How Cisco builds guardrails for contact center AI

Bobby, good to see you. Thank you for joining us on TTEC's CX expert series that we've been doing. So we're really excited to have you here. I'm excited about the conversation that we're gonna get into around everything Cisco's doing from an AI front. But just to start off, why don't you introduce yourself, and we'll go from there. Will do. Caleb, thanks for having me here. I'm excited for all the work that the both of our teams do, but more importantly, I look forward to this conversation. I'm I'm sure it's gonna be fun. So my name is Abhi. I lead the AI solution consulting practice at the Webex CX business in Cisco. And primarily, me and my team, we're involved in AI adoption, working with customers as they go through their AI journey, and specifically with those customers who are pushing the envelope, so to speak, in terms of AI products. So that's where most of my focus is. So I definitely look forward to diving deep into a couple of these topics and see where we go. That's great, man. I think I think you're gonna have a lot to share. So we'll we'll just kinda start here. There's there's obviously been AI is the theme, the topic of, I don't know, the decade maybe, but the last for sure, the last couple of years. We saw a lot of new announcements from Cisco and WebexOne late last year. A lot of exciting things on on where AI is coming natively into the product and some of the things you guys have been working on. So I'd love to just kinda start with what all are you guys doing right now? Kinda what's your vision a little bit for where you're going with AI within the the different Webex products, and and we can go from there. Sure. No. No. I think you're you're spot on. We had a slew of announcements in WebexOne last year in San Diego, and we've been working on the portfolio for quite a bit even before that. But, you know, if I were to summarize this on what our focus is right now and going forward, is really to see, like, how we bring AI into the world of CX, not as a bolt on. Right? Like you said, you know, AI has obviously garnered a lot of attention in the last couple of years at least and in a fair few before that. But in general, we've been in the CX business for a long time, you know, yourself and myself included. And at the end of the day, our focus needs to be on making sure that AI, you know, is one of many tools that ultimately contribute to a meaningful change in customer experience and or employee experience. Right? So we, so our focus has has continues to be we don't see AI as a bolt on. Like, we're not aiming to put AI on top of our existing products. We're not trying to take a feature and make it AI led, so to speak. But our focus is really to say, hey, if you were to relook at a particular experience, right, where you have AI at your disposal, how would that experience look like? Right? So if I were to kind of illustrate with an example, the typical fork in the road for most of the organizations is, hey, can I just take self-service and replace my good friend DTMF with IVR and then replace that with AI? That's one phase of evolution. But the way we are seeing it is, hey, if you really look at AI as this unlock of capacity, it's able to, you know, interact like a human, but also bring in the power of automation, etcetera. How would you rethink your whole flow of customer interaction? Right? Would you even want to have an IVR plus a handoff, so to speak? Right? So that's where our focus has been, where, like, all the announcements you saw in WebexOne have been the key building blocks. Like, we announced the Webex AI agent, the AI assistant that sits next to the human agent. But now the focus is how do you bring this in into the CX platform where, you know, enterprises can adopt this and not really have to struggle with, hey. How do I put AI, you know, wedge it into the existing ecosystem? But really to say, can I use this to make a meaningful difference to my customers and employees? Right? That's that's where most of our focus is. And I'm kind of excited on on the direction that we are on. Yeah. I look. I actually really compliment the way that you guys are are looking at it because as as we've dug into the product and the nice thing is that, like, I get to see AI in a lot of different forms and fashions. I get to see it deployed in, you know, many different products. I think the the direction that you guys have taken kinda like what you're talking about. I'll call it some of these journeys that happen within the customer experience where instead of looking at, like, oh, yes. We need to self-service, so let's go deploy self-service. But really the end to end touch points all the way through down the to even the data side of things, I'll call it kinda, like, the application of AI within customer experience. And and I think you guys are in a unique position because take Webex contact center. Like, you own the product. You own the desktop. You own the supervisor experience. And so you're able to control where AI is getting exposed, where you're introducing it into some of these flows. Yep. But you guys have also taken some of, like, an open