Derek Roberti - Conversational Automation, & Natural Language Understanding AI

(Joel Beasley at 00:00:02) Hello, my friends. Today we're talking to Derek, the VP of Technology at Cognigy. And we discuss Cognigy's advanced conversational automation, the future of natural language understanding AI, and how to motivate your team with what they're passionate about. All of this right here, right now on the Modern CTO podcast.

(Derek at 00:00:27) Here we go.

(Joel Beasley at 00:00:28) This is the Modern CTO podcast.

(Derek at 00:00:39) Everyone has their own childhood computer story. So for me, I stumbled into my love of computers on a Texas Instruments computer, which was one that you actually connected to your TV because no one had monitors at home at that point in time. And I learned basic programming, and I basically abandoned all of my friends for six weeks because I was just so engrossed in what I was doing. And since then, I've had a passion for technology. I've been lucky enough to work at a number of technology companies and build good relationships with people in the business.

(Derek at 00:01:20) In my latest venture, I would say prior to my latest venture, I was lucky enough to work at a company called Sitecore. And Sitecore is one of the leading marketing content management and enterprise software platforms out there, and I was one of the early employees here in North America. I stayed with the company for about twelve years, and that's when I met my colleague there. He was our CIO, Philip Heltvig, and he left Sitecore, took a year off maybe, and then started the company I'm at now, Cognigy. And based on our past history, he had started Cognigy in Germany, which is where the company is headquartered.

(Derek at 00:02:00) And when we were ready to get things going here, he gave me a call and we got the Cognigy show going here in the United States. And it's just been a wonderful time to be at another company that has outstanding technology that is not just fun or interesting. It's a great product. We all feel really proud of it. We got a lot of recognition for it.

(Derek at 00:02:25) And the more I get hands-on with the product, it's always fun to spend time with it.

(Joel Beasley at 00:02:30) That's awesome. I mean, it's really important to actually believe in it and really have a passion for what you're doing because people can tell. You can't fake that level of genuine interest.

(Derek at 00:02:43) Yeah. I would say that when I look at different technologies on the market and why I feel like I've just been lucky is that at Sitecore, my previous company, there was a magic to the product. Sitecore was able to develop a really passionate developer community over the years because there were certain things that Sitecore did right that were so powerful and put in the hands of developers. So people really had a passion for that product.

(Joel Beasley at 00:03:15) What were some of those things that they did right that really clicked with the developers?

(Derek at 00:03:21) So it's kind of a funny story. If you go all the way back to the beginning of the company, they early on decided to standardize on the Microsoft platforms. That was when Microsoft introduced the .NET framework as their first-class programming framework that followed some legacy products like Visual Basic and ASP. And they had a vision for how .NET framework would be used and how presentation components would be built out using the .NET framework, which they also had a vision as to how this would fit into Windows as well. What Sitecore did was they built out a rendering engine that worked based on Microsoft specifications before Microsoft had released it themselves.

(Derek at 00:04:13) So they built a lot of technology in anticipation what Microsoft would do. And because they were starting at such a foundational level, they created something kind of ultimately more powerful than the ASP.NET platform on its own. What really, as a developer, this was exciting to me. But what was also really exciting is you saw it with every developer who sunk their teeth into the product—the life cycle was confusion, frustration, questioning, like, "Why does this not work the way I think it should?" And there was always some turning point where the design of the product would click for them, and they'd realize, "Whoa, this is powerful."

(Derek at 00:05:01) And it was that turning point in people's heads that you could see over and over again that built that passionate following behind their product. And at Cognigy, it's the same thing. It's in a different way. What we face is we are a low-code platform for developing conversational automation. And for a business audience, they can intuit why a low-code platform is good.

(Derek at 00:05:30) So convincing people there is easy. For a developer audience, it's a little bit trickier because when you're used to coding everything yourself, the idea of using a graphical tool to do part of your job, a lot of people greet that with skepticism. Like, "How could this tool possibly give me the same level of flexibility I have of coding stuff? Or I'm going to have to do this stuff in the tool because a business is forcing me to, and then I'm going to have to end up writing a lot of code to compensate for the shortcomings of the tool."

(Derek at 00:06:01) So it's essentially going to rob me of productivity instead of enabling me, and creativity instead of enabling me. But you see that same life cycle when developers might have some skepticism about working with a graphical tool, start using the tool. You see how it really clicks in people's minds about how this makes them more productive and the power it puts in their hands. And you'll see people just genuinely have fun working with a product because it's so powerful. And they also see that the role we have for developers in the tool is very harmonious with how the tool works itself.

