentially a game-changer,” Clayton Utz Partners David Gerber and Lucy Terracall say in a joint report. “AI promises to increase by orders of magnitude the efficiency of collation, synthesis and transmutation of vast quantities of raw, relevant data, into forms useable by underwriters.”

Home and motor policies that are applied for by a customer, underwritten by an insurer and generated entirely through an AI technology platform without any human input “may well be not far away,” Clayton Utz says. 

Life insurance may quickly follow, helped by publicly available data – and disclosures may become “redundant altogether” for some classes of insurance. Underwriters of more complex commercial policies, including directors’ & officers’ liability and professional indemnity, are not readily replaced by technology but “there is much that can be done to help”.

Still, the law firm says the human underwriter and intermediary isn’t going anywhere soon and the value of the data analytical capability is “substantial but not determinative”.

“If you ask an underwriter whether their discipline is a science or an art, they will likely say that it is a bit of both. Insurance brokers and underwriters will not lightly abandon human interactions with their insureds, which can be so important to gain confidence.”

At MLC Life, AI augments the work of claims-handlers and “frees them to have great conversations the customer actually remembers”, Chief Claims Officer Andrew Beevors says.

“If you’ve got that little co-pilot on the shoulder that can prompt you a lot earlier, you can intervene a lot earlier. Therein lies where AI is supporting the claims-handler to be more proactive in terms of how they respond.”

QBE’s Mr Taylor also says that while AI will “disrupt and transform” the insurance industry, it will facilitate rather than “implode” current business models, and the “threat of [staff] being replaced is probably not real at the moment”. 

“It’s not like what the internet did to the media companies; it’s completely different,” he says.

There is particular promise in combining language models with embeddings and vector databases to help identify the commercial aspects of binding a policy – a task that QBE Ventures says can be done dramatically more quickly than by a human, extracting underwriting fields from unstructured text.

The other key change is the accessibility of use to all “without having to be a software developer or a machine learning expert or a mathematician”. 

“Platforms are emerging that can turn anybody into a data analyst,” Mr Taylor says. “That’s this inflection point that the world’s collectively coming to and asking, ‘what are the ramifications of this?’ What does it mean – that something that I did yesterday maybe a machine can do tomorrow, and it might be faster and cheaper? 

“Answering that question is precisely what the world is going to collectively discover over the next two to three years.”

He says the real opportunity – and unknown – is what the industry will do with all that freed-up time.

“Does that translate to more effective premium, or products that might emerge that simply aren’t commercially viable today? All of these things will be true to differing extents,” Mr Taylor says.

“To have all of these things offered by a single technology – this is the first time since the emergence of computerisation in the insurance industry nearly 60 years ago. 

“We have a unique, once-in-a-generation opportunity to do something extremely exciting, and we’re in the first year of it right now.”