Download the PDF
download the pdf
Download the PDF
Download the PDF
visit microsoft
visit microsoft
VISIT WIRED CONSULTING
VISIT WIRED CONSULTING
Find out more
Insurance is by nature a cautious beast. Adoption of AI-driven tools and techniques will be a slow process. Its trickle into new areas of the industry will start with experiments and proof-of-concepts – much like Brit Insurance’s use of aerial imagery to validate natural catastrophe claims. As the value of its various applications across everything from underwriting to customer experience is made clear, attitudes and appetites may begin to change. “Slowly but surely, we’ll start to move the inertia,” says Davison.
One might imagine that AI itself could come up with new forms of insurance altogether. So what was ChatGPT’s views about that? “Yes, AI has the potential to come up with new forms of insurance that are better suited to the needs of customers,” it said. “By analyzing vast amounts of data and identifying patterns and trends, AI algorithms can identify emerging risks and develop new insurance products to address them.” If that’s true, it’s yet another reason why the future belongs to those who embrace this technology at a business-wide level. “This isn't just an ‘IT problem to be solved’,” emphasises Wharton. “This is a fundamental business opportunity.”
The Way Ahead
AI is fuelled by data but insurance is subject to regulations that limit what data is admissible and how it can be used. That’s already having an impact on emerging AI applications such as fraud detection. “The ICO [Information Commissioner’s Office] is saying they're going to take a really dim view of applying a lot of voodoo to decision making,” says Bullers. “So if you're using voice analysis in claims to see if the person's stressed [as an indicator of potential fraud], they're basically saying that's pretty much an unlimited fine.”
To avoid being backed into a corner where restrictions stop AI in its tracks, Fretz argues that the industry needs to proactively implement sound practices at every turn. “Unless we self-regulate,” he says, “then regulators will slap us with something that we can't work within.”
Regulatory Barriers
One way to manage that might be to see AI as an advisor rather than a decision maker. “For me, it's all about the word ‘scoring’,” says Sawhney. “I don’t think you can just make decisions based on single factors like whether someone has leaves in their gutter on a satellite image.” Keeping a human in the loop can help ensure that adjustments are made where appropriate and outcomes are kept equitable. “Rather than just seeing AI as a way to reduce coverage, perhaps we can see it as a way to better assess the risks that can’t find cover today,” adds Sawhney. In that way, insurers may also be better equipped to decide to take on higher-risk consumers as a specific selling point.
AI adoption introduces a host of ethical questions, particularly around bias and discrimination. One risk for insurers is that the data they use may reflect social biases, particularly biases relating to gender and race, leading to those groups experiencing unfair outcomes.
That issue is exacerbated by what is known as the ‘black box’ problem: a user knows the inputs and the outputs of an AI tool but can’t explain how the software arrived at its decision. In financial services that’s a pressing concern, because it poses a liability question that can deter adoption. “I think there's a danger down the road of not being able to demonstrate your decision-making process in a particular situation and replay it back to the regulator,” says Bullers. “You have to be able to prove you made the right decisions on the day.” Explainability is a key research area for AI, and one that insurers will be watching closely.
Insurers face a further ethical conundrum. On the one hand, more personalised insurance products more accurately reflect individual risk – on the other, is it right that someone who is riskier through no fault of their own should be singled out for a more expensive policy or find themselves unable to get insurance at all?
Ethical Concerns
Ash Jokoo,
Group CIO of Direct Line Group.
“AI can help us with all the boring bits that aren't important, to get to the human interaction. And it's that human element that I think is super important.”
There are worries about outsourcing too much work to computers – and not only because of the need for empathy when a policyholder has experienced a loss. “I’m mindful of displacing the workforce because I think we could be losing good intellectual grunt and horsepower,” says Brockway. Despite the term, AI is not intelligent – it is simply powerful pattern recognition – and skilled people will be required to step in when a problem is beyond its scope or the system hasn’t reached a desirable outcome. For our industry experts, it seems the optimal solution is one in which AI complements, rather than replaces, humans. This might be Chat GPT helping a salesperson ask better questions or it might be human adjustment of an AI-triaged claim. “AI can help us with all the boring bits that aren't important, to get to the human interaction,” says Jokhoo, “And it's that human element that I think is super important.”
