It is now an accepted truth across the industry that artificial intelligence (AI), aided by machine learning and an increasing ability to extract patterns from big data, is in the process of transforming virtually all sectors of insurance, from claims automation to fraud identification. Travel insurance is no exception, with some insurers already deploying the technology, even though travel insurance is a unique product – there is no other insurance line that sells so cheaply and risks so much.
A UK-based traveller can buy insurance for under £10, or even get it free with a new bank account, then slip on a step in Houston, Texas, and break their back, and their travel insurance company will be on the hook for the cost of specialist care and support. That £10 premium income has brought with it a multi-million-pound liability. In that sense, travel insurance will continue to be a product that, by its nature, is always going to be at risk of a claim that is wildly disproportionate to the size of the fee charged. AI can’t change that.
Of course, all insurance is about pooled risk, with multiple premiums providing cover against the – hopefully – few catastrophic claims that do arise. Sometimes – catastrophe insurance often being a case in point – the annual claims total is well in excess of premiums garnered and the industry takes a hit. AI can’t do a great deal about events that happen in the real world either, though it looks certain to help insurers understand risk patterns more precisely and thus calculate premiums more effectively and individually.
AI is already helping to sharpen statistical calculations of the risk involved in a product, and can help with the premium pricing. But this is something that good, old-fashioned human actuaries, working with historical statistics, have been doing for years. The real impact of AI over the next few years is more likely to lie elsewhere.
One of the first places where AI’s impact on travel insurance is being felt is in the automation of claims. Any provider of travel insurance has a vested interest in building up a loyal customer base. An essential part of creating this loyalty is the ability to provide a rapid decision and payout when claims come in, improving customer satisfaction rates.
AI can help here by applying machine learning to the insurer’s massive database of past and present claims. It can spot anomalous claims that suggest potential fraud might be in play, and flag that claim up for further analysis. The basic idea is: pay simple claims rapidly and, in doing so, keep the customer happy. This may involve straight-through automated payment made direct to the customer’s account, with no further human intervention in the claims processing cycle. More complex claims, however, can be flagged as such and routed to a claims assessor, while potential fraudulent claims are routed to the fraud analysis department for further investigation.
Travel insurance is, by its nature, a high-volume, low-margin business, so anything that AI can do to drive cost out of the process will be – and is being – embraced by insurance providers.
Automatic for the people
One way of doing this, apart from adding AI to a workflow system, is to break travel insurance down to its various constituent parts, and then automate the simplest parts as fully as possible. Jamie Hersant, Head of Lifestyle Claims at AXA, points out that AXA has embraced this approach fully. Claims under benefits for missed flights, delayed flights, lost baggage and damaged or lost personal items can usually be automated in a ‘straight-through’ manner. Laurent Benichou, Director of Research and Development at AXA GIE, points out that the company was able to deploy a blockchain-based smart contract to compensate travel insurance customers for flight delays because a delayed flight is recorded directly on an external database. This provides instant proof that a claim has been triggered and the payment can be sent automatically to the user’s bank account without them even needing to make a claim. This is a real, zero- touch, web-based contract cycle. Once it is up and running, it is almost cost free to the insurer, which makes it a wonderful product from the insurer’s perspective, and a very satisfying one from the consumer’s. They got their cash compensation without doing a thing other than buying the contact online.
More complex ‘big-ticket’ claims are much tougher for AI to deal with. According to Hersant, AXA’s approach to date is to automate the process of submitting a claim, while relying on the expert assessor for judgements on more complicated claims.
Cutting human beings out of the equation can work. Fukoku Mutual Life Insurance laid off some 34 staff and set itself a target of saving around US$1 million a year through installing a £1.4-million AI claims automation system based on IBM’s Watson AI. The task was to analyse medical claims by looking at the associated documentation, in order to determine if the claim was valid. The system can read medical reports and ‘knows’ the recovery processes associated with specific treatments or surgical interventions. This understanding enables it to make a judgement on what constitutes standard costs incurred for valid treatments, and thus if a medical report appears that does not fit into the typical billing sphere, it can be flagged up as potentially fraudulent.
According to press reports, Fukoku Mutual Life Insurance achieved a 30-per-cent increase in productivity and saw savings of around $1 million a year. The insurer’s investment in the AI system, which was put at $1.4 million, was recouped inside of two years. The insurer makes some 132,000 payouts to policyholders a year.
A 2018 study by the global market analysis house McKinsey & Company set out to sketch the way AI was likely to be deployed across the insurance sector by 2030. The study identified four key trends and themes: the explosion in the number of connected devices, via the Internet of Things (IoT), producing mountains of data; the increasing prevalence of robotics in everyday life; the impact of open-source and data ecosystems; and advances in cognitive technologies.
As yet, travel insurance companies have not really made any noticeable or well-publicised advances towards linking data from wearables and the IoT to the pricing of their travel insurance products. However, it is relatively easy to come up with scenarios in which such information could be used to tweak travel insurance pricing in real time. A traveller choosing to go on a walkabout in an Alpine village when the streets are covered in slippery ice is clearly putting themselves at greater risk of a potentially damaging fall. The next generation of smart watches will probably be able to sense the lack of grip underfoot and flag up the danger to the individual.
From here, it is not a huge step to envisaging a travel insurance policy where the premium was attractively discounted if the user agreed to having their wearable device monitored and the price incrementally increased if the perceived risk to the individual increased. An AI system could then fire out a warning of the potential premium increase, which would be activated, in the above example, if the individual did not return to firmer footing.