(Derek at 00:06:40) So you always feel like if you are developing extensions to the product, it's always in service to making the graphical tools even easier to use. So I feel really proud of it, and I think that the company has built something special. And this is in contrast, you know, the startup world is a diverse world. MVP means different things to different people. Sometimes it's something that will barely work, and sometimes it's something that's well-tested.

(Derek at 00:07:11) We are far beyond that stage, and we have a product that's been developed by some of the largest organizations in the world because of its well-roundedness, I think, and the power that it brings to both the developer and the non-developer.

(Joel Beasley at 00:07:27) That's really cool. Before we really get into the Cognigy stuff, I wanted to ask you about one thing I saw in your background. Because I saw you actually went to Stanford Law School, and I was curious, were you planning on going into law prior to technology? Or, because I know you also said you were really into it as a kid, so how did you take that foray into law and then come back?

(Derek at 00:07:51) Yeah, so there's certainly different detours I've taken in my career. I actually started Stanford out of undergrad getting my PhD in a liberal arts program there. I was passionate about learning. That's always been on fire for me.

(Derek at 00:08:13) And doing more learning was something that meant a lot to me. It aligned with how I thought and what brought meaning to my life. What I found is that earning a living has little to do necessarily with what brings meaning to your life. So when I graduated, I realized that working in academia wasn't for me, and the market for academic jobs was not all that abundant. So I thought, "What is a discipline I can go into where learning is valued, intelligence is valued and fostered, but also has something practical at the end of it?"

(Derek at 00:08:57) And that led me to law school. And I was lucky enough to go to law school at Stanford, was surrounded by some of the smartest people I've ever met. It was amazing. But even more, or as amazing, I think, as my peers at Stanford, was the time I was there. So I was there as the .com boom was taking off.

(Derek at 00:09:21) This is the old, original .com boom. And what was fascinating about it is the people who came to our classes were venture capitalists, were people starting new companies or old companies realizing that they had to have a presence online and trying to figure out what that was. So between the law school and the business school at Stanford at the time, there was a tremendous amount of focus and energy around the set of emerging technologies that now is just kind of part of our day-to-day life. We had classes like, I took the first e-commerce class ever offered in a law school, at a time where the legal frameworks around e-commerce were being figured out. All of our classes were infused with a technological angle that really hadn't been there before.

(Derek at 00:10:23) In a normal career path through law school, especially if you're going to one of the top law schools, usually in your second year of law school you do an internship with a law firm that will ultimately hire you after your third year. So it's a very linear path and a very established and comfortable one. For me, in my second year, I already knew I had to do something in the technology space. I reconnected with my passions I was talking about earlier when I was younger.

(Derek at 00:10:54) And I wanted to do something in technology that connected me with what was going on all around me, and I didn't care if it had anything to do with law. It was just too exciting what was going on. So I worked for a company, which actually still exists—it's an online talent marketplace—when it was first getting started. And I had the fortune of being mentored by someone who shared all of his knowledge and experience with me and helped me understand what it meant to contribute in a technology business. So by the time I hit a fork in the road where I needed to decide, "Do I want to stay in law school, or do I want to jump into the tech world?"

(Derek at 00:11:41) I felt like I was too fortunate to be where I was to not finish law school. So I finished law school, went into the technology business, and have not looked back since.

(Joel Beasley at 00:11:52) That's really cool. With that context, it makes a lot more sense as a path that you followed than just looking at "went to law school, and now he's the VP of Technology." But so you mentioned earlier that Cognigy is a low-code platform for conversational AI, correct?

(Derek at 00:12:13) That's correct, yes.

(Joel Beasley at 00:12:15) Cool.

(Derek at 00:12:15) Sometimes we say conversational AI, and sometimes we say conversational automation.

(Joel Beasley at 00:12:21) Okay, yeah.

(Derek at 00:12:23) It's been a shift that we've taken because when the company was first started, it was started by an inspiration of AI back then. It's something that everyone was talking about—we've got to kind of get in on what's going on in the market. And so Cognigy and a lot of companies really centered our messaging, our product development around AI technologies. What we found was that from our perspective, whatever the technology is behind the scenes, whether it's AI or something else, is less important than the outcomes that we're going for.

(Derek at 00:13:07) So we often try to reorient people to say, while natural language understanding is important, while AI is important and deep learning is important and machine learning is important, none of these really point to an outcome. And the ultimate outcome is automating business processes, and it's that twofold benefit of, one, making it easier and faster for customers and employees to access the information and systems they need 24 hours a day. And then for the business, the ability to automate that so they can reduce their costs. We always tell people to focus their human resources on those scenarios where human empathy and human decision-making are important and automate anything where people are essentially order takers or doing data entry or doing simple lookups of information. So we are Cognigy.ai—that's the name of our platform—but conversational automation is really our mission above all else.

