Leadership Insights: Generative AI Compared to RPA

Transcript

Mike Vizard: Hello, and welcome to the latest edition of the Digital CxO Leadership Insights series videos. I’m your host, Mike Vizard. And today we’re talking with Drayton Wade, who is head of product strategy and business ops for a startup called Kognitos. So we’re talking about AI in the enterprise, especially this whole new generative AI platform that everybody’s talking about. Unless you’ve been under a rock, you’ve probably heard about ChatGPT somewhere. But there’s other ways of thinking about this stuff. Drayton, welcome to the show.

Drayton Wade: Hey, Mike, thanks so much.

Mike Vizard: So walk us through it a little bit. I know you guys are investing in this space. But it seems like there’s multiple platforms out there that are tuned for different use cases, and you guys are one of them. So explain where you fit in the AI landscape these days as it is continues to rapidly evolve?

Drayton Wade: Yeah, absolutely. There’s a lot of different applications. I think of this core technology called large language models, which is coming online. So the most familiar to most people would be GPT-3, then ChatGPT, which is a more user friendly version of GPT-3. Recently, we heard about Bard coming online from Google. And we can expect Meta and you know, even a lot of others to release their own large language models. When you have large language models, you have the ability to translate actual day to day English into different generative action and creativity that you may want to do. So you’ve seen that with DALL-E, where people can say what they want to create an art. And the artificial intelligence will create different art based on that. There are some around; there’s one called Jukebox that helps create actual music, a very famous one is Copilot by GitHub, which helps create and generate a lot of the baseline code that would go into products. And so we’ve taken a slightly different step, where we have made generative AI for automation. So the automation space has traditionally been plagued by having to have really, really trained people to help automate business processes, processes like claims or uploading purchase orders that the manual day to day things that people hate to do, don’t do or forget to do quite often. And so we’ve managed by building on our own large language models, and then also leveraging technologies like GPT-3, and other generative AI. We’ve made it where business users can say what they wish to do, what they want to have automated in their work. And our tool translates that into automation that then runs and automates that work.

Mike Vizard: So explain an example of how somebody is using that within the context of a business process? Is somebody just walking in and saying, I need to fill out my HR forms, and then this all gets generated? Or what does that look like?

Drayton Wade: Yeah, it’s good question. So most businesses start with their high volume, highly repetitive processes, because they tend to have the highest ROI. So a great example of this is – we have a trucking company down in the southeast, where they get these bills of lading, which have all the things that are listed on a particular truck that’s coming in. And previously, they were having to look at those, and then manually type that information into any ERP system. At the end of the day. You can imagine with humans, we’re not really good at this, we tend to make a lot of errors. And so instead, they have automated that entire process with us, which basically allowed them to reallocate half of that staff to other more meaningful activities. And it’s reduced their error rate by over 80%. So we see a lot in finance and accounting, starting off accounts receivable Accounts Payable a lot within HR. And then we’re increasingly seeing quite a lot within supply chain, which has a lot of documentation, a lot of data, moving from different systems and things that should be automated.

Mike Vizard: We heard a lot about this kind of automation within the context of things that people call robotic process automation platforms. Is this kind of the next iteration of that? Are we kind of moving the ball forward in a way that may be simpler and easier and possibly even more effective?

Drayton Wade: Absolutely. So we actually have several people on our team that come from the RPA space, myself included. I was previously UiPath in the early stages through their IPO leading their strategic partners team, and RPA. That version of RPA is a great tool for processes that are very standard that don’t tend to change very much, and that don’t have a lot of exceptions is a very common phrase within process improvement – things like that. So RPA requires developers – basically a business user has to tell a developer what they want to do. The developer goes in and builds like a flowchart based on that. And then the bot essentially runs that flowchart across different processes. There’s a few challenges with that, though. And that’s why we actually created Kognitos as well. One, and this is all the time, is what I’ve just mentioned; it requires developers. There are very few actual developers in the world relative to the number of business processes we have. So there was always a huge bottleneck, where you would have to train these developers to be able to automate anything, which makes it very expensive. The second issue is anytime a process changes. RP doesn’t handle it very well. Instead, tools that are built upon AI, and they actually can be taught and can learn to handle those different changes. So what we’ve done is we’ve made automation, all in English, to where a business user is now the person in control, they can build the automation, they can manage and handle any changes or exceptions. And they can go back and always see exactly what the AI has done. So it’s not a black box, where they don’t really know what’s going on. And that fundamentally changes the cost of automation, the accessibility for all users.