strategy where I think it's kinda like a best of best of both worlds scenario. You have something that's native, but then customers, companies can still come to the table. And if they have models they've built, they wanna leverage, they can they can still utilize that. At least that's that's how I see kind of what you guys have been, you know, working on. No. Absolutely. Because I think see, what you touched upon is quite important because the the the point around companies having their own strategy in terms of you know, we have seen companies who have an approach where they wanna build everything, you know, right from the LLM all the way till the application. Right? There are some companies who have a very different appetite because, you know, they they do wanna have multiple options when it comes to their conversational AI experience. Right? So the the way we always look at it is with the Cisco platform, you have the right to use our AI but not the obligation because we genuinely, you know, like, recognize that there is a world where, you know, there is a coexistence and a complement strategy that we have to do for the sake of the customer. Right? Because, you know, this might sound cliched, but at the end of it, we genuinely want the customer to be successful in their AI journey. Right? Sure. So the because if you look at if you look at the goal saying that, oh, I wanna push for Cisco adoption, it's very different from saying that I want the customer to be successful in their AI journey because everybody's AI journey is different. Like, there's some customers you know, like we started off speaking, there's some customers who are really pushing the envelope when it comes to AI. Like, we I'm like, I work with customers for building their specialist small language models as an example. Right? So some customers who have built custom TTS engines, but there are also some customers who do not want to go down their path, but they want everything in a single platform where they know for a fact that they get everything from one vendor. But in all of this, the one thing we kind of bring to market, in my view, is the fact that we underpin this entire platform no matter which part of it you use in the Cisco reputation of security, scalability, and observability. Right? Right. Because our our thought is, look, as a as an organization, right, no enterprise right now knows exactly how the strategy will evolve in the next three to five years. Because, hey, we don't know how the technology is going to evolve in the next three to five years. Right? But what no CIO or a CEO is going to compromise on is the is around security, scalability, and observability. Right? So our point is we wanna build a platform on which you can build towards the future, right, while you almost, like, guarantee your present by making sure you're building it on a good foundation. Right? And and we wanna kind of genuinely kind of flow with how the the customer needs evolve. Right? Because I think we can look. We can we can talk on this for hours because I genuinely feel there's a there's a growing evolution of how, like, actual consumer behavior is changing. Right? Like, I don't know about you, but I use a lot of like, I use more ChatGPT for even q and a than what I used to use, like, Google for. Right? I'm sure it's the case. And we have seen ChatGPT, like, you know, enter the business space where apps are being built into ChatGPT. So the future is not so distant where you will have end consumers have their assistants reach out to businesses on their behalf. Right? So all of these things are changing, but therefore, we want to build a platform that allows customers to be agile without actually compromising on the core of the infrastructure, which is the scalable, secure, and observable platform. Yeah. Well, I'm I'm glad you went down the security path because the more that we've seen, especially with generative coming into the business, there's a very common theme that we run into talking to companies, which is I mean, you know, three, four years ago, the AI security legal council didn't exist. Now now it's staffed. Now you have people with, you know, forms you've gotta fill out and evaluations and audits. Security is absolutely top of mind. Everyone's quite concerned about what happens with their data as it gets absorbed by AI. Right? Are these conversations actually secure? Am I exposing customer information into AI or, you know, vendors gonna use my data to train their models? Right? So I'm like, Cisco's obviously in a unique position because you guys, like, with the Splunk acquisition, with ThousandEyes, AppDynamics, you've got these tools that some of these other companies, like, they simply don't have, you know, natively or within the ecosystem that you guys deal do. So I'm curious to, like, how integrated or are you taking into account some of these new observability kind of security tools as you're looking at how you're applying AI within the whole Webex suite? No. Absolutely. I I think the see, this key this fundamental focus on security, right, it comes from the simple fact it actually comes from two parts. One is, obviously, we, as