Getting beyond the AI hype

Two years on from the launch of ChatGPT, and some of the hype has evaporated. Rather than being an end in itself, generative AI is increasingly another tool in the digital toolkit. This shift away from what Paul Hollands, Chief Data and Analytics Officer at AXA, calls ‘bright shiny-itis’, is to be welcomed as it means investment can be targeted on those projects most likely to drive the business strategy and deliver on the financial plan.

‘Has generative AI changed what the future will look like, or the pace we get there?’ asked AXA’s Hollands. AXA itself is betting hard on the technology, which is at the centre of all its transformation efforts across its retail, commercial and health lines with around one in three of its UK employees already interacting with the inhouse generative AI solution on a daily basis. ‘That’s the scale of it now,’ said Hollands.

Ensuring these investments add genuine value means building re-usable solutions so that each use case becomes incrementally quicker. AXA’s call summarisation tool for 500 agents in the UK was swiftly adapted, reusing the same capability and infrastructure, to deliver a vulnerable customer identification use case.

‘It means we were able to deliver at pace,’ said Hollands. ‘It just took a week from concept to delivery.’

Likewise, when the risks with reinforced concrete emerged, AXA was able to analyse 70,000 documents in just seven days, and from that identified a potential issue for four customers, who were immediately contacted with offers of support. As the industry moves towards a risk management partnership model, this kind of rapid and targeted intervention will be key – and AI is the enabling technology.

How is Allianz using generative AI? We caught up with Firas Ben Hassan, Allianz Technology’s Deputy Head of Data Science Services, to find out.

Companies across the insurance spectrum are reporting similar gains. Chris Pearce, Head of Data Science at esure, said the insurer is already seeing ‘very tangible benefits’ from the deployment of generative AI.

‘The main impact is in the streamlining and optimisation of digital customer journeys, where we’ve seen huge gains in efficiency and AHT reduction,’ he said.

Brokers are also gaining from AI-powered transformation in the underwriting function.

The ability of AI models to actuarially price and assess risk means we are getting underwriting decisions in almost real time and when you remove friction from a transaction then everyone benefits. It’s long been a trait of our business that we have to predict insurance costs and insurance capital and for that to now be in our hands rather than relying on anecdotal advice is very powerful.

LUKE VEVERS
VICE PRESIDENT,MARSH

Ensuring this potential is realised, rather than lost to ‘shiny-itis’ experimentation, means a focus on real-world business problems.‘You have to be clear what it can solve and what it can’t compare to other machine learning methods and analytics and insights,’ said Pearce. ‘Is it a deterministic problem, a stochastic problem, what is the variance of error you are willing to take operationally and commercially?’

He pointed out that given it’s now the end of 2024, the proof of concept has been made. Now it’s about impact and value.

You need to make sure there’s a strong business case for each use case and that you can measure that value or you will end up with a lot of MPVs and POCs, and you could spent your entire budget on these with no impact.

FREDERIC CRECCO
DIRECTOR OF GROUP APPLIED TECHNOLOGY,AGEAS

Keeping that focus on end value means a roadmap that defines the problem, involves end users at every stage and recognising that deployment of AI is a team sport.

‘It’s a misnomer that you can roll-out co-pilot and that’s it,’ said Chris Pearce of esure. ‘That’s the tip of the iceberg.’

We sat down with Darran Simons, Head of Insurance for EMEA at FICO to ask him in more detail about responsible implementation of AI.

Beneath that, he listed infrastructure, architecture, software development – both middle ware and front end – SRE and networking, data engineering, data science and then a supporting cast of Infosec, legal and compliance.

The latter point is important, and one that is also maturing as the technology is better understood. Danny Hoskin, Principal Data Scientist for Artificial Intelligence at Aviva, said that governance, guard rails and auditing are essential to ensure outcomes are accurate and ethical but that the world needs to recognise that humans are fallible too.

‘Hallucinations are the biggest unsolved problem right now,’ he said, ‘but the humans you put in the loop can make things up too. There is a case that we’re holding AI to a higher standard than we do humans. There’s s till a journey to go on.’

That journey is, however, increasingly informed by understanding, experience and pragmatism, rather than hype. And that can only be a good thing.