Joe Rittenhouse, CTP & Ram Rajagopalan, Zoom | The AI ROI in Contact Center Summit

AI has been part of the contact center conversation for years. You know, initially handling repetitive tasks such as call routing, basic self-service, and and simple customer inquiries. Today, however, AI is moving into more complex workflows that can improve agent performance, automate multi-step processes, and reshape the customer journey. The opportunity is significant, but success requires more than just deploying another tool. Organizations need the right use case, trustworthy data, a pragmatic implementation plan, and clear metrics for determining whether AI is delivering value. Welcome to the CX Summit. I'm Bob Laliberte, principal analyst with The Cube Research. I'm joined by my co-host Zias Carafallah, principal analyst and founder from ZK Research. Welcome, Zias. >> Hey, and thanks Bob. Great to be back here. >> Absolutely. And, you know, to help us explore how enterprises can move from this experimentation to measurable results, we're joined by Ram Rajagopalan, head of product AI, Zoom CX at Zoom, and Joe Rittenhouse, co-CEO, Converge Technology Professionals, uh a Zoom partner with extensive experience implementing contact center and AI solutions. So, Ram, Joe, welcome to the summit. >> Thanks, Bob. And uh Zias, thanks thank you for hosting me. Uh it's a pleasure to be here, and Joe, uh good to see you again. >> Good to see you guys, too. And thanks for having us. >> All right. So, now let's have some fun. Ram, I thought I'd start with you, and, you know, begin with the evolution of AI in customer experience. So, a lot of organizations started with relatively simple capabilities, as I mentioned, right? Routing calls, answering their frequently asked questions, and things like that. Any kind of repetitive interactions. But, how are customer requirements evolving today? And are there more sophisticated use cases that are becoming practical today? >> Yeah, that's a great question, Bob. Um yeah, you know, the industry is evolving rapidly um with the advent of Agent Tech platforms in in in especially in voice agents these days. Traditionally, you know, many customers even about a year 18 months ago many customers focused on uh purely measuring containment. Um but where I see conversations these days are going from not just looking at containment, but end end-to-end resolution of uh consumer inquiries. So, this is what we are calling within Zoom uh conversation to completion where we're not just looking at um just answering the basic inquiries, deflecting the call from going to a human agent, but actually uh completing the task that the consumers or users are calling into the uh calling into for support. Um and we're also seeing use cases evolve from basic uh inbound calls to also going towards more automated outbound calls, outbound dialing navigating um complex IVR menus uh things like that. Um and also we recently at Zoom uh this summer we even introduced um a new uh you know, uh you know, packaging called uh in outcome-based pricing, which not only just measures the call volumes or consumption, but also measures the outcomes that we are producing for customers through virtual agents. And then customers have an option to choose that if they want to just pay for the performance. So, we see you know, use cases evolving um both with inbound as well as outbound. >> Hey, Bram, the you know, the entire industry is moving to this vision of rapid gentle compacting CX. And I do think the conversations to completion narrative that Zoom has been working with is unique, right? And so, can you maybe give an example of how that might work in the contact center particularly for a more complex type of interaction where maybe you need to bring in a human agent as well. >> Yeah, I mean, I can tell you a couple of examples. We have one customer of Zoom who's using virtual agents today to perform, you know, political surveys in in a within a country. So, this is virtual agents going through a list of residents within county or a state and then qualifying by calling them out by making an outbound call, asking some qualifying questions, and then taking them through a very complex survey of questions where each question depends on your response to the previous question. So, this is not just completing qualifying the you know, the caller, but also going through a complex survey questions, but writing all the results into a platform from which you can then create a complete comprehensive report around what were the interests for this case for example, they were doing a political survey, where do they lean? Are they leaning left? Are they leaning right? Etc. What are the key topics, key issues that that they are that the you know, the population cares about and getting a full end-to-end report at the end of the day is is a fairly complex endeavor. If you think about it, sometimes you know, we call people go into voicemails. So, so the virtual agent needs to be intelligent enough to detect when a voicemail is encountered and leave a appropriate message in the voicemail in the voice message. So, all of this is one example of how we are kind of navigating complex use cases and we're producing not just um, a completion record, but uh we're giving a full report of what was the outcome of those calls back to the customers. So, this is where I think um, if you know, we are seeing use cases are evolving and in health care, too, we're seeing another uh place where we're seeing evolving where we're making outbound calls uh to remind patients of upcoming appointments, prescription refills, and even qualifying, you know, connecting between health care providers and insurance companies where uh we need to