a company, right, we have one of the largest security portfolios on the planet in terms of, you know, our security business being a leader in that space alone. So there's a lot of research and lot of intellectual capital in the big Cisco world that we are drawing from when we build our products. Right? Now, like, a few specific examples is we have this this framework called the responsible AI framework, right, in Cisco, which where every feature, right, no matter which part of the portfolio it is in, every AI feature that gets built, it needs to be built against the AI responsible AI framework before it gets shipped. Right? So what that means is features are looped, you know, all the way from, you know, language bias, you know, in terms of how features are built so that, you know, the way security and data is handled, you know, where data is stored and processed, all of these things are not handled at a feature level, but there's a framework across Cisco so that no matter which Cisco product you you touch, you know it's gone through as responsible AI framework. Right? Because that's one thing that's key for us because when you kind of leave this at the hand of individual designers, right, you don't have control on who does what and because, you know, it's just not a standard. Right? That's number one. Number two, even if you look at specific to Webex and the Webex contact center or the CX suite, Right? Like, a great example to talk about is all of our LLMs. Right? They they sit behind something known as the LLM proxy. Right? So that means any AI feature that's built being built in the CX world today, be it the AI assistant or the AI agent or anything else that we're working on, right, nobody in the business directly interacts with the LLM. Right? Even though, for example, we use a GPT model as sort of in in a at the back end, we don't interact it with it directly, but we interact through the LM proxy. Now what that what does that mean for a, let's say, a customer? It means that this is where Cisco's taking responsibility for, like, how your data is handled, how your data is processed. Right? Because we're not just saying that, hey. We're just building a wrapper around standard models. Right? Sure. Which, you know, to be fair, most of the market kind of operates that way where, you know, a lot of companies, including us, we consume frontier models like like, you know, GPT. Right? Right. And, you know, you build an ASR and a TTS around it, so it becomes an AI engine, right, which is the net of the experience. But the differentiation is we are saying that, hey. We are taking on the responsibility of how data is handled because all of it flows through the LLM proxy that's managed by Cisco at a central level. Right? Now what does that mean? It means things like your system guardrails, right, enterprise level guardrails. All of these are applied by default on any AI agent that you build. Right? Yeah. Now if you are a CIO right now, you have some level of comfort to know that the obligation or the responsibility rather to secure the AI agent is not solely on the developer himself or herself. Right? So even if you build a very vanilla AI agent and you haven't thought about security, right, you can rest easy knowing that the LLM proxy gives you a level of security and guardrails that will secure you for most of it except for your use case specific. So for example Sure. If you're building a, like, a use case in the, like, the like, medical world where you're, you know, doing prescription refills. Right? Yeah. The only guardrails you'll want to think about is to say, hey. What is it about this use case that is specific to my business? Maybe you wanna say that, hey, don't talk about pricing. Right? Right. Or say, don't talk about x, y, and z. That's a use case specific one. Right? But things like prompt prompt injection, things like, you know, somebody trying to say, hey, give me the name of your last five customers and so on and so forth. Even if you don't plan for it, it is taken care at the system level. So that's examples where, in my view, it comes from the fact that we're drawing from the security portfolio and how things are being built. That's a product example. But it's also like a cultural example where the responsible AI framework was put in place even before we started building these AI products to ship. Right? So from day one, everybody's thinking about this at the build stage, and it's not an afterthought, right, which I think is is is unique for somebody like Cisco, like you said, because we have both the portfolio and the actual, like, cultural experience of building things that are secure. Yeah. Or I I think it's a a hidden thing, a little bit the market doesn't even realize. Like, it's an extremely mature approach to a very complicated issue. You're absolutely right about the guardrails being applied where, the amount of time it takes to be aware of that, to develop against that. I think most companies are still trying to advance the skill sets of their employees to be ready to do some of this AI development. And and, look, at the mow the majority of AI that we see getting applied in customer