Mike Vizard: As we think that through, in my experience, there are processes that are very structured, but there are more processes where there are more exceptions than rules. So how does the platform know or understand when there has been another exception? And through a process? And how does it learn? I mean, do I essentially prompt it by showing it what that new process looks like? And how will they actually kind of figure out what’s going on?

Drayton Wade: Yeah, it’s a great question. Actually, your question comes to something that’s even more core that people should be asking. So if you look at computer science, traditionally, we as humans have taken the approach of building machines. And then we ask people to learn how to communicate with those machines, and to how to engage those machines. At Kognitos, we’re kind of flipping that around. We’re saying that machines should know how to operate and work the way that people do. So to take your example, or your question around exceptions, if I was working with an intern, and I asked them to do a process, and I told them, “Hey, I want you to go get these documents, find this information, and then put it into some system.” If they had a hard time finding that info, they would just come back and ask me a question. And then I would give them an answer, or I would tell them how to go solve that problem. All this takes place in English, that’s the way we as people operate with each other. Our systems is designed the same way. So in generative AI, whenever the automation hits a problem, it actually will create a sentence on its own asking, “Hey, I can’t find the supplier number.” It’ll ask that back to the human. And then the human can actually tell it, how to find that information in English and teach it so that it then remembers that in the future. So imagine you have a document, like that trucker example I had earlier as a supplier number. And then right under that is the trailer number for an installer, and I could just tell the intern, “Oh, the supplier number’s always right above the trailer number for this vendor.” And we’ve made it the same way with Kognitos. As you can tell, the English logic like that then stores and can use it in the future. So it’s able to have what’s called “learnings” the same way that a person would learn how to handle a particular process. This makes it really durable, and really easy to add all those exceptions. And you’re right. A lot of processes changed a lot. And they have a whole lot of exceptions in today’s business.

Mike Vizard: A lot of those processes are designed by people, and people are flawed. So can the machine figure out some of our processes are inefficient and make some recommendations?

Drayton Wade: 100%. And that’s really the vision for Kognitos is not only to provide on demand labor, which is basically what automation is, but also to provide on demand insights at the end of the day, to where a person can go back. And because we’re taking those processes and logging them all in English, that person can query and actually find where those issues are. So giving another example. We were on site last week in the RV capital of the world Elkhart, Indiana, which I’ve never been to before, but we were working with one of our customers who is an RV supplier and they’re constantly thinking process improvement; Manufacturing, lean manufacturing, how do we streamline things? And what they found is because now in their AP and AR processes everything is in English. Their AP people can go in and see when exceptions are happening very frequently. And saying, we probably should just redesign that, even though Kognitos can handle the exceptions and do it fine. It would be better if we just redesigned that process to make it more efficient and streamlined moving forward. So now you don’t have to hire armies of consultants to go back and try to map your processes and do all this. Because it’s operating the way a human performs it, you have basically a record of how work is actually happening in your business. And then you can query and find those inefficiencies. All in English, anyone can understand it.

Mike Vizard: As we think this through a little bit, we’ve been watching this low-code, no-code movement, wherever you’ve been trying to encourage citizen integrators and developers to kind of create some of these workflows themselves. And they haven’t learned a tool and basically started to think like a developer. If I understood what you just said, perhaps I don’t need to do that anymore. Because I can use a platform that will just in plain English, allow me to describe what it is I need to have done, and things will magically happen.