Fears Around AI Replacing Humans
Whether it’s rooted in fears over job security, resistance to change or a lack of understanding that makes the technology feel like a leap of faith, there can be a culture in insurance firms that is averse to new technologies such as AI. “I'm not sure everybody believes that this is a good idea,” says Wharton. “I don't see too many people saying, ‘Help me do this.’” This culture is compounded by industry norms. Insurers need reinsurance, and reinsurers trust conventional approaches to underwriting. Davison puts it straightforwardly: “If you went to a reinsurer and said, ‘I've got a machine that does all my risk selection’, they wouldn't support you.” He believes this will change, but it will happen slowly. As AI adoption continues to grow, and more insurers start to realise the associated advantages of greater accuracy, efficiency and reduced costs, we’re likely to see a groundswell of adoption as the laggards race to play catch up.
Reticence To Engage
One promise of AI is that it can unearth hidden correlations that can help predict risk, possibly drawing on data sets that previously weren’t thought useful. But there’s a problem: you don’t know if a new data set will yield those correlations until you look for them. Gathering, preparing and analysing data that adds little value is a waste of time and money. “That's my challenge for AI: can I use it to help me find the external data that's going to be useful?” says Gareth Wharton, Cyber CEO at Hiscox. “Because I could go and get data from anywhere, but how would I know if it's worth it?”
There’s a further consideration. The better the data, the better the AI, but insurers have to perform a balancing act. If an AI tool is too smart – if its ability to second-guess behaviour seems uncanny, or if its knowledge feels too intimate – it can prove disturbing to consumers. “We need to be able to have that depth of understanding in a way that doesn't freak them out,” says Jokhoo. In the past, when insurers have revealed how many data points they collect or stated that they analyse non-verbal cues to detect fraud, such as facial expressions in claim-report videos, it has caused outrage online. This is an area in which insurers need to tread carefully as they decide what data to use for what purposes, and how they communicate those decisions to the consumer.
Data Selection And Consumer Acceptance
Some insurers, and some types of insurance activities, have seen AI adoption accelerate in recent years. However, AI disruption seen within the sector has not been as rapid nor as broad as many had predicted. “I’ve been in car insurance for 27 years,” says Andrew Brockway, CTO of Confused.com. “We're asking customers the same questions now that we did 20 years ago.” There’s a reason for that: there are a number of challenges to solve which have, rightly or wrongly, squeezed the brakes on AI adoption. So what are they and how might they be overcome?
What’s Holding Back AI Adoption?
Customer retention is influenced by the experience that customers have at every touchpoint with their insurer. And AI offers opportunities (and risks, as we’ll see later) here too.
“We put an awful lot of energy into optimising customer journeys,” says Davison. “AI allows us to accelerate that optimisation and also deliver personalised experiences so not everybody gets the same journey.” One element that insurers are increasingly building into customer experience is AI chatbots. Done well, these can remove wait times, and rapidly complete tasks – such as adding an extra driver to a motor insurance policy – to the satisfaction of the customer. The AI can also route the interaction to a human should the need arise.
Optimising Customer Experiences To Improve Retention
AI’s ability to support personalisation of products and services could lead to the wider use of emerging products such as ‘highly dynamic, usage-based insurance (UBI)’. This is where premiums adapt constantly to changing conditions, and billing operates on an ‘as-you-go’ basis. “We're seeing it already in shipping, when the location of a cargo ship relative to pirate waters or land changes the policy in real time,” says Fretz. “For general insurance we’ll also see risk being assessed in real time and having that reflected in continuously changing policies.”
