How AI is altering insurance coverage, with Lex Sokolin (podcast)

Futurist and fintech entrepreneur Lex Sokolin explains the distinction between automation and synthetic intelligence (AI), and the way AI is remodeling the insurance coverage worth chain—from chatbots to claims.

Highlights

Automation is the method of translating human course of to machine course of. It’s programmed from the top-down, with a identified workflow and identified outcomes.
Synthetic intelligence (AI) is the digitization of human intelligence to machine intelligence. It requires in depth information, fed right into a mathematical algorithm, as a way to create correlations between hundreds of various parameters. It permits selections at scale, however not like automation, the workflow and outcomes aren’t identified.
Gross sales and claims brokers are two examples throughout the insurance coverage worth chain the place automation and AI might be utilized. Nonetheless, basically, chatbots haven’t been very efficient at replicating human interactions.

How AI is remodeling insurance coverage, with Lex Sokolin

Welcome again to the Accenture Insurance coverage Influencers podcast, the place we ask trade leaders about developments and applied sciences shaping the way forward for insurance coverage: self-driving vehicles, fraud-detection know-how, and customer-centricity.

Lex Sokolin is a futurist and fintech entrepreneur. In our final episode, he defined why developments in banking and wealth administration might maintain helpful classes for insurers, particularly in relation to working with—not towards—insurtechs. On this episode, Lex dispels some myths about AI, and appears at how AI might be utilized to the prevailing insurance coverage worth chain.

The next transcript has been edited for size and readability. Once we interviewed Lex, he was the worldwide analysis director at Autonomous Analysis; he has since left the corporate.

I really feel like lots of people confuse AI––the applying of AI––with the applying of automation. Are you able to spotlight the distinction between these two?

If you concentrate on digitization or automation, for me that splits into two forms of vectors. The primary one is human course of to machine course of, one thing that an individual does in a guide means, in a workflow. Take that and put it into software program.

We now have expertise with this each single day: take into consideration going into Excel and typing in a mathematical formulation. You’re defining a ruleset in keeping with which software program will compute one thing. Otherwise you go one step additional and say, “Let’s construct software program for account opening.” As an alternative of a human being coming into the workplace and filling out paperwork, you possibly can seize that on a cell platform.

That taking of that information and filling out types, that’s all programmed high down. We all know what the workflow is. It’s sufficiently easy for us to sketch it out and switch it into “if this, then that” guidelines, after which the end result is totally deterministic from the place we began and how much information we added. We all know the way it works. We are able to reverse engineer the code and perceive what’s taking place very simply.

The second means you possibly can have digitization is from human intelligence to machine intelligence. And inside machine intelligence, there are completely different approaches to creating outcomes that really feel like intelligence, that really feel like there is a component of judgment to it. The one which’s widespread proper now's machine studying enabled by arithmetic referred to as neural networks.

What neural networks do very properly is resolve an issue in a probabilistic technique to create an instinct for what one thing is. For those who’re a human being an image of a cat, you already know that it’s a cat and never a canine, and there’s a course of in our mind by which that occurs. The image of a cat has no that means to a pc until it's reworked into information. You want tens of millions of variations of that information cat to be aggregated and fed right into a mathematical algorithm that’s capable of create correlations between hundreds of various parameters as a way to say, “that is extra prone to be a cat” or “that is extra prone to be a canine.”

AI continues to be software program. It’s nonetheless a software, however the foundational piece isn’t the top-down logic of “if this, then that.” The foundational piece is huge information units on which the software program sits, or is skilled, and people foundational information units got here out of the Web. And after getting these information units, you’re capable of apply these completely different mathematical algorithms on high. You'll be able to basically put right into a ruleset how an individual would make a judgment, after which you possibly can raise out that judgment and you'll plug it in right into a software program course of. And so now, at scale, you are able to do issues like make selections on whether or not someone needs to be getting extra credit score and get their subsequent mortgage. And with each new piece of knowledge you replace that.

A whole lot of this got here out in promoting. Amazon is superb at providing you with ideas about what you can purchase subsequent, and Netflix and Spotify know your tastes in video and music in the identical means. And in insurance coverage, there are many completely different ways in which AI can be utilized on the manufacturing layer, on the working claims layer, on the portfolio administration layer, in addition to inside buyer distribution.

So, two very completely different worlds. Automation is the, “if this, then that” command, a Soviet Union central planning world, the place you outline all of the outcomes which can be deterministic. After which the AI world is probabilistic, based mostly on current information that you simply prepare the neural networks on, and it’s far more like codifying a human instinct after which deploying it at machine scale.

Lastly, one of many issues that plagues the thought management on this house is portray with a really broad brush. Photos of cyborgs and varied community diagrams to make it really feel futuristic. These items [AI], finish of the day, is all only a set of human instruments that individuals developed as a way to be more practical, as a way to scale their pondering and easily do extra. Although it sounds threatening or very formidable, I don’t suppose AI is any completely different than the invention of the cloud, or electrical energy, or the wheel or fireplace or language—or any of those foundational issues in human growth.

So in relation to AI, which options, or forms of options, are essentially the most mature?

I’d say that essentially the most mature components of synthetic intelligence are those which can be being constructed by the massive tech firms. The large tech firms are motivated to recreate numerous the human senses. They wish to determine find out how to present services to folks in a means that’s intuitive and is chosen by the folks in these platforms.