Similarly, if the traveller was doing a tour of bars and getting themselves intoxicated, the wearable could flag up the increased risk and the AI could increase the premium and automatically debit the consumer’s account – as per the prior travel insurance agreement.
In a crude kind of way, personal security firms providing protection services to firms with staff abroad already use GPS monitors on the phones belonging to the staff they are covering to track whether or not they are in high-risk areas. They, of course, are not adjusting premiums in real time, but that is simply because it is so much easier to work with a flat fee.
The reason for this is that they don’t have the systems to implement incremental real-time changes to the premiums based on the choices individuals are making. The whole point about AI-driven insurance is that it holds the promise of making these adjustments as easy to implement as a flat fee, while allowing a more commensurate balancing of risk and premium.
If this sounds wildly unreasonable, it is nevertheless exactly the kind of personalised minute-by-minute adjustments of motor premiums that the McKinsey report predicts will be commonplace by their 2030 target date. The three authors of the report, Ramnath Balasubramanian, Ari Libarikian and Doug McElhaney, argue that the flood of data coming in from the IoT ‘will allow [insurers] to understand their clients more deeply, resulting in new product categories, more personalised pricing, and increasingly real-time service delivery’.
They give an illustrative instance from life insurance, where, to quote them, ‘a wearable that is connected to an actuarial database could calculate a consumer’s personal risk score based on daily activities, as well as the probability and severity of potential events’. Developments like this would truly be a game-changer.
Rise of the machines
Life insurance companies are already looking at wearables with interest, and insurtechs are working hard to bring systems to market. Wearables can collect a great deal of information on a person’s state and immediate surroundings; instead of using statistical, historical data, AI looks set to introduce a shift across the insurance sector to models that operate using real-time data to price risk on the fly.
The second trend, robotics, will impact travel in obvious ways, and will be something insurers will have to consider. One can imagine that if self-driving autonomous vehicles become pervasive around the world, travellers choosing to travel in cars driven by other humans may well be regarded as putting themselves at higher risk, generating a commensurate real-time adjustment of the policy price.
The ‘open source’ and easy connectivity to external databases is already huge for insurance. One of the major ways external databases help insurers is by providing them with an alternative to having to ask people to fill in tedious forms when applying for products. The statistical information held in these external databases makes it easier to profile specific categories of risk. The ability to access external databases is already proving fertile ground for some insurers – more than one insurer is now offering immediate payouts for delayed flights using third-party information that will instantly confirm which flights have been affected.
Then there is the fourth McKinsey trend, namely advances in cognitive technology. This trend is already alive and kicking in the shape of chatbots, powered by natural language comprehension and driven by cutting-edge AI software. One of the primary innovations driving cognitive technology is what the industry calls convolutional neural networks, also called ConvNets or CNNs. The CNNs’ ‘partner in crime’ is the Recurrent Neural Network, or RNN. Working together, these two deep-learning technologies are providing some staggering depth to chatbots and to image recognition. Explaining these technologies would require a couple of features in their own right, but suffice it to say that they are now pervasive techniques driving massive advances in natural language processing – and insurers are already exploring ways of using chatbots to facilitate a more rewarding relationship with customers.
There are, in fact, a couple of insurtech startups in this space that are doing some very interesting things. Lemonade, for example, focuses on home insurance, but its optional add-on Anti-Theft Package crosses into one of the niche areas of travel insurance by providing compensation for valuable items wherever in the world they are lost or stolen. Purchases are all driven via an AI bot and the purchase is done in seconds without the need for any paperwork or phone calls. Claims are made by opening the app and answering some bot-driven questions. One-third of all claims are paid in 30 seconds, according to Lemonade.
Phoebe Hugh, Founder of the insurtech startup Brolly, claims that her company is already massively disrupting the life insurance market by using AI to get rid of complex policy documentation. A study by KPMG found that the average length of time it takes to read an insurance policy document is around 30 minutes, and many potential buyers either give up or buy ‘sight unseen’ without bothering to read all the qualifying clauses. Small wonder then, that this leads to problems and reputational damage at the claims stage.
What Brolly does is similar to Lemonade, and crosses over into travel insurance for the same reason. It focuses on contents insurance and allows consumers to choose the items, up to a value of £10,000, that they generally have on their person when they are away from home. All the user has to do is take photos of any expensive items, be it a watch, laptop or handbag, on their smartphone, and upload them to the app. Brolly uses image recognition to identify these, and the consumer gets a quote in under 60 seconds. One real innovation Brolly introduces, which kind of parallels the situational pricing McKinsey predicts in its report, is that the monthly premium declines month on month for no claims, up to a maximum discount of 25 per cent.
These are just a few examples of the kinds of transformations that AI is making. They affect everything, from product innovation and distribution through to back-office processing, claims analysis, fraud detection and payouts to customers – and these are just the transformations we already know about. What lies over the horizon as insurtech charges on will, in all probability, be eye-popping. The possibilities are endless. But, with the insurance industry hit hard by Covid-19 claims, especially travel insurers, which have been on the receiving end of millions of cancellation claims, will the current lower level of investment in insurtech be sufficient to keep customers happy? Bottom lines are under pressure, but insurers that choose not to invest in the future of their company now could end up paying a heavier price in five years’ time when the travel market is hopefully back up and running as normal.