The rise of chatbots will have ‘an enormous impact on insurance on so many fronts’, but insurers will have to get to grips with the ‘risks and ethical considerations’

Breaking news at the time of writing is that American artificial intelligence (AI) company OpenAI has released Generative Pre-trained Transformer 4 – more commonly known as GPT-4 (14 March 2023).

The launch of this latest multimodal large language tool further increases the AI opportunities and risks facing the insurance industry.

This latest version of OpenAI’s chatbot can respond to images and it processes around eight times as many words as the original ChatGPT model launched in November 2022.

Trained on text taken from the internet, ChatGPT has been designed to provide quick and understandable answers to any question.

Ian McKenna, chief executive of the Financial Technology Research Centre, said: “If you look at what some of these chatbots can do now and extrapolate what they will be able to do in four or five years’ time, it’s really quite scary.

“People won’t have to remember facts and data in the same way and it will have an enormous impact on insurance on so many fronts.

“For example, the lives of people writing marketing content will get easier at one level, but we’ll need [less of these employees].

“Under Consumer Duty, there’s a requirement for everything you produce to be understandable by the client and I’m convinced that insurance people will very quickly be using technology to produce content to make it so.”

But, for the time being, ChatGPT’s accuracy is highly questionable – currently, it is only trained with data up until the end of 2021 and the chatbot cannot tell fact from fiction, or right from wrong.

Therefore, today’s usage of ChatGPT in the insurance industry is for simple background tasks.

Haris Khan, manager of the insurance consulting practice at professional services firm Deloitte, has anecdotally heard of some underwriters using ChatGPT as one tool to help augment the underwriting process. But that’s about as exciting as it gets.

Keith Raymond, principal analyst at research and advisory organisation Celent, added: “I’m not aware of any insurer worldwide using ChatGPT in a consumer-facing way.

“I think they will, but I don’t know when. Insurers are currently looking at the risks and ethical considerations.”

Ghost broker potential

One such risk being considered by the insurance sector is the potential uptick in fraudulent activity arising from more empowered hackers.

Indeed, in March 2023, cyber security firm Darktrace reported the emergence of more convincing and complex AI scams since last November’s ChatGPT launch.

Margaret Scott, director of claims strategy and customer experience at Allianz Insurance, explained: “Ghost brokers may use ChatGPT to draft fake insurance policies or insurance certificates. Scammers could forge documents to purchase cover or commit other types of application fraud.”

However, some market experts believe the impact of AI chatbots on fraud could be neutral, or even slightly positive for the industry because ChatGPT can also greatly help anti-fraud efforts to spot suspicious patterns of activity (see case study).

Phil Price, vice-president of loss adjusting technology at Crawford and Company, said: “In the future, I would like to think [technology like ChatGPT] would benefit insurers the most as they are the ones with a whole host of data they can investigate.

“But they will have to put in the work to link all internal and external datasets to make this happen.”

Part of the broader AI family

The majority of the feedback from insurers on the opportunities and risks around chatbot usage tend to refer to AI in a broader context than simply OpenAI’s ChatGPT.

For example, Axa UK told Insurance Times that within the work being undertaken in its data innovation lab, Atlas – which launched in Q1 2022 – the insurer is exploring using generative AI to supplement the information available to its call centre agents in order to speed up interactions with customers.

Historic policy documents are also being used to train AI models to answer questions customers may have about their policies in easy to understand language.

Meanwhile, Chartered Insurance Institute board director Artur Niemczewski believes that technology such as ChatGPT will prove less relevant to insurance compared to the continued development of other narrower AI models that are designed for specific tasks in the value chain.

He explained: “Hundreds of AI tools developed specifically for insurance will become more appreciated because of the publicity received by ChatGPT. It was a phenomenal marketing campaign for AI, which went viral.”

Khan additionally emphasised that the real impact of ChatGPT has been in the way it has provided more industry-wide optimism in the ability of AI generally to help insurers.

He continued: “A lot of scepticism has developed over the past few years as to what AI could do.

“There had been a lot of hype with it and a lot of failed initiatives. But now, people on the ground are actually wanting to leverage AI.

“However, we will always need humans to oversee some of the tasks. A lot of jobs will change, but will not necessarily disappear. AI just helps humans do more human-like tasks and get rid of the more mundane ones.”

In a nutshell, AI is not expected to replace insurance workers - but those who use it could well replace those that don’t.

CASE STUDY: Peppercorn Insurance

Peppercorn Insurance, an insurtech which started trading last December, uses conversational artificial intelligence (AI) – such as chatbots or virtual agents – to offer motor insurance customers what it describes as a smart, intuitive and user-friendly experience involving no forms.

This approach, which is designed to keep operating costs and expense ratios low, aims to put customers in control and empowers them to ask unlimited questions without fear of being cut short.

Nigel Lombard

Nigel Lombard, chief executive of Peppercorn Insurance, said: “Conversational AI is a lot more specific than generative AI.

“You can’t really have a conversation with ChatGPT and its answers aren’t necessarily accurate. So, in its current form, I wouldn’t let it be used directly to give consumers answers.”

Peppercorn Insurance does, however, sometimes use ChatGPT behind the scenes for getting initial information, which it then checks.

The insurtech has even used the tool to defend itself against fraudulent consumers indulging in “quote manipulation”, such as changing parameters on different quotes to see if they can get a better premium.

“ChatGPT can analyse quotes and work out if they are genuine,” continued Lombard.

“We can then use conversational AI to ask why different answers [were provided] on different quote requests. You can’t do this via forms and the question wouldn’t be asked at call centres.”