What I imply by that's the sense of imaginative and prescient, the sense of listening to, the flexibility to create speech. These are issues which can be very mature when it comes to the know-how itself, how well-trained the networks are and what information is on the market for that coaching. When you concentrate on self-driving vehicles, that’s additionally a model of machine imaginative and prescient.

These are the mature applied sciences and largely as a result of they’ve been constructed by the massive tech firms, whether or not within the west or in or within the east. While you take that and apply it to the monetary companies trade and to insurance coverage, it’s not a shock to see the stuff that’s getting used essentially the most to be the one that almost all aligns with human skill.

That is sensible. So how do you see AI being utilized to the standard insurance coverage worth chain? In distribution, for instance?

If you concentrate on insurance coverage gross sales brokers, what's it that they do? Effectively they've a task of being bodily current the place a consumer is. You'll be able to consider that as nearly a billboard for the monetary product, and within the US alone I feel there are 370,000 insurance coverage gross sales brokers—so there’s a task for AI there. How do I discover the consumer? How do I get to the place they're? Synthetic intelligence might help you determine, based mostly on preferences and looking historical past and so forth, the place your buyer section lives.

The second step from that's taking the client and interesting them in some type of dialog. Within the bodily world you may need an individual that involves your private home or goes out to a website to do an evaluation. Within the digital world, the cellphone is your consideration platform.

In order that’s actually vital to wrap your head round as a result of there are solely 5 to 10 seats on the cellphone for monetary apps. Whereas within the bodily world you possibly can have as many branches as you need, and you'll ship out as many individuals as you want—within the cell world there’s solely 5 seats you possibly can take. It’s extraordinarily vital for monetary incumbents to determine find out how to dwell inside these consideration platforms and have attributes which can be native to these consideration platforms.

Chatbots are a type of issues. (And for me, chatbots and voice are basically the identical.) Chatbots as issues that dwell both contained in the cellphone as a standalone app, or that dwell inside one thing like Fb Messenger as a standalone bot. If you concentrate on downloading Lemonade or Leo or one thing like that and with the ability to talk throughout the app, that’s only a native function of how it's best to construct your customer support perform. We’re in a world the place many of the consideration sits with the massive tech firms, and never with billboards or different kinds of conventional media promoting. In order that’s massively vital.

In fact, the caveat is chatbots haven’t been very efficient at replicating a human interplay. It’s actually powerful to seek out the road between the human and the machine, and the negotiation of that line is the place you can also make or break the client expertise. If in case you have a buyer that’s coming into your app and making an attempt to debate one thing together with your chatbot and it’s a irritating expertise and so they’d somewhat discuss to an individual, you’re undoubtedly going to lose them. And for those who don’t have a simple technique to push out of that conversational stream right into a human channel, once more you’re simply going to lose that buyer.

After which in different instances, in addition to in keeping with generational strains, you may need a a lot better expertise with the client that is ready to get onboarded by way of the cellphone, is ready to get underwritten by way of a cellphone, is ready to take an image of their passport to get by way of Know Your Buyer and Anti-Cash Laundering compliance (KYC AML), or is ready to take an image of the harm to their automotive and push that by way of to the insurance coverage firm or for claims evaluation.

There’s undoubtedly a negotiation between how irritating it's to work with a chatbot versus how good it's to have the ability to do this stuff robotically and shortly. And I feel that’s nonetheless being found or explored. I’d say we don’t have a ultimate reply there but––partly as a result of the underlying know-how nonetheless has numerous room to go.

Amazon Alexa and Google’s AI assistants are nonetheless of their very, very rudimentary levels, and I might I might count on the following 10 years to be these platform shifts the place the massive firms compete for with the ability to do dialog properly. In order that’s the primary piece––insurance coverage gross sales brokers and the position that they play.

I’d additionally flag the claims course of. Inside claims processing, there’s about 250,000 folks, so the magnitude can also be fairly massive. After which for those who have a look at underwriting headcount and folk that work on the fashions, you’re attending to about 280,000 folks. There may be an equal quantity of alternative for automation utilizing this know-how and the entire completely different components of the worth chain.

I like your description of the cellphone as an consideration platform. Thanks very a lot for taking the time to talk with us at the moment, Lex. Some actually attention-grabbing issues to go away and take into consideration.

Great, my pleasure.

Abstract

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

The distinction between automation and AI. Automation is a case of “if this, then that,” the place outcomes are well-defined and understood. AI is a probabilistic end result from a skilled neural community, deployed at machine scale—the place the outcomes could also be surprising.
AI might be deployed as chatbots to interface with prospects as insurance coverage brokers do at the moment; nonetheless, there's work to be carried out to enhance how chatbots replicate human interactions.
Claims and underwriting are different factors within the insurance coverage coverage lifecycle the place there might be alternatives to deploy AI.
Within the digital world, the smartphone is an “consideration platform” with restricted actual property for monetary apps. Insurers can be prudent to determine find out how to dwell inside consideration platforms.

For extra steerage on AI in insurance coverage:

Within the subsequent episode, Lex will focus on the ethics of AI. How does bias creep into AI-driven selections and what can insurers do about it? Lastly, given the massive matters we’ve coated on this collection—disruption, innovation, insurtech and AI, to call a couple of—what can incumbent insurers do to stay aggressive with out sacrificing shareholder worth?

What to do subsequent:

Contact us for those who’d wish to be a visitor on the Insurance coverage Influencers podcast.