Futurist and fintech entrepreneur Lex Sokolin explains the difference between automation and artificial intelligence (AI), and how AI is transforming the insurance value chain—from chatbots to claims.

Highlights

  • Automation is the process of translating human process to machine process. It’s programmed from the top-down, with a known workflow and known outcomes.
  • Artificial intelligence (AI) is the digitization of human intelligence to machine intelligence. It requires extensive data, fed into a mathematical algorithm, in order to create correlations between thousands of different parameters. It enables decisions at scale, but unlike automation, the workflow and outcomes aren’t known.
  • Sales and claims agents are two examples within the insurance value chain where automation and AI could be applied. However, in general, chatbots haven’t been very effective at replicating human interactions.

How AI is transforming insurance, with Lex Sokolin

Welcome back to the Accenture Insurance Influencers podcast, where we ask industry leaders about trends and technologies shaping the future of insurance: self-driving cars, fraud-detection technology, and customer-centricity.

Lex Sokolin is a futurist and fintech entrepreneur. In our last episode, he explained why trends in banking and wealth management could hold valuable lessons for insurers, especially when it comes to working with—not against—insurtechs. In this episode, Lex dispels some myths about AI, and looks at how AI could be applied to the existing insurance value chain.

The following transcript has been edited for length and clarity. When we interviewed Lex, he was the global research director at Autonomous Research; he has since left the company.

I feel like a lot of people confuse AI––the application of AI––with the application of automation. Can you highlight the difference between those two?

If you think about digitization or automation, for me that splits into two types of vectors. The first one is human process to machine process, something that a person does in a manual way, in a workflow. Take that and put it into software.

We have experience with this every single day: think about going into Excel and typing in a mathematical formula. You’re defining a ruleset according to which software will compute something. Or you go one step further and say, “Let’s build software for account opening.” Instead of a human being coming into the office and filling out paperwork, you can capture that on a mobile platform.

That taking of that data and filling out forms, that’s all programmed top down. We know what the workflow is. It’s sufficiently simple for us to sketch it out and turn it into “if this, then that” rules, and then the outcome is fully deterministic from where we started and what kind of data we added. We know how it works. We can reverse engineer the code and understand what’s happening very easily.

The second way you can have digitization is from human intelligence to machine intelligence. And within machine intelligence, there are different approaches to creating outcomes that feel like intelligence, that feel like there is an element of judgment to it. The one that’s popular right now is machine learning enabled by mathematics called neural networks.

What neural networks do very well is solve a problem in a probabilistic way to create an intuition for what something is. If you’re a human being looking at a picture of a cat, you know that it’s a cat and not a dog, and there’s a process in our brain by which that happens. The picture of a cat has no meaning to a computer unless it is transformed into data. You need millions of versions of that data cat to be aggregated and fed into a mathematical algorithm that’s able to create correlations between thousands of different parameters in order to say, “this is more likely to be a cat” or “this is more likely to be a dog.”

AI is still software. It’s still a tool, but the foundational piece isn’t the top-down logic of “if this, then that.” The foundational piece is massive data sets on which the software sits, or is trained, and those foundational data sets came out of the Internet. And once you have those data sets, you’re able to apply these different mathematical algorithms on top. You can essentially put into a ruleset how a person would make a judgment, and then you can lift out that judgment and you can plug it in into a software process. And so now, at scale, you can do things like make decisions on whether somebody should be getting more credit and get their next loan. And with every new piece of information you update that.

A lot of this came out in advertising. Amazon is very good at giving you suggestions about what you should buy next, and Netflix and Spotify know your tastes in video and music in the same way. And in insurance, there are lots of different ways that AI can be used at the manufacturing layer, at the operating claims layer, at the portfolio management layer, as well as within customer distribution.

So, two very different worlds. Automation is the, “if this, then that” command, a Soviet Union central planning world, where you define all the outcomes that are deterministic. And then the AI world is probabilistic, based on existing data that you train the neural networks on, and it’s much more like codifying a human intuition and then deploying it at machine scale.

Finally, one of the things that plagues the thought leadership in this space is painting with a very broad brush. Pictures of cyborgs and various network diagrams to make it feel futuristic. This stuff [AI], end of the day, is all just a set of human tools that people developed in order to be more effective, in order to scale their thinking and simply do more. Even though it sounds threatening or very ambitious, I don’t think AI is any different than the invention of the cloud, or electricity, or the wheel or fire or language—or any of these foundational things in human development.

So when it comes to AI, which solutions, or types of solutions, are the most mature?

I’d say that the most mature parts of artificial intelligence are the ones that are being built by the big tech companies. The big tech companies are motivated to recreate a lot of the human senses. They want to figure out how to provide products and services to people in a way that’s intuitive and is selected by the people in those platforms.