experience is still I'd say it's I mean, at the such an early maturity stage. It's very front end. Let me build a an IVA concept that does steering or routing for customers, get them to the right place. And, like, just now we're starting to see translation start to come into play in the voice channel. K. So I'm curious, like because I think the foundation that you guys have is great. It also I'm assuming here, and you can correct me. It gives you the ability to take advantage of new models as things evolve. Like, whatever the next, you know, five point o, ten point o model that comes out next year, right, that you guys will have the ability to bring in the next best thing and apply it for for these existing agents that are already built in the product. No. Absolutely. And I think the the the way we do that, Caleb, is so for example, we started with four point the four o model. Right? And now you have a five point one, five point two. And like you said, this will rapidly advance. Right? Right. And this is something that at a at a Cisco level, we are we have access to all of the models. It's not just the OpenAI models. Right? There's the Anthropic model says there's a bunch of models that we can expose. But our goal is, like I said, we expose it behind the LLM proxy so that as a customer, while you have the choice, you're not compromising on the core aspects of security and scalability. Right? Yeah. And the next part really is I feel there will be we are going into a stage where use cases will start to become more and more refined, right, where we will want to like, you know, it's you know, like the saying goes, like, if you only have a hammer, everything starts to look like a nail. Right? Right. And and that's the part, like, with AI, like you said, the initial part was, hey. Like, large language models are the answers to most things. Right? Like, whether you do a basic IVR replacement or you're doing a complex fulfillment flow, it's the same product. Right? Sure. But we are going down the path where we will have more and more choices. Right? And and the choices are also quite complex in the sense you could have choice. Some customers want choices at the level of voices. Right? Like, some customers want, hey. Give me ten different voices where Right. I want the voice to show different levels of empathy, different levels of patience, and so on and so forth. Right? There are some customers, you know, and their needs are that, hey, I need choices in models. Right? Like, I still need large language models, but I need different models because of their performance. Right? And then you'll I I genuinely believe that you'll have enterprises where you'll also have to go into a path where you'll have completely different types of models. Right? It could be an industry specific model. It could be a different scale of a model. Right? So at the end of the day, we wanna be in that position where through this platform, you're able to consume the choice of AI because that makes sense to you. Right? But as a as a as a as a c suite, exec looking at this, right, you are knowing that you're doing this on a platform on which the the base is extremely strong. Right? Like the foundations are really strong. And I know that's what, like, example, at TTEC, you and the teams do, you know that you've got to get the data strategy right. You've got to get the knowledge strategy right. You've got to do all of those things right. So that you are in a place where you can capitalize on the AI developments and start giving real ROI back to your business. Right? But I fully agree with you. I think this is just at the infancy of what we are, you know, almost like scratching at the surface. And, I always like to say that with AI, the more you can do, the more you'll end up doing with AI. That means every use case you unlock, it just leads you to more use cases. Because the first one that goes all the way from POC to production is really the one that will push not just the product, but also the organization to say, hey. We are now ready with AI. Because like you said, like, three years ago, you didn't have you didn't even have teams in the security business that thought about AI workloads. Right? And and now that's completely different. Right? So I think it's it's only gonna accelerate. It's like a flywheel. Right? The more you do, the more you can do with this. So it'll be a very interesting year for sure. Oh, it's it's gonna be an interesting year. And and I'll go back to one of the early things you said about kinda reimagining some of the business processes because the I was I was speaking at an event last year, and I kinda made this, you know, a comment, but jokingly saying that I think you could pause all development with the large language models at this point, and it would still take us about five years to really kind of actually utilize and push the boundaries of what the technology can do. Like, at this point, I really do think it's our own creativity that's the limitation. So as people do start to roll out some of these basic use cases with AI, it does the light bulb comes on. They start going, oh, well, we could do this differently, and