AI and Insurance: Hype or Reality?
Despite having repeatedly failed to live up to the hype, AI’s central place in the future of insurance is secure.
Artificial intelligence has been a hot topic for several decades by now, but the last year or so has seen a huge jump in the public’s awareness of AI as real practical technology rather than mere science fiction.
Behind the scenes and out of the spotlight, however, AI technology has been steadily improving in capability and growing in scope. The insurance industry, for all its gigantic size and reputation for resistance to change, has made considerable strides in incorporating AI in various ways. As we shall see, these efforts have led to mixed results. Nevertheless, AI’s potential is too great for it not to play an ever-increasing role in coming iterations of insurance technology.
Before we get too deep into our discussion, a definition may be in order. So many people talk about AI, but fewer talk about what it is. Microsoft has come up with a fairly simple and clear definition: Artificial intelligence is the capability of a computer system to mimic human cognitive functions such as learning and problem-solving. Through AI, a computer system uses math and logic to simulate the reasoning that people use to learn from new information and make decisions.
The big AI news these days are the free online beta platforms that almost instantaneously produce uncannily human-like results. DALL-E generates images from any text input. I tried “praying mantis riding a bicycle” and “ice cubes on fire” and the results were as impressive as they were creepy. If DALL-E has not yet replaced Getty Images and Shutterstock, it (or something like it) will soon.
Even more newsworthy have been ChatGPT and a couple other similar programs. Their ability to scour the internet in response to the user’s request and create natural language output is nothing less than uncanny. It can even tailor its results to match specific writing styles, not only poetry vs. prose, but even as far as mimicking specific people’s writing styles.
Disruption is a grossly overused buzzword in tech, but if DALL-E and ChatGPT aren’t harbingers of disruption across multiple fields of human activity, nothing is. The precise nature of the changes that AI will bring can be much harder to predict.
The last decade has seen an explosion of startups that, though enormously diverse, fall under the new category of insurtech. Many of them have attempted to leverage artificial intelligence and/or machine learning to streamline one aspect of the world of insurance or another. The goal is typically some combination of reduced costs and increased speed, efficiency, or personalization. Let’s look at some examples and see how they worked out.
One of the larger and better-known insurtechs promised to use machine learning to re-define an old, fragmented industry and focus on prevention and smarter underwriting. The results of these efforts led to a loss ratio of 161%. The company was forced to reduce the number of products it offered and change its distribution models by focusing on indirect ones. Taking these steps succeeded in improving their loss ratio to merely 110%, which remains unsustainable.
Another insurtech trailblazer promised transformational motor insurance pricing using machine learning to price drivers based entirely on behavioral attributes. Technology such as telematics would result in month-to-month pricing variability, yet the traditional attributes of consumer pricing still are largely utilised.
As a final case study, perhaps the best-known of all the insurtechs promised to reduce claims ratios and costs through better analytics and claim automation. However, despite achieving good claims resolution times, both their claims ratios and their operating ratios remain stubbornly in the same range as the traditional insurers with whom the company now partners.
What do all three above cases have in common? A growing use of technology as a basis for generating additional insights, or better processes – but a new insurance company that is still exposed to all the same macro conditions (specifically inflation, NatCat and customer acquisition) as all the largest established players that are focused on direct, agent and digital distribution.
Siri, What Insurance Should I Get?
We can see that trying to shoehorn artificial intelligence into insurance has not been an unqualified success. Does that mean that the project is doomed?
Quite the opposite.
The recent failures are merely pointing us in the right direction, which is away from hype and towards a steady development of substantial and useful capabilities.
The reality is that artificial intelligence will penetrate every single aspect of the insurance industry to offer better products and services to the customers while improving the bottom line for the shareholders. But these changes will not happen overnight and cannot be forced. Full transformation will take closer to a decade than a year, and we need to let the process play out as capabilities continually improve.
Some changes will happen sooner than others. Parametric insurance is one such. Flight delays are a live and real-world example, where no claim even needs to be filed; rather the system automatically informs the traveler of both the delay and the compensation.
A little farther down the line, similar tools will be developed that expand on established use-cases with more complexity, such as home insurance. Imagine being informed by your insurer that it has detected a water leak in your home, assessed the floor damage, called for the relevant repairs, and issued an immediate prepayment to get sorted out.
Eventually, the total interconnectedness of all our devices and activities will reach the stage where all we need to do is say out loud, “Siri (or Hey Google, etc.), what insurance should I get?” And that’s when we will have achieved total AI insurance transformation.
Of course, we can’t forecast where this will end. One important unanswered question is the extent that legislation and regulation will play in determining AI’s role in the industry. In the meantime, we can prepare the data foundation that is the prerequisite for any robust AI system. Doing that requires determining, defining, and activating as many data sources as possible, both within and without the insurer’s databases. This will include learning how to add market-level (public) data at scale to make the models and machine queries ever better.
Even as the technology continues to develop, there is much for us to do so we can prepare for the future and meet it head-on.