What I mean by that is the sense of vision, the sense of hearing, the ability to create speech. Those are things that are very mature in terms of the technology itself, how well-trained the networks are and what data is available for that training. When you think about self-driving cars, that’s also a version of machine vision.

Those are the mature technologies and in large part because they’ve been built by the big tech companies, whether in the west or in or in the east. When you take that and apply it to the financial services industry and to insurance, it’s not a surprise to see the stuff that’s being used the most to be the one that most aligns with human ability.

That makes sense. So how do you see AI being applied to the traditional insurance value chain? In distribution, for example?

If you think about insurance sales agents, what is it that they do? Well they have a role of being physically present where a client is. You can think of that as almost a billboard for the financial product, and in the US alone I think there are 370,000 insurance sales agents—so there’s a role for AI there. How do I find the client? How do I get to where they are? Artificial intelligence can help you figure out, based on preferences and browsing history and so on, where your customer segment lives.

The second step from that is taking the customer and engaging them in some sort of conversation. In the physical world you might have a person that comes to your home or goes out to a site to do an assessment. In the digital world, the phone is your attention platform.

So that’s really important to wrap your head around because there are only 5 to 10 seats on the phone for financial apps. Whereas in the physical world you can have as many branches as you want, and you can send out as many people as you like—in the mobile world there’s only five seats you can take. It’s extremely important for financial incumbents to figure out how to live inside those attention platforms and have attributes that are native to those attention platforms.

Chatbots are one of those things. (And for me, chatbots and voice are essentially the same.) Chatbots as things that live either inside the phone as a standalone app, or that live inside something like Facebook Messenger as a standalone bot. If you think about downloading Lemonade or Leo or something like that and being able to communicate within the app, that’s just a native feature of how you should build your customer service function. We’re in a world where most of the attention sits with the big tech companies, and not with billboards or other sorts of traditional media advertising. So that’s massively important.

Of course, the caveat is chatbots haven’t been very effective at replicating a human interaction. It’s really tough to find the line between the human and the machine, and the negotiation of that line is where you can make or break the customer experience. If you have a customer that’s coming into your app and trying to discuss something with your chatbot and it’s a frustrating experience and they’d rather talk to a person, you’re definitely going to lose them. And if you don’t have an easy way to push out of that conversational flow into a human channel, again you’re just going to lose that customer.

And then in other cases, as well as according to generational lines, you might have a much better experience with the customer that is able to get onboarded through the phone, is able to get underwritten through a phone, is able to take a picture of their passport to get through Know Your Customer and Anti-Money Laundering compliance (KYC AML), or is able to take a picture of the damage to their car and push that through to the insurance company or for claims assessment.

There’s definitely a negotiation between how frustrating it is to work with a chatbot versus how nice it is to be able to do these things automatically and quickly. And I think that’s still being discovered or explored. I’d say we don’t have a final answer there yet––in part because the underlying technology still has a lot of room to go.

Amazon Alexa and Google’s AI assistants are still in their very, very rudimentary stages, and I would I would expect the next 10 years to be these platform shifts where the big companies compete for being able to do conversation well. So that’s the first piece––insurance sales agents and the role that they play.

I’d also flag the claims process. Within claims processing, there’s about 250,000 people, so the magnitude is also quite large. And then if you look at underwriting headcount and folks that work on the models, you’re getting to about 280,000 people. There is an equal amount of opportunity for automation using this technology and all of the different parts of the value chain.

I love your description of the phone as an attention platform. Thank you very much for taking the time to speak with us today, Lex. Some really interesting things to go away and think about.

Wonderful, my pleasure.

Summary

In this episode of the Accenture Insurance Influencers podcast, we talked about:

  • The difference between automation and AI. Automation is a case of “if this, then that,” where outcomes are well-defined and understood. AI is a probabilistic result from a trained neural network, deployed at machine scale—where the outcomes may be unexpected.
  • AI could be deployed as chatbots to interface with customers as insurance agents do today; however, there is work to be done to improve how chatbots replicate human interactions.
  • Claims and underwriting are other points in the insurance policy lifecycle where there could be opportunities to deploy AI.
  • In the digital world, the smartphone is an “attention platform” with limited real estate for financial apps. Insurers would be prudent to figure out how to live inside attention platforms.

For more guidance on AI in insurance:

In the next episode, Lex will discuss the ethics of AI. How does bias creep into AI-driven decisions and what can insurers do about it? Finally, given the big topics we’ve covered in this series—disruption, innovation, insurtech and AI, to name a few—what can incumbent insurers do to remain competitive without sacrificing shareholder value?

What to do next:

Contact us if you’d like to be a guest on the Insurance Influencers podcast.

 

Submit a Comment

Your email address will not be published. Required fields are marked *