we could go use it for this kind of use case. And have we actually thought about changing like, why do we actually have the the customer call in and go through these ten steps to actually get something done? Why can't it be done in two steps? Right? Yep. I think all of these things are starting to get exposed, but it does start to actually make visible all of the data challenges that companies have, to your point, the knowledge management, the content. We were working with a customer last year, and they were frustrated because the responses that they were getting in the chatbot, the the response was really long, exceedingly long. And so we actually went and found the source content for it, and it's like a ten page document. And it's very verbose. And you're like, of course, it's getting a really long response. You've got really long content with really complicated answers. So, like, the AI is not gonna just be able to completely overcome that without you actually having to go in and do some level of work to optimize what's going into the AI. And so there's I think there's a lot of change that's gonna happen within companies, but the ones that are adapting to it, the the ones that are actually trying to use it now are gonna be the ones that kind of accelerate and and take the the leap forward. So Yeah. No. For sure. Getting I'll I'll kinda transition the topic, I guess, little bit because I like to kinda understand what's next. Right? So I think about in the customer experience where we've seen AI starting to get a getting used for what I'll call, like, three major areas. You got the conversational agents. You have the agent assist kinda sitting next to you doing basically, it's listening to a conversation. It's finding content. It's finding answers for you. Conversational analytics kinda being that third pillar, if you will. Obviously, excuse me. I think the agentic piece is definitely going to come more and more into the contact center. We're starting to see that with some back office automation, taking more journeys end to end. But Yeah. From from what you're working on and kind of what you see from a vision standpoint, I'd I'd love to know what kind of what do you project out for me a little bit. Sure. Sure. No. Of course. No. I think the in in if you if you were to anchor this conversation with what we started off with, right, that is everything that we build is geared towards making a meaningful difference to ideally both the customer experience as well as the employee experience. Right? That's the agent experience. Right? And if you see the product evolution, the way it's being built out, if you just take what you've announced so far, like the AI agent and the AI assistant. So even that, while we market it as two separate products, if you really look at how it's built, like, agents that are being built out, you know, are being can be exposed via the AI assistant next to the agent. Right? Now Sure. The the idea is it's one, like, if you can call it that, like a super product that has different surface areas. Right? Yeah. Now I think this is a very important, like, deliberate design decision for us. Because at the end of the day, if you keep the data flywheel going, right, that's where you get the compounding effect of using the same platform. Right? Because if not, if you are a customer, you could just use vendor A for AI agent, vendor B for AI assistant, something else for some because there are a lot of point solutions in the market. Arguably, you know, each one Yeah. Is a leader in their space. Right? But the reality is if you run this for six months, nine months, and twelve months, right, as an example, if you run all of your AI solutions for twelve months, in the ideal world, you should have gathered enough intelligence from these conversations so that every one of your next conversations with the same customer has to be that much more contextual, that much more intelligent. And the customer feels like, hey, I've spoken to this brand before. Because if you really look at it, look, the technologies have changed. Customers still have the same problem. Right? They feel like every time they call into a brand, you know, they feel like they're talking to them for the first time. Right? Yeah. They don't they they feel like they're having, like, like a disjointed number of experiences with the brands. Right? And I think, genuinely, Caleb, I think we're at this inflection point where this could get worse as brands start to make lot of experiments. Like, there are cost like, I've seen brands who are doing a lot of POCs with lot of vendors. There's one POC happening with vendor a in one part of the department. There's something else in a different part of the department. But at the end of the day, it's important to realize that it's one customer that's, you know, talking to one brand. Right? Sure. The customer is not waking up every day thinking, oh, today, I'm gonna have a support experience with brand. Right. They don't wake up and say, I'm gonna have a marketing experience. They just have one experience with the brand. Right? So I think that is a really hard but an important problem to solve, where