Paul Ford:
Everyone should stay really focused on incrementally using machine learning and AI to solve really basic problems. You thread enough of those together, it looks transformation.
James Kotecki:
This is Machine Meets World, Infinia ML’s ongoing conversation about artificial intelligence. I am James Kotecki, and my guest today is Paul Ford, the CEO of Traffk. Welcome to the show.
Paul Ford:
Hey, thanks so much, James. Pleasure to be here, and I look forward to our conversation.
James Kotecki:
Traffk, as I understand, it is using AI in the context of insurance — health insurance, life insurance. So what are you doing with AI that other companies in your industry are not?
Paul Ford:
Quite frankly, as you know, insurance companies have a ton of data, but what’s a little known secret is it’s rarely used amongst all the different departments. We like to think of ourselves as incrementalists, rather than innovators. So we just say, “Hey, let’s start to make use of these vast amounts of data.” We’re all about creating and designing new innovative insurance products and launching it to market with our insurance carrier reinsurance partners.
James Kotecki:
So is the idea that now people will be able to get insurance at better rates that they couldn’t before, or maybe people that couldn’t even get certain types of insurance will be able to get it now, just because you have the data to be able to provide them with an accurate assessment of the risk?
Paul Ford:
Absolutely, so more competitive rates because largely we’re using AI and machine learning today to replace and automate several functions in the supply chain of manufacturing and insurance products that are manual, high overhead, lots of labor involved and automating it, and using data as a proxy. Typically in life insurance, you might have to have a medical exam, they run blood, urine samples and collect some lab results, push you over to an underwriter. They kind of look through this PDF form and figure out what kind of rate they should give you based on your health. We replace all of that framework with just a digital proxy pulling in real-time APIs and pushing that data forward and automating those explicit lab values and returns directly into the decision-making of underwriting.
James Kotecki:
So instead of a person squinting at a PDF, it’s now an AI assessing patterns across data and making similar decisions that a person used to make.
Paul Ford:
Of course, yeah. There’s underrepresented, underinsured, non-insured segments of the population that we can now feature engineer product around with a more intelligent eye.
James Kotecki:
Do you have a sense of the head count that you have been able to replace so far or could theoretically replace?
Paul Ford:
We don’t believe that AI and machine learning will lead to a dystopian future where agents don’t exist or underwriters just don’t exist. We actually believe that their roles will be augmented by new tools and processes, and maybe there’s new roles that are created. 80% of written premiums in insurance is produced by agents. New tools and things of that nature doesn’t affect head count as much as it does produce more opportunity.
James Kotecki:
Let’s talk ethics. If I think about an area where AI ethics is stereotypically important, it has to be insurance…
Paul Ford:
Of course.
James Kotecki:
…where people could be approved or denied based on data and AI and black box algorithms that they maybe never get to see. How do you approach that?
Paul Ford:
Yeah. Well, number one, our team cut our teeth on HIPAA-related data, PHI, the most sensitive data known to insurance. Our mission is to make insurance more inclusive and accessible. To an insurance regulator, they love to hear that, right? Because the concept of AI ethics, machine learning being used for bad instead of good means, oh no, more people will be discriminated against. In the cases of providing insurance, certain demographics or geographies might get redlined. In our case, we’re talking about a situation where typically an insurance company might have five classes of rates, whether smoker, non-smoker, preferred. Whereas with the kinds of machine learning we’re doing in applying pricing mechanisms and creating a spectrum, you can go from having five classes to 20, which means you can actually price risk more appropriately and smooth it to make it more equitable across more people.
James Kotecki:
And do you have a monitoring system in place for how you catch potential unintended biases that might creep into the system?
Paul Ford:
We have several models that we run to kind of sweep and look for biases in the model. There’s many insurance hypotheses that we run to make sure that we’re not running afoul or going down some rabbit hole. There’s plenty of insurance regulation and legislation that literally will say you can’t do X. So we kind of make sure our models aren’t necessarily moving around it to get to the same effect. There’s prohibitions around certain kinds of data being used for underwriting. We make sure that we have data proxies that are comparable. They don’t run afoul of the insurance regulators, where they could still be used in the future safely.
James Kotecki:
And how has COVID changed the data that you’re getting, and has it changed the way that you’ve built or deployed your models for evaluating the data?