(Joel Beasley at 00:14:13) That makes sense. That's a smart way to angle yourselves there. Yeah. So I feel like there's a lot of bias against automating customer service calls or communication in general because people will have a bad experience with a conversational AI in the past. How do you approach that when you're talking with prospects?

(Derek at 00:14:36) It's a really important thing to address. We're talking about a lot of history here, and I always tire of people telling me about the good old days. But some historical perspective is, I think, helpful. If we look early on in the days of web development and web design, even if that's what we would call it, we went from text-heavy pages with just a lot of stuff on them to some basic sense of information architecture and hierarchical navigation. Along the way, tools came up, WYSIWYG editors that let you build web pages and help you overcome some of the technical challenges of getting online.

(Derek at 00:15:22) So there's a lot of pieces that were being assembled. Mobile app and mobile web development kind of followed a similar path, because a lot of times, the first hurdle that's overcome is a technical hurdle. But the subsequent hurdle and the far more important and perhaps difficult one is that of user experience. So we had our fill of websites that had the left nav with a hierarchical representation of the information architecture of a website—

(Joel Beasley at 00:15:53) Yep.

(Derek at 00:15:54) —that did not feel like it really connected users with your content. And even though it was logical, it didn't mean that things were actually easy to find. And even when we introduced search, the same problem was there. And the real innovation on the web, both on mobile and on the desktop, was figuring out how can we engage users with content, find the information they want when they need it, and present it in a visually appealing way that's going to work on any device. And if you look at where we are now versus where we were years ago, you can go to Squarespace, set up an account, and create a beautiful website in a few hours.

(Derek at 00:16:38) Use my coupon code. But in conversational development, we are very early on where the technological hurdles are the ones that have been addressed first. So the earliest technological hurdle was figuring out how can we get natural language understanding to a point where it can do a credible job of taking a freeform user input and figuring out a user's intent. So it's kind of step one. Step two was, how do we build some tooling around it in code?

(Derek at 00:17:13) And then how can we move to a point where we can have low-code platforms that not just handle the natural language understanding part, but can actually fulfill on users' requests, walk them through business processes, et cetera. Now we're at that next step, which is how can we educate people, provide best practice and models on what good conversational interactions look like. People have never had to design a conversation before. So a lot of what we see on the web—and I think, actually, the voice examples are often better—but a lot of things just through basic chatbots aren't examples of good user experience.

(Derek at 00:17:55) What will happen, and I think what's going to be exciting is kind of this next chapter in the industry, is when we do really start foregrounding the user experience and looking at how can we create conversations that solve problems for people, that don't frustrate people, that understand people enough to help them get the thing done that they want to get done, and that when you hang up the phone, you feel good, or when you finish with a chatbot, you feel good. And also that you're connected to a person when you need to, without friction. I think once we get to a point where that user experience gap has been bridged, the whole mentality around what conversational automation is and the value it produces will change.

(Joel Beasley at 00:18:44) That makes a lot of sense. And I think what you guys are doing is really smart. Taking the low-code approach to maximize the customization options for each individual client being able to design their conversations. Because it seems like in the past, conversational AI is something that just requires so much customization to each use case that it's really, really hard to make a blanket solution. But taking the low-code approach is absolutely a great way to go about making that blanket solution, because you don't have to be an expert in coding or in neural networks to design the conversation.

(Joel Beasley at 00:19:31) You can just take this awesome basis that you guys have created and drag and drop your conversation design into it.

(Derek at 00:19:41) And I think there's two benefits there. The primary benefit, I would say, is that you can have, you know, on the one hand, less expensive resources, so you don't have to hire developers to do everything to build out your conversation, so it can save you money. The other thing is that you can push—you can empower other people in your company. We have companies we work with where it's the finance team or the tax team who are building chatbots to automate parts of their job, and they have no coding skills. But they're not thinking about it as, like, we don't have this—you know, hire a developer.

(Derek at 00:20:19) They're thinking about it like, oh, great. We can solve a problem quickly without having to have advanced skill sets. The other benefit from a—I think of it from a design perspective is when you do everything in code, it does give you a tool to immediately visualize and prototype what's happening. What a low code platform does, like Cognigy, is it gives you that visual component. You can be in that whiteboard session prototyping, build out conversations in real time, move dialogue from one point in a conversation to another, and have it be a really seamless and fluid and visual experience.

