From pilots to scale: your roadmap to integrating AI and RPA into the underwriting process

Artificial intelligence (AI) is much discussed in the insurance industry but it is often in the context of small-scale projects, pilots or experimental phases. To make the most of the transformative potential of AI and associated systems such as robotic process automation (RPA), insurers need to bite the bullet and move towards a more integrated approach.

At this year’s Underwriting Innovation Europe Virtual Event, Lewis Z Liu, co-founder and chief executive officer of Eigen Technologies, will join a panel of innovation, insurance and technology experts to discuss how carriers can move their AI underwriting strategies forward to make the most of these innovative processes and technologies.

Why is now the time to be discussing taking AI mainstream?

Insurance, like other financial services businesses, is at a crossroads. There is increasing competition from insurtechs and plenty of room for them to disrupt the insurance model. New players—Bought By Many, for example—are achieving significant investment. It is an existential threat.

Risk dynamics are also changing. Climate, geopolitical and cyber are now much bigger threats than they were before. And finally, carriers and reinsurers are no longer simply relying on long-standing relationships to ensure ongoing business. For a reinsurer to reinsure a package, there is now much more focus on delving into the detail of the deal.

Using AI to transform the business from top to bottom can seem daunting. Can the task be broken down?

AI is just software—there’s input and there’s output. The only difference is that the inputs and outputs are perhaps “messier”. In the real world, the inputs are more complex which is why we rely on human cognition and AI tries to replicate that. Trying to find out levels of cyber exposure, humans need to be trained, but the more they learn, the better they get. The AI is the same.

The challenge is that the AI needs a lot of data to do that learning—tens of thousands of data points. But we feel it can be done with much less: just tens.

The point of departure is what problem do you want to solve. Let’s not do AI for AI’s sake. Challenge it to solve a problem that humans might normally tackle or is very convoluted and you want to simplify. Start with a smaller problem but make sure that it does have wider-reaching implications. Start small but don’t stay small.

Commercial lines carriers often claim that their business is too complex to automate. Is this true?

Commercial lines is probably the best candidate for AI. There are plenty variations in those inputs and outputs but you still have the scale you can get value from.

Commercial can particularly benefit from automated document analysis. There are so many complex pieces of information, either paper-based or via email that need to be fed into the underwriting model and AI can increase both speed and accuracy.

Similarly, there’s lots of value to be had from AI extracting information during the claims process. But we need to get rid of the idea that AI does everything. You still need a human in the loop.

How can we demonstrate the value of AI to the business?

There are several layers of value. In terms of cost savings, carriers can sometimes see a 60 to 90 percent reduction. Then there is workflow consolidation: 300 types of document generate 300 manual workflows and AI brings all that together into one.

The next is accuracy. There is the myth that humans are more accurate than machines. A report from Deloitte in The Actuary magazine suggested that humans are 70 percent accurate—AI can boost that.

Then there is a reduced time to market. Insurtechs are able to create 10 to 100 times faster reactions to changes in the marketplace. And finally, there is better overall decision-making. A particular use case has been in Solvency II in the area of asset liability. We’ve been able to see £7 billion of capital relief using AI.

It’s not just about cost savings. Done well, you also gain strategic return on investment.

What should attendees take away from this presentation?

That in order to scale, the first task is to find the problem and the second is to define the right criteria for success—criteria that are realistic. Do not have higher expectations of AI than of humans. You don’t need 100 percent accuracy in any scenario—humans aren’t always right, so why expect the machine to be?

Once you have real-world success, you can iterate quickly and get the result into production. AI is here and it is necessary for incumbents to fight off disruption from insurtechs.

Lewis Z Liu will be speaking at Intelligent Insurer’s Underwriting Innovation Europe Virtual Event (June 28–30, 2021). The event is free to attend for insurers and brokers/agents, but you must register in advance. Sign up to access the content live and on demand here.