navigate complex IVR menus. All these use cases are becoming more and more common and more and more common these days. >> And Joe, I want to bring you into the conversation here and one of the reasons I like talking to partners is because you're actually measured on whether getting this stuff to work for customers, right? And we've talked about this a lot where um, and the executives of companies today are under tremendous pressure to deploy AI and move quickly, but that rubs up against the concerns of a contact center leader that might be concerned about disrupting customer service or perhaps just trying to do too much at once. And so, when you work with customers, how do you help them identify the the best first use cases for AI to help them get started? >> Yeah, and it it also disrupts the IT team, too, right? It's it's it's the disruption of the business, right? And so, and and there's an anxiety, there's a push, but usually where we try to start is where's the low-hanging fruit? And it's in areas that you're usually not even typically thinking about of like and it it could be and we see this from all different verticals, from health care to banking to professional sports to manufacturing. Just a simple question of like, what do you do after hours? What do you do after hours today? Well, we're usually staffed, we follow the sun from the East Coast to the West Coast, but then after hours, we just have a general mailbox. How many calls do you get? What type of calls do you get? What's happening after hours? And the answer is typically, I don't know. Well, we can see that volume. We can project that volume. And so, let's just go put a containment out there and go see what's happening. And some of the results we've seen, especially in the medical industry, of like just being able to book appointments or being able to have a follow back with a nurse or something along those lines, those are instant savings and instant revenue contributions. But what that's not what we're looking for. We're just looking for a place that you can start simple to get these tools in place, have a pragmatic approach, get everybody comfortable, and get the organization used to having a choreography of how do we solve these complex business needs because it's not just the tools that are answering the stuff or the questions. To Ron's point of like the integration of like maybe we want to write this to a ticket to Salesforce the following day. So, when an agent comes in, that's the first thing that they do. And in the methodology and the coaching that we have during that pragmatic approach is to say, as we get this in, it's like getting a new pair of shoes. You feel like you can run faster. You're going to want to do this. You're going to want to do that. And we can do all those things, but we have to prioritize and we have to have a plan. And as long as we have a plan and we're working within the business, we can execute. We can start to take on those larger and more complex projects. But this is an evolution and it continues to go post deployment. And it is not something that you're just going to set it and forget it. And so, you got to get used to that mentality for the business, too. There's change. And this orchestration takes every level of business. It is not just IT. >> Hey. Hey, Joe. Because you're involved in the deployments and so forth, I'm wondering if you could share some knowledge with organizations out there that are thinking about adding AI and trying to do this, you know, I'm wondering if you could talk about some of the common mistakes that you see organizations making, especially when they're trying to boil the ocean and they want to do everything at once, and and how to overcome that. >> Yeah, and I I I think the the first question that we ask and and where we're seeing some some clients have some success is, do you have an AI committee? And the answer is about 50/50 right now. And then the 50% that's saying, "Yes, we have an AI committee." It's like, "We're starting to put it together." But the fundamentals are there and they're starting to put it together. The groups that don't have an AI committee, it's for us to sit down and be like, "How are you managing your AI strategy? Like, is it top-down? We just have to do it. We have to do it." Or is there an orchestration? And the reality is like, you can't boil the ocean. And there is a lot of orchestration between the various departments in what you're doing. But to to present to the C-suite of like, "This is the lift internally you're going to have to do. These are the requirements that your teams are going to have to participate in. They're going to have to drive this. These are the metrics we're driving. This is how we're prioritizing it. Do you guys approve of this?" And then these are the deliverables that we'll give. The reality is is if we can't get that, that's not an implementation that's going to have success. And so it's it's more on the front end of the organization of like, we just have to re-kind of teach of how do we want to handle our AI strategy? And there's an orchestration level from the C-suite to every department. And this is not just an IT project. This is not just a CX project. And there's a lot of overlap. And so you just have to have visibility of what's going on with the organization and have an orchestration. Start small and go from there. >> Got it. And you come in as the voice of reason that can help organizations get through that and get down that path. >> Yeah, I think our pragmatic approach is like, we're not here to sell you the product. You already want to buy the product. But how are you effectively going to do this and grow it? And And it's it's going to evolve with you. And so it's not