Hi, and thanks for joining us. I'm Mike Bond with TTEC Digital, and I lead our customer experience transformation studio. So our consulting team focused on implementing technologies that actually achieve their goals. So today, we wanna talk a little bit about knowledge engineering for AI, and we're talking specifically in this case in the CCaaS space. And what tends to drive AI performance, especially when you're looking at contact centers, isn't as much the technology itself. It's the knowledge behind it. So what we see is a lot of teams that run into trouble because they tend to treat AI as if it's all one capability. And in reality, there's kind of three different problems to be solved or three different applications for AI, and that's what I wanna start with today. So for context and and for language throughout, this presentation today, we're gonna talk about three different types of AI that all live within contact centers today. So the first is conversational AI. Those, you know, you'll hear referred to as chatbots or automated virtual assistants, IVAs. The case there or the the use case for those is that it needs to understand the customer's intent and then manage the dialogue and get us to an outcome, whether that's a self-service automated outcome that satisfies the customer's intent or getting us to the right human who can help us resolve whatever the the reason is for our inquiry or our reaching out to the brand. Those are all things that fall into that purview of conversational AI. Another very popular use of AI in contact centers, is agent guidance. So to guide the agent through the conversation. So that's really supporting the agent so they don't have as much dead time in the call. So that they don't have to put the caller on hold while they go look up a a piece of knowledge or a piece of information on how to satisfy the caller. And then the last that that we're seeing, you know, very prominently now is around, autonomous process AI or agentic as you'll hear it called. So that's AI that can actually go execute tasks. So while you have either a virtual or a human agent engaging with your customer in the contact center behind the scenes, you can have a virtual agent going and taking actions and actually completing tasks on their behalf, again, to satisfy that caller's intent. But the key takeaway from that, and and while you're thinking of those three types of AI deployment, is that they all draw on one knowledge foundation, but they do it in very different ways. And that's one of the problems we see is that we keep treating AI as if it's all one thing. But, really, what I want you to do when you're designing the implementations, designing the support there is to think about which category you're designing for. So let's talk a little bit more about what these three categories mean or the three operating demands. So I'm not gonna read every word on this slide. I know there's a lot here, but I encourage you to, you know, take some time and and go through these three so you really have an understanding of what is the primary role for this type of AI. I talked a little bit about conversational AI and how it needs to really be your language. You know, it needs to speak the language your customer speaks. It needs to understand the languages your customers present. When we talk about guidance, that needs to support the agent in the context of the conversation they're having with the client. It needs to give them next best guidance. Sometimes we call that next next best action or next best conversation. Whatever it is, it needs to be guiding the agent through that conversation, again, to make it as efficient and effective as possible for the caller. And then lastly, when we talk about process AI or autonomous or agentic, that needs to be able to orchestrate and execute those defined tasks, those workflows. So when you think about their primary role, you can understand where now the knowledge base and what those tasks need to draw from that knowledge base would differ. So if you look at what it must interpret and what it must produce, that needs to be influencing your strategy for how your knowledge is curated, how it's created. You can also then see some ideas here of where it fails. And a lot of times, this is because the underlying knowledge supporting the AI isn't sufficiently training it. And then with all of these, the human still matters. And I think it's important that we spend some time and as we're doing these designs, thinking about, okay, how is the human going to, be involved in this process? So as we're thinking about these patterns, each of the AI having its own job to do, now we need to think about, okay, how do we organize our knowledge to support that mission? So what engineered knowledge needs to look like in each of these categories, you know, around conversational AI, it's that language understanding and response orchestration. So when we talk about things like intense intent models or utterance variation, you know, and we're not talking about a lot more than just, different accents, different dialects, things like that. I'll give you an example. We had a health care client, and one of the things we found was they supported both, employees with employer sponsored health care plans and Medicare, Medicaid patients. And what they found was those two populations could be having the same intent when they called, but would describe it differently when asked what can I help you with by the virtual assistant? So understanding the language models of those two populations was hugely important for creating successful conversational AI. That was part of the knowledge that needed to be curated. And when we talk about agent guidance, that needs to be context context rich. And and it's in the moment of decision support. What we mean by that is very often content or knowledge articles as we all think of them were written by a human for a human, assuming that a human would be reading this. So they're not always shaped and put into the context of a machine readable format. So they're not always indexed properly. Things like that that help the AI rapidly search and retrieve. So, you know, very lengthy text based knowledge articles generally don't even serve the human agent very well because it's too much to read, but it certainly tends to confuse the the virtual agents. So we need to make sure that things are properly annotated and put into the context of being machine readable. And then when you're talking about the autonomous or the process AI, that knowledge that needs to serve that needs to understand the processes and the requirements and what are the business rules. What is an exception? What needs to be escalated to a human versus what can be automated? All of those types of rules, which generally are not covered in the majority of knowledge bases that we encounter. So when you think about, you know, good knowledge engine engineering, it means that you've designed your knowledge content fit for the purpose of the AI it needs to also serve. And, of course, it also needs to serve our human agents because we want our virtual and human agents to all have the same information and arrive at the same answers and provide those answers to our customers. But think about it. They consume it in a very different way. So at the very bottom of this slide, it's really one knowledge layer, but it needs to be structured properly. It needs to be contextual. And one thing we also see with a lot of clients, it needs to be governed. So we need to have the ability to know who is maintaining and managing the that knowledge, which is, you know, the basis on our what our AI is going to achieve. Who's managing that and how? And are they sticking to the guidelines of creating and managing the knowledge based on the jobs that the AI is gonna need it for? So one of the things we want we always like to say is, you know, it needs to be be built to be read both by human and virtual agents. So when things are built as knowledge documentation for humans, they tend to be static, written for comprehension. You know, I mentioned the not having the contextual tags, and they're usually not in a a machine readable structure. So those are all the things we need to address when we think about knowledge as being executable by AI and logic yeah. Supporting the logic of the AI. So I'm I'm gonna close with this. It sort of is the takeaway. Again, you're when we talk about knowledge engineering for AI in the contact center space, we really it's important that we understand what type of AI we're talking about, what type of tools and capabilities we wanna implement, and then make sure we have properly engineered our knowledge to sufficiently service that type of AI. If you have questions on how to do this in the real world, we'd love to have you reach out to us. More than happy to engage in conversations and help make sure your knowledge is engineered appropriately to satisfy your AI. Thank you very much. I look forward to talking to you next time.