Drayton Wade: You are spot on. And this was a challenge that I saw on some of our other team members that previously came from RPA. I was that RPA company for a long time, which is a no-code, but a low-code technology. It’s definitely using Python. It even at the end, even though it’s a great company and good tech, I couldn’t really build my own processes. And one of the problems is what you just mentioned, we’re trying to force people to think like developers and to think like machines which should be the other way around. We build machines to serve us to help us be more productive. If I have to learn how to operate like a machine, that actually doesn’t make me more productive. And in fact, economists have shown that’s had a big effect on productivity, flatlining over time, because we’re trying to teach people to operate like machines, and they just end up typing more things all day. Instead, we’re operating the way people operate, we’re intentionally thinking through the psychology of how people work and designing our product around that to where people just use Kognitos as the way they would interact with another human. So that’s conversational. Work is conversational, either through email or through talking to people. We think in conversation, we think in sentences. And that’s how we’ve made Kognitos – to where it operates just the way people think. So people should not have to quote, learn how to use an automation technology; it should operate the way that they operate.

Mike Vizard: What will happen to the legions of IT people that we have hired and rely on today as interfaces to the machines? How will those roles evolve?

Drayton Wade: Yeah, and this, this question goes, I think even more, not just into automation, but broader generative AI. So we’re starting to see generative AI, not Kognitos, is one of the tools that can create user interfaces, by you telling it what you want to see. And based on all the examples seen on the internet, it can then create some of those interfaces. I actually look this much more positively than others. I think you are now democratizing the ability to automate work that humans aren’t designed to do to begin with. We’re not designed to do manual repetitive tasks, we are creative, we are logical, were designed to do bigger things. And we’re bringing essentially the means of production down to average people. And so basically, up-leveled everyone in terms of productivity, because now, with us, an individual user can automate parts of their work and have a lot better insights into making decisions day to day. A developer can automate their baseline code so that they can think about more creative code and operate much faster. And I look at it the same way with IT teams – there are so many other problems that they need to solve day in and day out that managing RPA bots was not a good use of their time. And it’s a very frequent complaint. So now we’re enabling them to think bigger. It challenges what they may have, how should they deploy new technologies in their business? How do they make sure their business is prepared for the future? So to me, it upskills, essentially everyone, which is really exciting.

Mike Vizard: We may still need systems of record, but all these quote unquote, systems of collaboration that we’ve been trying to build from the ground up the last decade or so, could just become highly automated, if I understood you correctly.

Drayton Wade: Yeah. So essentially, automation and Kognitos shows us what RPA tried to do and did to some extent, right? It is that connective tissue between different applications, which enables data to go back and forth and across all of your different applications. We do it all through API. We think that’s much a much better approach than screen scraping or anything like that.

Mike Vizard: Alright, so ultimately, how smart can smart get? Because the things about these AI platforms is they get smarter at a faster rate than we do. So what comes next?

Drayton Wade: Yeah. So this brings up like a very interesting concept that’s starting to be debated around generative AI, ChatGPT and AI in general, which is how do you use AI in the enterprise in a way a that you can trust, right? With a person you’re able to interact with and develop trust with an individual, but you have to be able to do the same with AI; it’s gonna get very, very smart. It’s going to supersede I think our intelligence in time. However, if you intentionally design the systems to where the human is always in control, there’s ways to manage it and to utilize it into something that’s more powerful than you. And we’ve been doing that with machines for a long time, with machines that are much more powerful than us. So we make it to where AI with Kognitos, we tell him what you want to do, it populates how it would automate. But ultimately, you have the control over running that automation. Likewise, everything has been done, you can read in English, so you have full audit records; you can go back and change things. AI that’s in a black box, which is where I get a little bit nervous about some of the other applications of generative AI, you can’t really use in the enterprise, because in the enterprise, you need high levels of consistency, and you have to be in control.

Mike Vizard: All right, folks, a wise man once said, “The future is here. It’s just unevenly distributed.” Hey, Drayton, thanks for being on the show.”

Drayton Wade: Thanks so much, Mike. I really appreciate it.

Mike Vizard: And thank you all for watching the latest episode of the Digital CxO Leadership Insights series. I’m your host, Mike Vizard, and you can find this episode and others on the digitalcxo.com website. We invite you to check them all out. And once again, thanks for watching.