(Joel Beasley at 00:20:59) So tell me—understand a little bit better, because it sounds like a lot of the conversation design is done on the low code platform. So what is the AI aspect of it actually doing?

(Derek at 00:21:15) That's a great question. Yeah. So in our kind of—in this discipline overall, AI shows its face most in the most pronounced way in conversational automation through the natural language understanding. So that is letting someone—I actually have a client right now who I called into their IVR just to test out their current, you know, process. And they literally say press 1 to do this, press 2 to do that.

(Derek at 00:21:46) They go all the way through to number 8, and the only reason they stop at 8 is because press 9 is for other questions. But I had a feeling that if you could press 10 or 11 or 12, they would kind of keep getting you through to more and more options. What the natural language understanding component lets you do is it lets you start off by just saying, I need to check my balance. I don't need to know how to traverse the IVR to be able to get to that one thing I want. But to be able to take however I want to express I want to check my balance—which is not always going to be as simple as a few words—to be able to take that unpredictable user input and figure out what that user is trying to do is the heart of the AI.

(Derek at 00:22:34) And there's more to it. What if I say something like, I just got my bill, and my balance was higher than previous months—or my cost was higher than previous months—and I want to find out why. So you've given us a lot of information there. Essentially, you want to find out why your balance was higher this month than the last month. But you've given us a lot of little details that you've received your invoice, that we're talking about this month and last month, not six months ago and today.

(Derek at 00:23:05) And you're giving us a lot of little contextual clues along the way. So it's not just that we need to understand your intent. We need to pick up on all of the data points that you're giving us so that the next question we ask is not going to be, oh, you're asking about your bill. What month is it for?

(Joel Beasley at 00:23:23) Right. Right.

(Derek at 00:23:24) So the NLU needs to be intelligent enough to understand the whole context of what I'm presenting and then be able to fulfill on their request, again, without frustrating the user or asking them for information that they've already given us. That is the heart of the AI part of conversational AI. Now there's a satellite—or, many satellites of AI technologies that surround conversational AI that can be part of it or not, depending on what solution you're looking at. So for example, in a AI interaction, let's say in a chatbot interaction, I might get to a point where I need to upload a document. So let's say if I am filing an insurance claim, maybe I need to—I have a police report to do the claim—they need a police report.

(Derek at 00:24:16) So maybe I scan that with my phone and upload it. Well, you might have AI at that point in time that can analyze the documents, figure out what kind of document it is, validate that it's an official document, pull out all of the key data points from that document to be able to know what are the next steps in processing the claim. You might have AI that analyzes the knowledge base. So that way, when I know the user's intent, I know the best content to go back to them. I might have AIs that analyze the user's sentiment and can use that, for example, to give actual agents insights on what's been going on in the conversation.

(Derek at 00:24:54) So there's a lot of AI that surrounds the heart of the conversational AI, which is really about making the dialogue really work for a user.

(Joel Beasley at 00:25:05) The AI is doing the heavy lifting in terms of understanding what's going on and extracting the data points so that the conversation you design, you can design it to react to those data points and give the proper responses.

(Derek at 00:25:22) Interesting thing, though, is we are, again, still kind of in this in-between period because you can have very competent, automated voice experiences. For example, if you call in to United Airlines, you can say, hey. I need to reschedule my flight. It can understand you. It can walk you through the whole end-to-end process in a fully automated fashion.

(Derek at 00:25:44) But what we still do is because users aren't used to expressing themselves in a freeform way, a lot of times, the voice bots will kind of coach people. You can say things like, I want to check my bill, or I want to reschedule my flight, whatever. Right? So we're at this point where, from a user experience perspective, we're still having to, ironically, teach users how to interact with natural language. So I think that the—once the NLU is largely there.

(Derek at 00:26:18) It's not perfect, but it's largely there. It's really about kind of marrying that user experience component of users trusting that the system can now understand them. And it does when it works out well.

(Joel Beasley at 00:26:31) That actually—it reminds me of, like, I don't remember where I was reading this. But just the other day, I was reading that, like, in the 1950s, there was like a big push in the media because we had to teach Americans how to throw away their trash. Because more waste was being produced as, like, we were industrializing. And now that's just something that you don't think about.

(Joel Beasley at 00:26:56) You're just like, yeah, that's how society works. I feel like that's similar. Which—I feel like I'm doing a terrible job as your host here comparing this awesome technology to throwing away trash. But we're learning how to just be more casual with our machines, and that's something that's going to be so normal in the near future that you won't need to be presented with your options of, hey. Don't worry.