Paul Ford:
COVID hasn’t changed what we do, I think if anything, it’s led to more adoption of what we do. One of the effects of COVID, in the life insurance process, you need to go see a nurse or a doctor, go to a lab, have your blood, urine sample ran, and we’ll collect that information and process it. Once coronavirus hit, there was social distancing, labs were shut down, nurses weren’t going into people’s homes. That whole thing was just broken, which led to an extreme backlog of insurance applications for many carriers. So you’re seeing this doubling, tripling effect of a wait time to get an insurance policy. We actually use these digital proxies to replace that physical ask for data by reaching out to various databases and partners in a compliant manner to pull in comparable data. We’ve been able to help some of our carrier partners clear their queues and backlog by using these digital proxies. So we’re starting to see a lot more adoption of that by various insurance companies.
James Kotecki:
What are the next logical obvious places for AI to take insurance?
Paul Ford:
Typically there’s no end-users involved with the creation of the product and design. How do people understand which products fit the right person? We’re working on delivering — we think the rest of the industry will do it too — instead of having 20 different products with different kinds of naming conventions, we think it will actually run under one brand. And then AI is used to guide that individual through a series of questions that essentially Plinkos them into the right policy without having to say, “Oh, did you want this plan?” They have no clue. So you can kind of guide that and create a more profitable scenario, a better fitting financial product for the individual, competitive pricing. We think that’s a winner. So we think that’s going to end up happening in the industry with the more advanced use cases of AI.
James Kotecki:
So everybody’s on the same airplane. They’re paying different prices for their seats and their seats might be at different parts of the plane, but it’s all under the same brand of the same plane, so to speak.
Paul Ford:
Same brand, same plane, and we’ll get you to the same place based on your situation and what party you’re going to.
James Kotecki:
Are there other AI trends that excite you or especially scare you that you think about?
Paul Ford:
One of the trends that scares me still today is just that no one really knows what AI is, we’ll call it the layperson. Even many data scientists, especially insurance actuaries, have no clue what AI is or machine learning. The resistance to understanding it or the overwhelming feeling of “wow, it’s going to change everything.” Everyone should stay really focused on incrementally using machine learning and AI to solve really basic problems. You thread enough of those together, it looks transformational. That’s one of our design and guiding principles of using AI and machine learning. Without that, I think people start to take miscalculated steps in the wrong directions, which could set things back a little bit.
James Kotecki:
So before, you said you were an incrementalists, but it sounds like that’s in service of being transformative.
Paul Ford:
Absolutely. Absolutely.
James Kotecki:
Fill in the blank for this question: Future AI historians will look back at 2021 and say, “I can’t believe they blank.”
Paul Ford:
I can’t believe they made it through 2020.
James Kotecki:
Fair enough.
Paul Ford:
In all seriousness, 2020 did bring about catalyzation of the use of AI and machine learning against different data proxies. I think they’ll say, I can’t believe they got through 2020, and they were able to leverage in a matter of months the same effect that it probably would have taken six years. So I think that’s what AI historians will say when they look back.
James Kotecki:
On your Twitter bio, it says, I’m going to read this, “Ben Sisko’s alter ego,” which I assume is a reference to a Star Trek: Deep Space Nine?
Paul Ford:
Oh, most definitely. Most definitely.
James Kotecki:
Does that — does your being a fan of that show influence the way you do this job?
Paul Ford:
It really does. Ben Sisko, for all intents and purposes, was a Starfleet captain that they sent to the edge of the galaxy and they had no idea what was next. So a lot of what I do, in at least my business life and sometimes my personal life, I’ll go to the edge and say, “Hmm, I wonder what’s beyond that?” So not afraid of the unknown, love ambiguity, it allows you to write your own rules and dictate your own destiny.
James Kotecki:
Paul, anything else you want to bring up or just want to talk about when it comes to AI here?
Paul Ford:
The fear that exists within actuarial, it’s really just the unknown. We did a lot of work with the Society of Actuaries using their tables, enhancing them with more data, and when we start to show that, and the convergence of the two, that seemed to be an unlock. I think that’s where you’ll see from the Society of Actuaries, more use cases and effort of making sure there’s an awareness of actuarial science, machine learning and AI having a world that exists together. That’s super exciting and I think that’s where you’ll see further catalyzation over the next few years.
James Kotecki:
Well, Paul Ford, the CEO of Traffk, thanks so much for joining us today here on Machine Meets World.
Paul Ford:
Hey, it’s been fantastic. Lots of fun and thank you so much, James.
James Kotecki:
And thank you so much for watching and or listening. You can always email the show with your thoughts and suggestions. It’s mmw@infiniaml.com. Please like this, share this, give the algorithms what they want. I am James Kotecki, and that’s what happens when Machine Meets World.