just a day one, "Hey, we're done. We're collecting revenue." This is an ongoing process and you're going to evolve with it. >> Excellent. Now, that sounds great. Uh Rob, I wanted to come back to you. You know, when we talk about, you know, contact center data, right, really fragmented across virtual agents, human interactions, workforce management apps, right, knowledge bases, customer systems, all sorts of different different places. How does your integrated CX platform help customers connect those interactions and turn the resulting data into better outcomes? >> Yeah. That's a good question. I think Joe touched on this a little bit. Um, you know, one of the things that I think many many, you know, as customers consider about their AI solution, uh, you have to think through the entire customer journey end to end. Uh, you have to think about the journey, uh, as they start. I call it the before, during, and after, but, uh, before, during, and after they interact with a human being. But, you have to think through that entire process in terms of how, uh, what are the kind of use cases where a virtual agent can best serve a customer. And then when does it get, um, you know, elevated to a human being and what happens after that uh, engagement is concluded. And AI really plays a role in all of these three segments. And many times customers tend to think about, um, you know, many there are many solutions out there. Many vendors bring in very, um, you know, point solutions for each one of these stages, either it's before, during, or after. But, the challenge that they will run into is, uh, the data common data layer that connects all these three. You want to see the trans the entire life cycle of an engagement. When the customer is with a with a virtual agent and what happens afterwards. Uh, with the Zoom uh, CX, this is one area where I think we tend to uh, have a unified layer where a customer can see the performance of a virtual agent and see it side by side with a human agent. We have quality measurement tools that evaluates virtual agent and human agent side by side and tells you uh, what was the CSAT at each stage, what was the, you know, sentiment at each stage, etc. So, you have a automated way of evaluating both the human agent and the virtual agent. Another thing which I think Joe touched upon was, you know, when we speak to, you know, particularly large enterprises, like you said, this is this is an area which touches multiple subgroups within a company. Because, for example, for a virtual agent, you need to provide it context and knowledge, right? But sometimes the knowledge is maintained by a different group and there is often disconnects between the validity of that knowledge. Maybe the knowledge base is not up to date or is not updated or there is conflicting information. But when you inject all of this into a virtual agent, it the responsiveness of that agent can determine can be determined by the the quality of knowledge it's been provided. One of the examples that we ran into with a large consumer products company was the knowledge base was quite not, you know, it was not up to date or it was not updated for a for a virtual agent to understand. In many cases, we were kind of escalating the call to a human being only to find out that the human beings were using their own in, you know, tribal knowledge or knowledge that they're gaining through experience, which is not documented elsewhere. But being part of the same platform allows a virtual agent to see not just its conversation, but the conversation that is handled by a human agent and learn from that conversation. We look at that, we look at, you know, whether is there a gap in the knowledge base and then surface it back to the human administrators to see if that gap can be addressed by looking at the human transcripts and then feed that back into virtual agent so that they learn from that. So that in the future, a similar question comes up, they don't have to elevate it to a human being, but they can answer it themselves. So, that feedback loop is very critical and that's what one of the uniqueness of the connected CX platform from Zoom, where the virtual agent and human agent are always talking to each other and learning from each other. >> Hey Ron, there's there's more though in a contact center than just Zoom, right? You've got your application, there's a lot of third-party CRM systems, things like that. And so, talk about how you integrate with those third parties um just to make sure that there's a complete view of customer experience and companies can get away from the silos that they've had for a long time. >> Yeah, absolutely. >> [clears throat] >> So, you know, many customers have different uh system of records. It could be a ticketing system, a customer information system, um and the and to provide a more customized experience for every time a call your a customer of yours is calling into your uh contact center, we need to know who you are, what kind of uh you know, uh previous history that you've had with the uh with the with the brand. And virtual agents can look into that by, you know, we have nearly 40 out-of-the-box integrations with all kinds of uh you know, system of record from Salesforce, Microsoft Dynamics. And in the future, we'll have our own data layer, which kind of remembers the context of the caller. If they call back in the last 48 hours, there's a history of conversation that we need to keep into account. And using that in a way to kind of provide a very tailored experience back to the caller is becomes very unique, and this is what customers expect. Uh even as a consumer, I would I would feel I would feel happy if if I'm speaking to an agent that knows my history. Uh I so