(Joel Beasley at 00:27:26) You're allowed to talk to this robot like it's a person.

(Derek at 00:27:28) Yes. Exactly. Exactly. And, two things to say about that. One is there might be a marriage in the middle.

(Derek at 00:27:37) So if you think about how you, as a, you know, technical person, interact with a search engine, we kind of intuit the right way to frame our searches with a sense of how we think the search engine thinks. And for me personally, that works. I want to say this in a way, or enter my search term in a way that's going to increase the chances I'm going to get the information back that I want. It's very possible that we might mutually train each other, that the NLU engines will get used to how they think—

(Joel Beasley at 00:28:12) Yeah. Yeah. Kind of—

(Derek at 00:28:12) The inputs that they work best with, and we'll learn to kind of talk with them. People always make fun of me because I have my Siri voice, you know, which is like when I'm talking to Siri, I'm super articulate, and I speak loudly and clearly because I really want Siri to hear me and fulfill my request. Another side story about the trash and the progression of how cultures deal with trash: I have a friend. He was traveling in another country, and this was maybe ten years ago or something.

(Derek at 00:28:43) But he was taking a bus, and there's a sign on the bus about how to deal with your trash. It said, don't be a pig. Throw it out the window. So the idea was, like, don't be, like, uncivilized and keep your trash on the bus.

(Joel Beasley at 00:29:00) Wow. The window is for—

(Derek at 00:29:03) So, just a funny random story that you made me think of. But, yeah, there's progressions of how we interact with the world.

(Joel Beasley at 00:29:12) Man, that's really interesting. I guess let's get back on to AI and data rather than trash here. But thinking about your the AI models behind the natural language understanding, I was actually just speaking with Steven Davis. He's a VP of technology at a company called InnoData. And they focus on, like, cleaning the data that goes into the algorithms before it gets there with—really they have a lot of subject matter experts that they work with that do data annotation that just makes the models work better.

(Joel Beasley at 00:29:51) And after that conversation, I'm just curious with your datasets. I mean, with your AI models, how do you train them? What do your datasets look like?

(Derek at 00:30:02) So there's different—there's different problems that we solve with NLU and with automation, each of which is going to take some different approaches. So there are broader, kind of, you know, broader knowledge-based areas where you're really looking at a significant corpus of information and trying to predict or be ready for a diverse range of questions that could come in. So these are the scenarios where you do have to invest and think through data annotation, looking at existing transcripts, human-to-human transcripts, looking at search, how people interact with your search engine. So they're kind of a bit more data heavy in terms of how you train your machine learning models.

(Derek at 00:30:53) There's other problem spaces, which is if I take the airline example, people are going to call in, and there's probably 30 things people are going to ask about—things having to do with flights being late or scheduling flights, things having to do with baggage, things having to do with upgrading seats and rewards programs. It's a much smaller problem space that requires a much smaller investment in being able to train the machine learning model. So people will often come to us with the expectation where they say, how long does it—how much time does it take to train the NLU? And it's really going to vary based on the problem space that you're looking at. If you're really looking at automating business processes, there's typically a finite amount of business processes that you're trying to solve for when people call up on the phone or when people interact with your chat.

(Derek at 00:31:51) And if they get more complex than that, then human agents can take over those conversations because those are going to be the ones that require more sophisticated judgment or analysis. If you're saying, hey, we have a knowledge base of, you know, 10,000 articles, and people could ask any question in any of those 10,000 articles, that's where you're going to invest more in building your NLU models, have more specialized models, and maybe join a number of different technologies, like relying on search to a degree for the strengths of search, relying on machine learning for the strengths of machine learning, and marrying that with the natural language understanding engine as well. But the key thing that I would want to leave people with is don't come to a conversational project with the expectation that you're going to have to hire data scientists or invest a ton of time in building out the machine learning models. Start in a different way, which is saying, where are our human agents spending the most time doing repetitive work that doesn't require human judgment or human empathy?

(Derek at 00:33:05) And then how much money are we going to save if we can automate some of those interactions? And then build back into that, what are the use cases then that we need to automate, and what are the dialogues? What are those user-friendly experiences that people are going to have so that working in a self-service model is actually preferable. So that's how we frame conversations with customers rather than kind of starting off with a more intimidating conversation about how are you going to train your machine learning models.

(Joel Beasley at 00:33:39) That makes sense. And it comes back to how we started the conversation when you were telling me about your position as conversation automation. You're starting with the objective rather than the hiring the data scientists. That makes sense.

