Ankur Goyal: A More Productive Future for Humans + AI

Okay. Can everyone hear me? Perfect. Okay. So nice to meet you all. My name is Ankur. I'm the founder and CEO of Empira, and we're a company that helps other companies use machine learning to automate and analyze data and documents. Um, so we help more than 10,000 customers do stuff like automate the underwriting process, um, automate claims management, automate invoice processing, automate customer service requests, just all kinds of crazy stuff. And insurance is a really big area of focus for us, especially because there are some really mundane parts in, you know, the claims process and the underwriting process specifically. And, you know, this talk is not going to be too much about Empira. It's more just thinking about the future. And as an AI company, we do spend a lot of time thinking about the future and, you know, AI's role in it and our role in that. So let's set the stage for the future. It's 2032. Are we living in The Matrix? Are there self-driving cars? What's going on? So let's start with self-driving cars. Cruise was founded in 2013, and it was one of the first companies to really set off the self-driving revolution. I live in San Francisco, so I see some of their cars floating around, but it's still not really in production yet. I think by 2032 though, we will have self-driving cars. They will be in production. And it'll be thanks to AI technology, but also better city planning and road engineering, which a lot of cities are working on right now. So another question is, are we going to talk to customer service bots or are we going to talk to people when we have questions? And by 2032, conversational AI will be good enough so that by default, you're going to talk to some kind of customer service bot, if at all. But at the same time, you know, the underlying technology is going to get better too. United will have a better app, so hopefully you don't have to call them and complain as much as you do. Planes are getting faster, trains are getting faster, logistics are getting better. And so a lot of the customer service requests that we have today hopefully don't exist in the first place. And then an issue I think that's probably near and dear to our hearts in this room especially is, are things like, you know, lost run reports, Accord forms, invoices, are they still floating around? And I think the answer is no, or at least a human isn't going to be looking at them. By 2032, computer vision and NLP technology, which is already good but getting better every day, will be mature enough so that people don't have to handle paperwork on a day-to-day basis. However, there will be a lot of standardization too. And, you know, maybe you don't need to generate a PDF file for a lost run report or fill out an ACORD form because you have lots of great IoT data that answers the same questions in the first place. So, I think there's some common threads here that are worth thinking about. The first is that AI will become the default for a lot of things that have humans in the loop today. For example, we're the first line of defense for a vast majority of tasks in customer experience and data entry, just because AI isn't good enough yet and it's not deployed widely enough yet. So, by 2032, I think we'll be off the critical path to a large extent. However, AI will still need human input when things go awry. Why? And I know this has been a topic of significant debate recently, if people are following the Google stuff, for example. So this is subjective to some extent, but modern AI is really good at perception and not reasoning. One way I think about it, for those that maybe have been following the NBA Finals, which just ended, there are a lot of sports people in the media that watch the games and then try to explain what players are thinking about. And that's actually a pretty good model for what the capabilities of AI are in the first place. It's able to watch stuff and explain it, but it'll never really be the player on the court that's figuring things out in the first place. And the last thing, which I think about a lot, is that AI is really a tool to allow you to iteratively make things better. So big ambitious technology innovation is actually often at odds with AI. Self-driving cars would be a solved problem by now if roads were designed to support them in the first place. United wouldn't need so many customer service reps if their airline, you know, got in on time in the first place and their app didn't suck. Sorry if anyone works there. I had a really bad experience on my honeymoon a month ago. And again, you don't like, you know, there are all these questions on Accord forms, maybe you don't need to answer them in the first place if you have IoT sensors and other incredible data sources that can answer a lot of these questions to begin with. And so these things are sometimes at odds. And another question I think about is, if you call on the phone to buy something and you have to read your credit card number over the phone, should that be solved by AI or should it be solved by Apple Pay? Well, which solution is actually better? So all three of these use cases at their core are really about communication. And right now, there are things we communicate with each other through, you know, documents as businesses, or peers on the phone, or even drivers on— as we're driving around, sometimes passive-aggressively. These things are hard to automate because standardization is really, really, really, really hard. I probably don't need to convince anyone in this room of how hard standardization is. And unfortunately, when standardization doesn't exist, we're left communicating, you know, our credit card numbers on the phone or emailing each other court forms or lost run reports to try to get our work done. So historically, information that we communicate requires data entry. And I actually think this is a really great application. It's certainly our North Star at Empira is thinking about how AI can solve the data entry problem. Some forms of data entry are very, very obvious. For example, typing in a court form submission into 5 different policy management systems. That's, you know, obvious and painful. Some of them are more abstract, like doing sentiment analysis or doing data entry on a steering wheel to drive your car. I think a good framework is, you know, you're doing data entry in 2022 if you're sitting at your computer or in your car and thinking, you know, why am I doing this? I shouldn't have to do it. And again, at Empira, we really think about the data entry problem today with respect to things like documents, but we have very big ambitions for the future to think about, you know, the overall scope of the problem and the new problems that'll arise in the future. Okay. So now let's keep thinking about the future and let's say data entry is automated. Where does that leave us? What, you know, what are we doing? Well, I think every year iteratively we'll be doing less and less data entry. And that means, you know, less punching data from Accord forms and more time thinking about how to build a better user experience or how to actually analyze the data. You know, some people were talking about that earlier, like how do I actually make use of the data in the first place? I actually can't think of a single customer of ours, Jason, works with me, correct me if I'm wrong, who actually hires people specifically to do data entry. All of our customers, you know, where people are doing a lot of data entry, they're actually really smart, creative professionals, and that's the worst part of their job. They actually want to be doing that other stuff. And so yes, I think there will be less data entry work to do, but I don't think, at least in this industry, that's going to put a lot of people out of jobs. Instead, I think it's going to result in more time to build better user experiences, make smarter decisions with data, and so on. So by no means do— oh, this slide we removed. There we go. By no means do I think that every job is safe. I think eventually Uber drivers, some folks in customer service, data entry professionals, and so on are going to have to find something else to do. And I don't have all the answers here, but I can share some anecdotes about what we've seen at, at Empira. Um, you know, a company with 10 very overloaded operations people might sign up for our product, uh, online, um, implement it for 1 or 2 use cases and start saving a bunch of time. And what we've seen is there's a mentality shift, um, among these folks to sort of elevate themselves from doing back office data entry to try to make a deadline into being the professionals on their team who are actually able to help their businesses scale more efficiently. And so we're very excited about basically enabling and upskilling people to be the forces that can drive their companies to be more efficient. And finally, you know, tying it all together, we now live in a world where AI can solve this data entry problem. The types of data we're going to automate will change over time, but the factor that drives it, I think, is always going to be the same, which is a lack of standardization. AI is a way of technology— sorry, of allowing technology to function where standardization has not caught up or may not even be possible in the first place. And I think that's always going to be the case because we are human after all. That's it. Thank you very much.