that I don't have to repeat myself or I can continue from where I left off. So, all of that uh needs to happen within like say, between uh I would say between 500 to 2,000 milliseconds of a time of of every turn. We need to be especially on voice conversations. That is the latency through which the virtual agent has to learn and respond to a customer. And we have a number of connectors, but we also support many custom scripting uh within the product. So, we have a lot of choices. It's a very open platform. And for customers, they can also rely on people like Joe who can come in and hook it up and make it even more customizable and more personable for customers. >> Yeah, that now let's go back to Joe. Joe, you talk about the importance of picking the right use case to get started with, right? So, once you help a customer walk through that and understand the right use case, typically how long is it before the customer gets it up and running and then it actually starts demonstrate value and and a return on that investment? >> Yeah, with the low-hanging fruit methodology, that's why we start there because it's it's easy to get it in place and to to [clears throat] start capturing the art of the possible, right? And so, what that also does is it's just starts to get the wheels in motion for everybody dreaming the art of the possible, which can also be dangerous, but it's it it just really starts to grab momentum. And so, like on the low-hanging fruit analogy of like after-hours messaging and containment, it's going to continue to evolve, but you're just going to put basic containment out there of like what can we do? What do we think they're asking for and how can we do it? Roughly start to finish, that's 30 days from start to finish where you're going to start seeing containment, start seeing data, and start seeing results. There was one analogy that we did for a physical rehabilitation facility and all their physical therapists were answering the phones. And so, we asked what happens after hours and they're like, "Nothing." And I'm like, "What are people calling for?" And they're like, "Well, they just got out of the ER at 9:00 at night. They broke their and they were told to call us. And if they don't get us first thing in the morning, they're just going to call the people next to us and book with them." And just putting that containment on that front end resulted in millions of dollars in revenue. So, you just don't know what you're going to find, but you just start small. But to start small, roughly about 30 days to launch. And then from there you go. And then from there it gets into a conversation of what are we going to prioritize and what's the most important and how can we leverage this tool more? >> And that's usually leading to data integrations and all that stuff that Ron was talking about. >> Yeah, yep. >> So, Joe, I like your your example of the the containment. I'm thinking about as organizations are doing this and Zeus was talking about how do we get to value? Are you seeing with AI the metrics are changing? Is it, you know, resolution rate? Is it, you know, handle time? Customer sat, right? CSAT? Is it agent productivity? Is it cost per interaction? For the organizations that you've worked with that maybe are a little bit mature than others, are you seeing a trend in the metrics? Are they are they changing in what's what they're looking at to define value of the of the solution? >> Yeah, they the the easy button answer is all the above, but the organizations that are a little bit further along and more mature are are really making some good strides. And a big part of where some of the other early adoption started was just over consumption and not understanding pricing and volumes and understanding that there are consumptions. Like there's outcome-based consumptions. There's metered consumptions. There's conversation consumptions. There's There's different ways that you have billing. And so, understanding like what are those results? To Ron's point of like outcome-based rates, like what defines a successful outcome? Like you have to be extremely detailed of what you're contracting for. And so, what we've learned from those mature more mature organizations as they continue to progress is really understanding their volume and understanding their pricing models because a lot of people through the agentic practice that have ballooned up, they just got surprised costing of like, "Yeah, it's great. Yes, we have all this containment. Yes, we're driving revenue, but oh my god, my bottom line has gone up so much." And so, it's because there wasn't a pragmatic approach to understand like what is the volume going to be and what is how is the pricing outcome going to be related to. So, the more mature organizations are really focused on one, we know we can execute these things, and we know we're going to get these returns, but let's be extra critical of what we know our ongoing costs are going to be. >> Got it. Great. And I'm sure that's something you're helping organizations with, right? >> Yeah. >> Yeah, exactly. That That's where we come in to just kind of sit down and and help kind of break down the puzzle piece. The product's going to work. What are we going to do, and how are you going to accomplish it, and how are you going to budget for it? Because the reality is it is going to grow. >> Yeah. Yeah. But, if I may add, one of the things uh we are seeing, and I think it's becoming more and more mainstream, is um you know, definitely And this is one of those products where we have uh extremely data-driven, right? You have the call volume, you have the uh handle time, you have the containment uh rate. Uh but, we are going further uh and going beyond Some customers