What does AI mean for Insurers

From Vicky, the humanoid in 90’s kids show Small Wonder to Ava, the sentient, Turing test cracking and starkly human looking robot in Ex-Machina, we have always fantasized a world with Artificially Intelligent computers. While the former example was a lighter take on what AI could look like in everyday world, the latter was a more serious attempt to rethink the same concept. But between these two stark contrasts, AI is being effectively used by businesses today to perform several tasks that would have required humans to sweat tirelessly for weeks and sometimes even months.

On a high level, the term Artificially Intelligent can be attributed to any computer or machine that replicates a human behavior. With that broad a definition, even an IVR system is intelligent. After all it tries to do what a human would have done i.e. try to ascertain the purpose of your call and connect you to the right department. However, given the advances that the field has made in the last decade and for the purpose of this article, I will refrain from using such a broad brush to paint the industry practices. The widely accepted and more refined definition of artificial intelligence goes as below.

A system’s ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation.

An average Joe’s life is surrounded by intelligent machines. For example, the YouTube recommendations that pull you into a never-ending spiral of binge that by the time you look at the clock it is already 3am or the innocuous spam filter in your email that quietly works in the background without you ever noticing it. More complex AI algorithms are being employed by businesses to predict stock movements, presence of cancer, real time face recognition etc.

While other businesses have made significant advances in adopting AI in everyday tasks, the insurance industry is still seriously lagging behind in making any meaningful mark in the discipline. In my modest 7 years experience working for large insurers, I have not come across any path breaking implementation of machine learning, deep neural networks or other relevant concepts in real life. Some carriers have realized that they are late to the party and hence are rushing to acquire the necessary skills and teams to build a data analytics and AI practice but their progress is still in nascent stages. Lack of a holistic strategy is what keeps the players from making game changing progress, something that AI is absolutely capable of delivering.

I have tried to list some use cases that are radical but totally worth exploring for establishing the next generation of Insurance products and services.

• Custom risk assessment for accurate underwriting – Since the early days of life insurance, most insurance companies have relied on four parameters for a customer’s risk assessment – Age, Gender, Smoker Status and Occupation. These parameters determine your basic mortality risk, which is sometimes peppered with a few more historical parameters like a health assessment and lifestyle preferences questionnaire. In our opinion, these methods of determining insurance risk is outdated and needs a massive overhaul. Life insurance risk assessment can incorporate so many relevant data points available from wearable devices, lifestyle expenditure patterns and social media feeds for a more accurate and customized profile. Just because I am a 40 years old smoker who works in sales doesn’t mean that the level of risk I pose is the same as another 40 year smoker salesman who regularly runs half marathons, shops at organic stores and is a regular at his gym.
• Remote surveillance after a natural disaster – Imagine the plight of property owners who have been hit by a massive hurricane like Dorian or Katrina. In the midst of all the chaos and loss suffered by the owners, they also have to deal with insurance providers who are not able to reach the premises due to infrastructure collapse. It takes weeks and sometimes even months to get the infrastructure up and running before any kind of inspection can be performed on the property. Some insurers have started using aerial drone footage to do preliminary assessment of the destruction, but it is still limited to mostly manual identification of broken roofs and fallen trees. Deep neural networks are capable enough to identify damaged houses, vehicles, swimming pools
• Intelligent telematics – several insurers have started to invest in solutions and devices that allow them to monitor customer’s driving styles. Though most of these programs are in nascent stages, their fundamental premise is far from leveraging the true capabilities of intelligent telematics. My auto current insurer offers a ‘program’, which when enrolled in, gives me 5% discount on my insurance premium and also rates me on my driving every time I take out my car. However, the system is far from being ‘intelligent’ as the insurer claims it to be. I understand the ‘under-the-hood’ working algorithm that powers the program. Rather than assessing my driving style and marrying it with the driving context, the program only uses an algebraic formula to give me points on my speed, cornering, acceleration and breaking. The cumulative score I get for every driving session reflects the discount that I will be entitled to at the end of the year. Instead of such half-baked algebraic telematics, the new age object recognizing algorithms that power self-driving vehicles can be used to better assess a driver on the road. Things like how long after the front car’s break light went off did the driver apply breaks is a far better representative of the risk drivers pose than a driver doing 5 kmph on a highway.
• Intelligent digital assistant – An insurer provided digital assistant, akin to Google’s assistant or Apple’s Siri, can smoothen out many kinks in customer interactions for which they have to call the contact center. Simple pieces of information like my next premium due date or tasks like updating my address should not require me to pull out my policy documents that are lying in a pile in storage or call a contact center, only to be put on hold for 30 min (true story). Add to that the fact that most life insurance policies require customers to make payments once a year, makes it even more unlikely that customers will reach out. No wonder, majority of adults do not relate with their carriers in the same way they relate to other e-commerce partners like Uber or Amazon. An intelligent digital assistant that is available 24×7 and is intelligent enough to answer basic financial questions will make the customers feel connected with the carriers. Empowering the customers with technology will not only result in a recurring relationship but also impact the bottom line of the company.
• Pay as you use insurance – Life, auto and several other types of insurance are typically renewed after a certain period of time. My life insurance contract is renewed every year when I pay my premiums and my car insurance every 6 months. It doesn’t matter if my car spent 2 months in winter sitting in my garage because I prefer to take an Uber when it snows. I still pay the same amount of premium for my car insurance. On the other extreme, it doesn’t matter if I suddenly decide to take a long road trip across the country to see my college friends. Though futuristic, usage-based insurance products are possible with the amount of data and connected devices that are available. Insurers can develop, with the amount of data that is available with them, models to accurately predict increase or decrease in risks for departure from regular trends in customers’ lives.

I have been working with Insurers for the better part of the last decade and have seen a recent uptick in technology exploration and adoption. A lot of my clients have shown interest in building an AI lab that gives them immediate result on the balance sheet. My recommendations to all my clients have been the same – AI is not something that you can install in a year or two and keep upgrading every 6 months thereafter. It is a continuous journey that requires a fundamental overhaul of how we perceive data, leverage it and keep improving our models for better accuracy and relevance.