anything and everything you do, it adds to your understanding, your context and your intelligence about the customer. So that's where if I were to project out what we're building towards, right, of course, we're building, like, core native products, like, in Cisco Live, etcetera. We spoke about we gave the at UN Webex one, we gave a preview of the AI quality management. Right? And goes back to our point around saying, hey. It's not about adding an AI feature on existing quality management, but it's us reimagining what does quality management look like when you have both AI and humans interacting with AI and humans. Right? Because if it's a human only con conversation, maybe four to five percent of calls being quality checked was okay for contact centers. Right? Sure. But in a reality where you potentially could have close to a hundred percent of your calls handled by AI, Right? Hundred percent quality management is becomes a nonnegotiable. Right? So, therefore, the these are the products that we'll continue to build. Even with AI agent, we are heavily investing in, like, a two a and MCP, so on and so forth, because we believe that not only vendor coexistence, but AI agents will only be successful if they can coexist and interact with each other. Right? Right. There is obviously a world where a Cisco AI agent will interact with a Salesforce agent to give an outcome to a customer. Right? Or Cisco AI agent will interact with the sales ServiceNow agent to give an outcome to a customer. So we we wanna unlock those use cases because, like you said, we see use cases evolving beyond IVR replacements and and the simple stuff. Right? Right. But the really cool stuff that we are going to continue to invest in. Right? And and this is similar to the security topic, but not the same, but that is it's foundational, it's important. Right? And sometimes it doesn't show up in the features that get announced, but it'll add to a compounding effect for the enterprise, which is use of data that comes from interactions, right? And using that data to create this flywheel effect where you truly become an intelligent contact center. Right? And there's a lot of focus on that. Because at the end of the day, this is my belief as just a CX practitioner, right, is if you start peeling the onion of all of the products in the market today when it comes to AI agents, you can distill them to saying, hey. Most of them use similar vendors for, you know, the ASR. They use similar vendors for the TTS. They use similar large language models. But the differentiation comes from saying that how do you make all of these work together? Right? Yeah. And make a meaningful, like, a human centric ROI. Right? Because, look, in reality, this is one of the most demo able, trial able technology of our lifetime. Right? Yes. Like, you know, possibly my my son can build an AI agent in less than five minutes with the tools there are. Right? Yeah. But you and I know the trick to getting this into an enterprise and doing it at scale requires us to go back and solve the real hard problems. Right? Like solving AI for voice on PSTN. Right? These kinds of things is where sometimes it's not front and center, but we are solving those problems that are really standing between a POC and a production setup. So that's what the future looks like for us, where we really move down the path where it becomes an intelligent flywheel, where all of our products come together, where the more you do, the platform effect kind of gets unlocked. And that's where we're kind of going towards. Well, that that's that's amazing. I mean, you you clearly have a strategy and a vision for where you're going, which is awesome. I'm gonna absolutely steal your concept of the compounding effect. I love that idea. And I think you're spot on with it that you do you can silo these things. You can fracture a brand experience as a result of because you it's a it's a constant reward trade off, a risk reward trade off of, yes, there's been there's immediate short term benefits of using some of these AI tools, but there's also the downside of it's a very different experience. Every time that you're doing it, it's less personalized than it was even before, and you're not gaining the the centralized intelligence of everything that's going on. Yep. So Exactly. Awesome. So That's it. Yep. Bobby, this has been awesome. I really appreciate you cutting out some time for me to go through this. I'd love to have you back on here as you guys continue to evolve and and release new things. I'm sure six months from now, the world will look completely different. And so I'd love to pick your brain up. I've learned a lot today. So thank you for for doing this. No. I think six months is still a long time, Kailua. I think the world will look different next month. But hey, look, thank you for having me. I'm sure we could have spoken for a few more hours, but it's great to work with you and TTEC in general. But again, I know this is a fun conversation, I'm sure we'll talk more. Again, thank you for having me, I look forward to more of this. Awesome, man. Bart, we'll talk soon. Thank you. Cheers. Thanks, Caleb. Take care. See you.