(Derek at 00:33:56) There's some conversational automation problems that—that's kind of a scandalous thing to say in our line of business—but that don't require an NLU, really. So as an example, there are some interactions that are survey-like interactions. From a low code platform perspective, you want to empower the business users to make changes to those kind of more linear parts of the conversation. So you want to empower them with the tools for automation, even if there is some NLU going on, but even if the NLU isn't really front and center of what the interaction is about.

(Derek at 00:34:35) What's front and center is saying, alright, you want to book a flight? Well, guess what? I need to know your departure airport, your arrival airport, the day you want to travel, and the airline you want to travel on. And, you know, there's some NLU there in that I can give you those in any order or whatever. But, really, the heart of what it is is walking you through a process. We have customers who are using automation to qualify people. So if someone comes in and they're looking for a particular product, it could be, alright. Well, we need to ask you these three questions to get you to the right product specialist.

(Derek at 00:35:07) So it's not really NLU focused, but it is automation focused. And empowering those nontechnical users to create those automated parts of the conversation is a huge value that the product brings.

(Joel Beasley at 00:35:20) It makes a lot of sense. So recently, I was listening to this podcast called HPE Tech Talk. It's a tech podcast put on by Hewlett Packard.

(Derek at 00:35:30) Mhmm.

(Joel Beasley at 00:35:31) And there was this one really cool episode where the host of the podcast was actually interviewing his daughter who's pursuing a master's in applied data science. And together, they were discussing how to keep bias out of AI datasets and AI training models. And I was just curious, what tactics do you employ to keep bias out of your NLUs to make sure that everyone can be understood?

(Derek at 00:36:02) So there's a couple of facets of this. One is transparency. And machine learning models, in essence, are the opposite of transparent. They're designed to be a black box that you don't have to understand. It's not fixed rules, so it's all doing all of the thinking behind the scenes. And yet, when it comes to something as important as bias, being trustworthy is important, or being transparent on how you're analyzing what's happening and the assumptions you're making is important.

(Derek at 00:36:41) That's why it was really exciting for Cognigy to be the first AI company to be tested for security and reliability for what's called AIC4. This is a standard—it's more of a European standard for trustworthy and safe AI that tries to address things like security, reliability, and explainability so that what tools the AI is deploying behind at the scenes are transparent. Now that said, when we approach a customer and kind of are listening from them, how can we build the best NLU models, we want them to provide data from actual chat transcripts that they have. We don't need to necessarily use the transcripts verbatim, but we want to take their actual customer information to understand how conversations happen with them. And we have a diverse range of customers.

(Derek at 00:37:41) For example, we have a customer who is in the healthcare space, and most of their caretakers are from other countries. And so one of the concerns that they had was that the speech-to-text engine could actually understand their employees, and it wasn't just trained to be able to take my scary, friendly voice and recognize that, but then frustrate everyone else in the company who doesn't speak English as a first language. So you have to take these considerations into account when you're building out your models and testing them. And even when you're creating your training data, making sure the training data reflects how your users or your employees are actually going to interact with the bot. I see conversational automation as not the worst or most risky area for AI in terms of bias in the sense that there are plenty of other AI areas where this bias can be made material in ways that can have negative effects.

(Derek at 00:38:49) So for example, even classical statistical models around insurance rates or bank loans, other kinds of things like that where you can have a lot of intelligence that takes the data you have that just validates cultural biases that already exist. I think conversational automation is less prone to that as long as you're able to seed the models with actual data from your users that takes into account the diversity of users who are going to be interacting with your bots.

(Joel Beasley at 00:39:21) Yeah, that makes a lot of sense that it is lower risk, and you are able to, with those transcripts, just make sure that you're training it on exactly who's going to be interacting with it. So that's really good to hear. Makes you feel more at ease implementing the conversational AI.

(Derek at 00:39:45) And I think it highlights the fact that just like in our interactions with each other, in our interactions with technology, we want to feel understood. And when we're not understood, there's one thing of wanting to kick your computer because it's not working the way you want to do. It's another thing of feeling like the system is not responding to you because of how you express yourself. So I do think it's really important to keep in mind. But the tooling that's available lets you do that if you're just trying to design a good user experience for your actual customers.

(Joel Beasley at 00:40:23) And once Cognigy is deployed, it continues to learn, right?

(Derek at 00:40:29) Exactly. The best engines out there do. And the learning path starts off the voice context with the speech-to-text engine. So being able to take something that a user says and being able to turn that into text that then can be processed by natural language understanding model. And sometimes those can work hand in hand as well.

