Industry grapples with ‘scalability challenge’ around generative AI

Some market participants now question whether the excitement generated by the November 2022 launch of ChatGPT – and by the subsequent availability of other generative artificial intelligence (AI) models – has really been justified.

In particular, an article in the July 2024 edition of The Economist argued that the supposed AI revolution has, so far, had almost no economic impact.

However, most professionals involved with generative AI in the insurance sector are sure that it’s here to stay and that its impact will eventually be far reaching.

Generative AI refers to deep learning models that can generate a range of new content – for example, text, images and audio.

Martin Stockdale, partner and AI programme lead at law firm Kennedys, explained: “People tend to overestimate [generative AI’s] impact over two years, but underestimate it over five years.

“Many organisations have found they could solve certain problems via experimental usage, but the next stage is scaling it up in an organisation to provide commercial value.

“This is the real business challenge and I don’t think any sector has cracked it yet as the technology is still in its infancy. Insurers may need to have an eye on this scalability challenge [while] still experimenting.”

Scaling up the use of generative AI in organisations will involve looking at everything from the company’s skills, talent base and processes, to establishing what stage its data is at, as well as how accessible and appropriate it is for AI algorithms.

Do companies have a clear programme for change across their organisation to accommodate generative AI and do they have the right relationships up and down the value chain to drive financial benefits?

Scaling up

Because such large, internal tasks cannot be completed overnight, many insurers and brokers are still way short of creating scalable value from generative AI. But others claim to have taken the plunge.

For example, insurer Zurich introduced ZuriChat to all UK staff in May 2024 following an extended nine-month pilot to establish where benefits could be gained, identify pitfalls of the technology and develop an eLearning module, to ensure employees understood the power of the tool.

Staff can use ZuriChat for activities such as writing LinkedIn posts, summarising meeting transcripts and booking desks. At a corporate level, it is used to summarise and compare large documents and to extract insights from unstructured data to improve underwriting and claims processes.

Similarly, broker Miller launched MillerMo to all staff in December 2023. Its functionality includes acting as a personal assistant, proofreading, content generation translation, analysing datasets and software development.

In fact, research by advisory firm Celent – conducted in June 2024 across 24 Europe, Middle East and Africa-based insurance executives in roles tied to innovation – found that 42% of respondents said their organisation is in production with generative AI, implying that the technology has already been adopted at scale in the sector.

Grey area

However, the boundaries between experimental and scaled activity are extremely blurred and some organisations actually experiment with widespread usage.

For example, insurer Axa – which launched its Secure GPT generative AI tool in July 2023 – is unusual in acknowledging that it is still on a journey to scale up generative AI across all its operations.

Secure GPT has comparable capabilities to Zurich’s ZuriChat model.

Paul Hollands, chief data and analytics officer at Axa UK and Ireland, told Insurance Times: “It’s hard to see how anyone has already reached scale with generative AI. We are more likely talking about pockets where smaller use cases are emerging.

“We are at scale in the sense that 10,000 users in [the] UK and 140,000 users globally can use Secure GPT as part of their day-to-day roles. But we have 12 use cases we are progressing right now to embed generative AI into our operational processes and to truly scale.”

No one disputes that current versions of generative AI are still far from perfect.

Data is typically many months out of date because of the need to train generative AI before releasing it for use. Plus, the technology is often still subject to hallucinations – where AI presents false, nonsensical or misleading information as fact. But such issues should be seen in perspective.

Andre du Preez, head of innovation at Miller, said: “In life generally, we tend to follow the idea that if 80% is perfect then we can live with the 20% that isn’t.

“But if generative AI gets one thing wrong, those who don’t like it will oppose it. It was designed to make people more productive, not to be perfect.

“MillerMo has demonstrated very high accuracy rates – in some cases over 90% – and our rule that everything must be double checked before it leaves the building gets around the hallucination problem.

“Nothing is bang up to date right now, but I think that will change in the future. I’m sure that it will eventually be in real-time.”

Future generation tech

While acknowledging that the insurance industry has only just scratched the surface with AI, Du Preez is adamant that MillerMo has already paid for itself. So, the mind boggles at the type of productivity gains that could be around the corner.

Keith Raymond, principal analyst at Celent, agreed that generative AI will in due course use real-time data and feels that, as training evolves, hallucination errors will definitely become less significant than human error.

He explained: “AI is already making efficiency gains in some cases, but the technology is evolving incredibly fast and, at some point in the future, I expect it will touch every aspect of insurance processing.

“Ultimately, it will replace jobs in certain functions, like call centres.

“With underwriting, it’s right now largely an efficiency gain. But, over time, greater automation will increase the cost avoidance of bringing in new people and it could even replace a percentage of people.”

No one is scaremongering that virtually all insurance jobs are going to be replaced by generative AI – but those who don’t use it are likely to be more at risk than those who do.