are happy with that, but some customers want to go beyond that. And when it go when you go beyond that, we look at resolution. And resolution is measured in multiple ways, right? Like, every customer has a definition of resolution, right? And it varies from industry to industry, vertical to vertical, customer to customer. So, you want to have a platform that kind of scales with that, and also works out of the box. So, we have what we call as uh implied resolution and explicit resolution. Yeah. Explicit resolution is at the end of the call, uh you know, we have a two-question survey. Hey, did we solve the call? How do you rate it from one to five, you know? So, you get a very very uh objective score, and yes or no answers, through which you know whether you solved the customer or not. This is one way, and this is kind of tells you, but it's not everybody's going to stay at the end of the call to give you a survey response. So, that's one signal. But, we complement that with what we call as implied uh resolution, which using an LLM to evaluate the outcome of a call. So, a different LLM to evaluate uh the outcome of the call, wherein you can we have some you know standard signals, wherein we won't don't want to call customers being frustrated and dropping off as a call resolution. It is certainly contained, but we didn't resolve the customer and we didn't answer the question from the customer. And sometimes we provide an answer and the customers just you know they are they might get the answer and they drop off. So, we have built a composite metric which kind of takes into account different behaviors patterns and at the same time we also also allow customers to customize and define what are the key signals that they want to get from a call and allow us to measure that automatically for every call. So, you have a very thorough way of measuring not just containment but also resolution. Then you combine combine this with CSAT and sentiment score, you have a very you know deep level understanding of how the virtual agents are performing in the contact center. >> Okay. Now, I think that that's great and it actually leads a little bit into my my next question I wanted to ask you, right? As as these agents, right, are taking on more complex tasks, interacting with all these different systems, right, you're coming up with different metrics for them to track, but organizations also want to have confidence that everything's performing correctly across all these everything. So, how are you helping also customers build, test and and continuously improve the agents before and after they're in production? >> Yeah, that's that's that's another excellent question and I think Joe uh touched upon it. This because this is not once you deploy once and it's it's done. Um so, you know, you know, a few even about a year ago, everybody was rushing into how can I build an agent you know, that is voice agent or a chat agent and deploy it very quickly. The testing was manual calls or manual engagement you do you know, your team maybe makes a few calls, see whether Uh, agents are performing more or less the way you want it and then you go into production. But, where Zoom is going and where we see the future is, uh, we we want agents to do two things. Be able to solve complex problems and be able to build these agents and test them, uh, programmatically and build the confidence. What I mean by that is, we want to go away from just making manual testing calls to programmatically using LLMs to evaluate LLMs, uh, for example. So, we have a voice agent that, uh, makes programmatic calls based on the goals and the criteria that you define and the outcomes that you expect. And then we we run that with a simulated number of calls against the virtual agent that you have built. And then we have a scorecard and a and a metric that tells you how the agent is performing. So, you can really stress test the agent and then and you can test test it under different conditions, accents, background sounds, etc. And then see how the agent is performing. And then we also have recommendations on if there are gaps in knowledge, for example, or gaps in the way certain tools are being called, we make we identify those gaps and recommend what kind of, uh, remedies can be put in place in terms of prompting that can be adjusted into the agent guidance. So, this is something that we've, uh, championed and we're putting it out there for customers to test. And this is coupled with the native AB testing that is already available in the product. So, uh, you can really, you know, as you evolve from one version of an agent to another version with newer models, uh, you can test it along with, uh, you know, our built-in AB testing where you can split the traffic and test the outcomes of the agent and then decide whether are you ready to upgrade to the newer model or go to a newer version and do it, uh, you know, uh, in a very methodical data-driven way. >> Got it. >> Yeah, Joe, you know, we I want to wrap this up with, uh, some advice to to users. So, there's a very fine line between AI success and AI failure. I think everybody's got good intentions. And so, you know, you've done a lot of implementations. So, so from your experience, what separates the companies that deploy it and generate meaningful value to those that start the projects and then stall or fail? >> Yeah, I I feel like I'm kind of a broken record on that, but it's orchestration. But, I'll go back to to another layer there of like where we're seeing successful organizations is there's and you know, in large part when this AI stuff came out, there was a there was a fear that it was going to replace all these jobs, and we haven't seen that. We're enhancing the the