(Derek at 00:40:53) You can also train your NLU with training data that reflects how people say things. So sometimes we use bad grammar, or we refer to things in terms that may or may not make sense, but it's kind of a term that we use. And as long as you provide that in your training data, it can start off in a good place. And then as it has those successful and unsuccessful interactions with users, it can improve the model automatically over time. But that said, to your point about bias, you also want some human curation in there. So that way, real people are looking at this data to make sure that if something a user has said is ambiguous, they can use human judgment and human empathy to be able to coach the computer on how to better understand this kind of an utterance next time.

(Joel Beasley at 00:41:43) Yeah. That's really crazy thinking about how we're teaching computers how to understand context clues. Because I don't know. That's just I feel like a really advanced level of understanding and awareness of cultural awareness. Yeah.

(Joel Beasley at 00:42:02) It just really amazes me.

(Derek at 00:42:05) It's quite powerful. And as we think about it and are exposed to the imaginings of what's possible, so we watch TV shows and stuff like that, the power of the AI is there. But again, if we think about that automation lens, there's a question I have about how good it actually has to be to be effective. And if you look at something like search engines that we commonly use, I might not be able to put a free form user input or talk to my search engine or maybe have the ideal human interaction with it. But they're very powerful at finding the information that I want in the top three to five search results.

(Derek at 00:42:52) And that's good enough for me. That makes me productive. And I think that when we're setting our sights for what helps us today in our business and making our lives better, that close enough might be close enough.

(Joel Beasley at 00:43:09) Yeah. You're really smart. You're really good at pulling back to the goal-focused technology here.

(Derek at 00:43:19) It's very practical, thinking about practical applications of it for sure. Absolutely. In our space, there are as many conferences that focus on the pure academic side of things as do on the business side of things. And at Cognigy, we're always following both, but we want to make sure that the end goal is always in our minds, which is making life better for companies and the customers and employees they serve.

(Joel Beasley at 00:43:47) Cool. So before we wrap up here, I wanted to ask you a couple leadership questions. Is that okay?

(Derek at 00:43:53) Sure.

(Joel Beasley at 00:43:54) Cool. So if you could design the perfect leadership training for your team and the leaders that are your direct reports, what would the most important couple concepts be?

(Derek at 00:44:12) Yeah. I think fundamentally, leadership is grounded in the fact that understanding that the people who work for us are fundamentally people with a broad set of motivations and passions and limitations. So when we recruit people and when we look to retain and develop people, we need to plug into their human qualities. In the, sadly, in the software business, there is an expression that says salespeople are coin-operated, and it's a shorthand way of saying that any financial motivation you put in front of a person in sales will dictate their behavior. And while there's some truth to that for anyone in a capitalist society, it misses the bigger point, which is that the people who work for us are fundamentally complete beings with a wide range of motivations, and we need to honor all of those in order to get their loyalty and their passion and to enlist their support when we have to work really hard to get something done.

(Derek at 00:45:34) So I think, fundamentally, a people focus is one part, one ingredient. The other thing that's aligned with that is motivating people by connecting to what they're passionate about to the degree possible. We all have aspects of our jobs that we're not passionate about. But to the degree that our jobs can give us meaning beyond just the financial enablement that they give us, which is actually also quite meaningful. But the degree that we can create meaning around work, through our relationships, through, in my example, through a connection to the product that I feel passionate about, or for other people, it might be in their knowledge domain or, for yourself, in your career development. These are all things that you can connect to that motivate you, that that's what makes it worth it to get up every day. So people is the hardest aspect of work life, but it's the most important and an extremely rewarding part of it.

(Joel Beasley at 00:46:32) Yeah. And we're seeing a lot more of employers having to look beyond the financial incentives today because with remote work, we're seeing wages kind of flatten across the board for a lot of jobs that are remote-enabled. And so if everyone's offering the same amount of money, you have to figure out what are the other ways we can attract the top talent. Are we advertising our mission? Do we offer good time off?

(Joel Beasley at 00:47:04) Do we offer remote enablement? Because even companies that are fully in office are still competing with companies across the country for the people in their town. And yeah, that's just something I've been reading about and seeing all over the place today.

(Derek at 00:47:20) I think along with that, we need to encourage people and create a culture in, I think, the business community overall of risk-taking. And while this past year and a half has been tragic and a disaster, to the degree that there's something recoverable from it, I think it's the vulnerability that we're able to show each other in the business context that we haven't been able to do before. So I've seen so many people's pets or kids or guitars and drum sets. I've been able to peer into people's lives in ways that I wouldn't have been able to otherwise. And when you create that connection with people, both with your peers at work as well as your customers, and when they can create that connection with you, it builds lasting relationships that make you happier, make business better, and make you more connected to what your company is doing.