Motoring is the obvious use case, particularly if car sharing becomes more commonplace in the future thanks to autonomous vehicles. A real-time, ‘pay per trip’ insurance model benefits insurers by allowing them to evaluate risk on a case-by-case basis – if a motorist tends to speed and is driving in the rain, for example, then the price can reflect the elevated danger level. But it also rewards customers: those who drive safely pay less, and those who drive infrequently don’t bear the cost of an annual premium.
Enabling Real-Time Policies
A make-or-break moment for customer satisfaction is when making a claim. “Claims is really the product that we're selling,” notes Sawhney, “and improving the customer experience here can pay dividends.”
Insurers are already using AI for rapid damage assessment, particularly for vehicles – policyholders submit photos or videos and the algorithm estimates costs. And further use cases are emerging. “I have been approached by start-ups playing around with computer vision on building sites,” says Boyd. “From a claims handling perspective, if you were able to take a pre-event and post-event image of that site, and have an AI run the differences, you could then report back to the relevant insurer a proposed estimate for the damage quite accurately.”
While speed is of the essence in claims, customers also need an experience which is empathetic. That’s why the human touch is important in claims processing – but AI can help by improving that interaction. “Using the technology to be the ‘third person in the room’ is quite interesting,” says Simon Bullers, CTO of Hastings Direct. “If you think about most claims, there are multiple calls to the customer to gather the correct information – sometimes quite stressful calls. The more you can use AI to analyse the conversation in real time and prompt the contact centre agent to actually get the right information, the better.”
Rapid claims fulfilment is not only good for the customer, it also brings down costs for the insurer – the faster you know that a car is a write-off, say, the less time you have to provide the policyholder with a hire car.
Expediting Claims For Better Customer Outcomes
Insurance is fundamentally about judging risk. The price and scope of a policy ought to reflect the likelihood of a customer making a claim plus the likely cost of that claim. The better an insurer is at evaluating this, the more competitively they can price their products without jeopardising their balance sheets.
AI can improve the accuracy of risk assessments. It is a powerful tool for identifying correlations that are not immediately obvious, and it can do so across a wide range of data sources. In the future these sources could include data from wearables, smart home devices, vehicle data and social media. “AI enables insurers to make use of those more abstract data sources by pulling out the pertinent information to better assess risks,” says Fretz. There are obvious questions over what data is legitimate and where consent must be given. But as this trend develops, it will likely lead to more tailored policies that are priced based on an individuals’ specific attributes rather than a broad profile. This would not only reduce the insurer’s loss exposure, but also improve terms for less risky customers.
Improving Risk Assessments
Fraud costs the UK insurance industry over £1 billion each year, according to the Association of British Insurers. Thanks to its pattern-identifying powers, AI can help to tackle this costly issue by detecting and predicting fraudulent behaviour. That could involve everything from recognising when multiple claims have been made for the same incident, to using image analysis to judge whether car damage is consistent with the accident described.
Since the correlation between policy fraud and claims fraud is high, it pays to predict fraudulent activity at the application stage. “If we can uncover and stop that risk then clearly we turn off the tap at the earliest opportunity,” says John Davison, CIO at First Central Group. “A good example might be if the same device is buying multiple policies for different people. If you can spot that in real time, which is where the machine learning comes in, that gives you a material advantage.”
Stamping Out Fraud As Early As Possible
AI has a potential role to play at every stage of the customer acquisition process. Firstly, it can help identify potential customers. Just as Brit is using aerial imagery to assess damage, for example, satellite imagery can also be used to generate leads – if someone has a swimming pool, say, the AI can target them with a pool-insurance offer.
The sales process, too, can be enhanced. AI can offer relevant products that perfectly suit individual needs and it can lower the cost of sale. “There is definitely a banging on my door from within the company to use innovation to deliver really efficient sales,” says Ash Jokhoo, Group CIO of Direct Line Group. This might involve using AI to intelligently design cover based on a small number of questions, before rapidly providing a quote and processing the contract.