the the response we're giving to customers, we're enhancing the ability to drive sales. We we haven't seen a a replacement of jobs, but what we have seen is a a pretty significant creation of jobs in project management. And it's project management within the organizations that report to the C-suite that manages their AI infrastructure. And these PMs within these organizations that are really executing are the orchestration level between the various business units and understanding requirements and prioritizing and understanding getting back to the AI committee of like what projects are we going to take and then that the PM is responsible for making sure that the appropriate parties from the organization attend the meetings that are required to execute these projects. Because that's usually where most projects fail. Is there's a complete misunderstanding of the commitment of time that your business is going to have to put in here to Ram's point of what happens from start to finish and containment. And there's multiple departments on almost every [clears throat] implementation with different knowledge sets. And all that has to be choreographed. And so, how do you coordinate that orchestration between the business to effectively affect change? And it's measured. And so the the the organizations that we're seeing that are really starting to thrive have a PM based model now that reports to the C-suite that is the choreographer of the business and they own the projects. And that's a significant step that wasn't there before. You always had PM potentially in IT in larger enterprise, but now we're seeing it in the in the mid in the major market, too. And that's the difference in success that we're seeing. It's execution. These tools work. You just can't boil the ocean. You know, we use the analogy of you eat the pizza piece by piece. You don't fold it in half and scarf it down, typically. But it it's one one bite at a time. Start small and it'll go. >> Yeah. And then Ram, to to wrap up the let's pick up on the thread that Joe had about the human still being important. And so companies are going to have to manage their agentic deployments um as well as managing the relationships with the human agents as well. So, when you look ahead, um how do you expect agentic AI to change that relationship between human and virtual agents? And give us some advice on how companies should think about managing them. >> Uh yeah, definitely. I think um you know, you know, we have to think through this in two two ways, right? Like like what Joe said, um this when we you know, with virtual agents, you have an opportunity to serve more customers that are who are calling in or engaging with your brand either online or on the call. Um what we are really seeing here is the transition between virtual agents and human agents. So, when we transition or when the teams that are building these agents, whether it's building agents for uh for for the virtual virtual agents or for human agents, having that coming a common layer that kind of you know, where you're dipping into the same knowledge source, you're using the similar tools that are available between human agents and virtual agents, you basically reduce the burden on the IT administrators who are managing these uh the agents that are being deployed for different stages in the CX you know CX flow. And what more importantly when the call or the engagement is transferred between between a virtual agent and human agent, you want to transfer the full context. The call, the reason why they're calling, the context of collecting variables that you need to pass it on to a human being so that you know that human being who is answering or helping that customer can get on with the job and get it done without having to repeat themselves or look at 10 different places of records to just to answer a simple question. So, that relationship is critical and being and this is a two-sided story where you have a customer facing experience and an admin facing experience. So, as customers think about deploying you know agentic AI within CX, they have to think about one system where they can have one place to manage the entire AI for the CX and with a common infrastructure, common responses from you know from the language models, common set of tools that is deployed between virtual agents and human agents. >> Well guys, hey, I really appreciate this. This has been a fantastic conversation. So, thank you Ram, Joe for joining us. >> Thank you Bob and Joe. >> Yeah, thanks for having us. This was fun. Thank you. >> Absolutely. And you know that the key message is that contact center AI doesn't have to begin with a massive transformation, right? Organizations can start with a focused use case, establish measurable goals, and expand as you gain experience and confidence as Joe repeated multiple times in this call. >> Yeah, no, but at the same time though, it's important to understand that long-term success does depend on more than just automation. Enterprises need to connect the data and reliable knowledge, disciplined testing, and an architecture that can support an increasing amount of different customer journeys as well as that combination of human and virtual nations. >> Yeah, absolutely. I mean it's clear AI is raising expectations for both customer experience and operational efficiency, right? And the organizations that can create the greatest value value will be those who move with purpose, really start pragmatically, measure the results, and continuously improve. And you hear heard today about how the tools and the solutions, the platforms enable you to do that. So, again, thank you all for joining and thank to everyone watching this CX Summit, and stay tuned for more.