(Joel Beasley at 00:48:24) Yeah. That is a really positive thing. I love the higher level of empathy across the board for just knowing that everyone is an individual with a busy life and stuff going on. I love being on calls with clients, and if a kid screams in the background, they don't even apologize because they don't have to. And now they're more comfortable knowing that they don't have to.

(Joel Beasley at 00:48:50) Who cares? Kids yell.

(Derek at 00:48:54) Yes. Dogs bark. Yeah. I was on a call with a colleague yesterday, and it was evening his time. And he was taking a walk. He had his baby, his kind of newborn strapped to his chest, and he was just walking, FaceTiming me with his kid while we were on a call with a customer. And all of this, I think, makes our jobs more human. And I think all of the other, I mean, certainly, all of the other business needs are there, like accountability, doing what we say we're going to do, being able to meet deadlines, being able to show up when we need to show up, overperforming when the situation demands it of us. All of that is still there. It's just easier to do when we know that the people we're working with understand the full context of us as people in our lives.

(Joel Beasley at 00:49:52) Yeah. And it's been interesting to see how a lot of companies were really scared to offer remote flexibility because they were worried about their output going down. But then, suddenly, you realize when your employees get to sleep an extra hour because they don't have to drive to work, wow, they have more energy, and they can do their jobs better. And that's just been a really, really cool thing to see that not only can it work and can it hold you over when you have to be remote, but it can actually be better for the bottom line. And that's just always a really good thing when the bottom line can line up with what's actually best for each individual.

(Derek at 00:50:40) Yeah. I think across the board, people have seen productivity increase this year despite the fact that we've all been working under stresses that we've never had to deal with before.

(Joel Beasley at 00:50:52) Yeah. So what are you learning right now as a leader?

(Derek at 00:50:58) One thing I think I would say is what you just said. I used to be a passionate "be at work" person, part because working at home kind of drove me crazy. But also, there were specific ideas and outcomes that literally happened by the water cooler. And I just felt like there was no way to replicate that kind of impromptu experience. But I think what's really changed that for me is, A, just my lived experience of the past year, but also the critical nature of video, I think, in our communications.

(Derek at 00:51:43) So being able to see each other as we talk, that was the missing ingredient, I kind of feel like, in being able to connect with people, feel more spontaneous, know that people are actually paying attention to you when they're talking rather than writing an email. Or when they are writing an email, you at least know that because you're watching what they do. And the ability to have those same kind of spontaneous ideas and heart-to-heart interactions that could happen online was a big revelation for me. And I see it culturally as I interact with different companies as well. Sometimes, because we always have our cameras on, and they'll say, "Hey, you know, feel free to turn your camera on." There's some companies where it's like, no, we don't do cameras. Everyone's different, and so it's okay. But it also it's a missed opportunity for people to be able to connect with each other if they can't see each other face to face.

(Joel Beasley at 00:52:45) Yeah. Absolutely. And I think just the acceptance of having cameras on and not feeling uncomfortable with that is huge too. Being able to just Slack a Zoom link to someone with no context, and they're like, "Oh, cool," and they join. Rather than having to schedule a call, "We need to talk about this, this, and this at this time." You mentioned it a couple times, spontaneity. I think that's really key in those water cooler interactions that end up being so impactful.

(Derek at 00:53:19) Yeah.

(Joel Beasley at 00:53:21) Well, before we wrap up, is there anything that we want to get out there for Cognigy that we didn't hit on today?

(Derek at 00:53:29) Yeah. Thanks so much for the opportunity to say a couple of things. I think the point of what we're trying to accomplish at Cognigy has come out here, which is really about providing better experiences for customers and employees by having 24/7 access to self-service while also saving companies money and saving employees the toil of having to just be data entry people or order takers. I think that point has probably come across in the benefits of the low-code platform. What I would say is a lot of companies are in the stage where we know we need to do something, and yet you haven't taken that next step to actually get it in the budget to do a pilot project or start implementing something.

(Derek at 00:54:15) So what I always try to coach people on is two things. One is that it's easier than you think. It takes less effort than you think. It doesn't have to be anything close to a million-dollar project. So in terms of addressing those initial objections, realize it's easier than you think. You can get results quickly. You should be able to get results in three months or six months, not 18 months. And you can create better experiences than what you've seen elsewhere. So good user experiences are available in this world, and we work with you and our partners work with you to make sure that what you're creating creates the best possible representation of your brand to your employees and your customers. So take a chance and give us a call.

(Joel Beasley at 00:55:07) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn, or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.