Enhancing Customer Acquisition
Artificial intelligence is able to process data on a scale that humans could never handle, and find patterns that humans could never spot. This can unlock significant efficiencies, saving time, reducing costs and improving profitability across everything from customer acquisition to claims processing. Critically, AI-enhanced processes can also improve customer experiences to the benefit of both insurance companies and their customers.
The effectiveness of AI is a function of the quality and quantity of the data it’s fed, so as more data from more sources becomes available to insurers, the potential for AI to transform the industry only grows.
So what are the use cases that we expect to realise the greatest benefits?
So, what are the emerging use cases, and what are the challenges that will need to be solved? In partnership with Microsoft, we convened a roundtable of industry leaders to explore their vision of AI and the future of insurance. And yes, we asked ChatGPT for its opinion too…
ChatGPT brought home the power and possibilities of AI to society at large. That included insurers.
It’s perhaps counterintuitive, then, that AI deployment in insurance is not more widespread than it currently is. Implementation has been relatively slow thanks to a number of obstacles ranging from regulation to corporate inertia. But recent developments are galvanising the industry to explore the technology further.
When OpenAI’s chatbot ChatGPT took the world by storm in December, it turned AI into the biggest story in tech. Thanks to the intuitive interface, the readily appreciable natural language results, and the fact that anyone could use it for free, ChatGPT brought home the power and possibilities of AI to society at large. That included insurers.
This step-change in awareness of AI, its accessibility, and its use cases, has captured imaginations in boardrooms which were once reticent about the technology, spurring them to think more broadly about how it could improve their businesses. “When I was starting my career, the feedback from the top was all about mobile apps,” recalls Richard Boyd, Head of Digital Claims at Lloyds of London. “The more I reflected on that, it was not really about ‘building a mobile app’, it was about the executive recognising that there was a broader strategic opportunity that mobile devices were presenting to us. I think ChatGPT has done that for the AI conversation.”
Leon Fretz, Senior director of
financial services at Microsoft
“Insurance is a business built on data – so AI has potential applications across the board.
I fundamentally believe it’s going to dramatically change insurance products.”
Simon Bullers, CTO of Hastings Direct
“The more you can use AI to analyse the conversation in real time and prompt the contact centre agent to actually get the right information, the better.”
John Davison, CIO at First Central Group
“AI allows us to accelerate that optimisation and also deliver personalised experiences so not everybody gets the same journey.”
How AI Could Change The Game For Insurance
When Hurricane Ian made landfall in Florida late last year, it went on to wreak havoc.
With sustained winds of up to 240 km/h, it was the joint fifth most powerful storm ever to hit the US – and the country’s third costliest weather disaster in history – resulting in estimated losses of $112 billion.
To insurers, this kind of natural catastrophe presents not just a financial challenge, but also a practical one. How do you expedite claims and help people as fast as possible?
For Brit Insurance, the answer came courtesy of AI. Its rapid damage assessment tool, nicknamed ‘Golden Eye’, uses a machine learning algorithm to assess post-disaster aerial imagery and group properties by level of damage. This enables claims to be triaged before they are even reported. Brit had many policyholders who were affected by Hurricane Ian. When the tropical cyclone dissipated on 30th September, the company was able to make its first payment just eight days later. “We could never have been able to do that with humans,” says Sheel Sawhney, Brit’s Head of Claims and Operations. “It has been transformative for us.”
Damage assessment is just one area where AI is reshaping insurance. It’s an industry predicated on using massive data sets to model scenarios, screen customers and improve outcomes. Since AI is effectively an advanced form of data analysis, it is well suited to enhancing insurers’ operations. “Insurance is a business built on data – so AI has potential applications across the board,” says Leon Fretz, Senior Director of Financial Services at Microsoft, whose Azure cloud computing platform underpins Brit Insurance’s tool. “I fundamentally believe it's going to dramatically change insurance products.”