CX is reaching a tipping point - Interview with Jonathan Rosenberg of Five9

0:00Jonathan Rosenberg:So welcome to the next edition of the Punk CX podcast. With me today I have Jonathan Rosenberg who is the Chief Technology Officer and Head of AI at Five9. Now, Jonathan has dedicated his career to transforming the telecommunications industry and joins Five9 from Cisco where he was CTO for the Collaboration Technology Group, otherwise known as CTG. And Jonathan is well known for his authorship of the SIP protocol which is the foundation for modern IP-based telecommunications. Prior to Cisco, Jonathan was the Chief Technology Strategist at Skype, remember that one? Where he guided this company's technology strategy.

0:36Jonathan Rosenberg:He received his bachelor's and master's degree from MIT and holds a PhD in electrical engineering from Columbia University. Welcome to the... I was like, take a breath there. Welcome to the podcast, Jonathan. How are you doing?

0:48Adrian Swinscoe:Excellent, Adrian. Thank you. Glad to be here and doing very well. Excellent. Now, that was a pretty, I don't know, preeminent sort of introduction.

0:59Jonathan Rosenberg:Is there anything I missed out that you wanted to add? The only thing I guess I would add is also been working on AI.

1:05Adrian Swinscoe:And I will say this, there's a lot of people who just started working on AI sort of like three years ago when ChatGPT showed up. That's, of course, when it got really exciting. but at Skype I ran the research team that actually had a small AI group who were doing like early day stuff on churn prediction and some audio quality improvements and similarly at Cisco yeah I drove the AI program there that led to things like noise removal and face identification and of course here at Five9 also work in AI so so I'd add I'm a 10 plus year veteran of AI technologies applied to telecommunications as well because that's what we're going to be talking about here today so yeah that to the mix well let's get straight into it shall we i mean so

1:49Jonathan Rosenberg:in the setup to this podcast i know that that you said to me that you think that cx customer

1:55Adrian Swinscoe:experience is reaching a tipping point and i wanted to understand what does that mean what do you mean by that the the tipping point i think that's happening here is that consumer expectations have seen like a sharp change the last few months. And those expectations are translating into their expectations about how they're going to get customer service from businesses they engage with. And if those businesses are not providing a customer experience that sort of align with their expectations, it's going to really be paint. Like we'll see increased churn rates. We'll see customers, customers have always been fickle, right?

2:40Adrian Swinscoe:They easily, you know, change around between brands. That's going to get a lot worse. And it's approaching a tipping point because of this exponential change in the improvement experience. And what I specifically am talking about is like, everyone's using AI. They're using it in their personal lives. They're getting a lot of benefit of it. People are now used to lengthy, complicated chit chat conversations with these things that do amazing results for them. So imagine Adrian, you know as a user you log in in the morning on chat gpt and you have it plan your day and make your recipes and do a bunch of work for you and do some research for you and then oh you know you gotta update your cancel your health care appointment you log into your health care provider and you log into its bot and it looks like ancient history by comparison right so it's exposing this embarrassment uh around it and so that we can talk more about like what the exactly users are expecting all these things but but that's the tipping point is that these past generations just not meeting customer expectations and we're entering this new agentic cx era where they you're going to get this incredible customer experience um that and customers are expecting it

3:49Jonathan Rosenberg:demanding it and can you kind of pay as a picture i mean and so we know that sort of i think those things are being driven by the possibilities of this new technology particularly kind of ai generative AI and agentic AI, and also changing consumer expectations and adoption of this kind of technology, but particularly from a consumer perspective, but then that's driving into expectations around what kind of enterprises and brands need to deliver. That's right. How do you think, what we were likely to see as a change, how is experience going to change as a result, and what's going to be the impact on customers, agents, businesses.

4:31Jonathan Rosenberg:I mean, kind of pain is a picture.

4:33Adrian Swinscoe:Yeah. Well, what we'll see is what we're already seeing, which is a rush from enterprises to get on this bandwagon and adopt this AI agent technology to up-level their customer experience. And so what we can think about that from the perspective of the consumer and from the perspective of the brand that the consumer is engaging with. And from the perspective of the consumer, it's going to have a lot of properties that we sort of come to expect from these consumer chatbots. So it's going to be personalized, right? It's going to remember you, it's going to know you, it's going to remember your past conversations and your past purchases.

5:09Adrian Swinscoe:You're going to come to expect that. It's going to be highly conversational, right? No longer this sort of rigid flows. You can chat back and forth, ask different questions. It's really flexible, a lot in the same way that you talk to a human agent. In fact, that's another part of the expectation is like consumer expectations would be, it's a lot more like talking to a real person, except, and this is another thing, I think it's fascinating. It will in many ways be better because it's almost like a super agent, right? Because the human still operates at human speed, right? They can't read a pile of data, analyze a bunch of things.

5:49Adrian Swinscoe:You know, if they want to perform a transaction in a backend system, they have to they'll log into a user interface, click around, put you on hold, do some stuff. That's going to go away. Like these new AI agents are going to be able to collect your data, analyze it, make recommendations, do things for you. You'll get these superhuman experiences. We're already seeing them, right? We're already seeing them in our consumer experiences. And those will translate into the consumer direction for the prince. That's the consumer side. Okay. And then I'll just touch on the enterprise side. We can talk more about that too.

6:22Adrian Swinscoe:what the enterprises are going to see is they're going to see the two things that they've always wanted. They're going to see reduced costs and increased customer satisfaction. Like all things distilled to that. And they'll see that from reduced need for human agents, they will not go away and it will never go away. And it's a terrible idea to think that they're going to go away, but there'll be less need for them. And they'll be handling much different types of engagements that are higher value, more emotional and pathetic, but they'll get a lot of reduction in cost. And of course, customer experience will be great.

6:54Adrian Swinscoe:And the ultimate pinnacle of this, my prediction is the best of breed in businesses will perfectly blend the human and the AI agent experiences together facing the concern. That's going to be where the best success is.

7:09Jonathan Rosenberg:So I want to kind of just pick up on something you mentioned. We talked about most of these experiences will be conversational. And I wanted to ask, get your take on this because, you know, so I look at this sort of stuff and I just think about it from my, particularly from my customer's experience and also from the things I've gone before in terms of technology and things. And we're entering into an agentic era before the last kind of few years has been around the use of how do we harness the potential of generative AI. And prior to that, if you think about particularly chatbots, it was very much rules driven type of interfaces.

7:45Jonathan Rosenberg:however those rules driven interfaces were quite good at doing sort of predictive sort of stuff i looking at sort of demand and going actually are you are you calling us about x y and z or our data tells us most people are talking about x y and z now if you think about like a customer's experience actually typing something or it's either you know people are getting better at introducing voice interfaces to this so which is then doing you know real-time translation but it's quicker to read something and then choose something that is to speak something or actually type something and so are you also kind of seeing a maturation in this of the design of these interfaces and because sometimes i i'm i'm frightened that we throwing the baby out of

8:32Adrian Swinscoe:the bathwater type of thing yeah exactly what i mean absolutely so i think the future is what we've been calling multimodal. And again, I think this isn't here yet, right? So this is looking forward to, I think, where it's going to go. Is what'll happen is we used to think about it from the perspective of channel choice. Like as a consumer, I could call or I could chat and you could pick, yes. And it was one or the other. And what we're now going to see with these new AI agents is that you and a turn by turn basis can speak something, type something or send something. And similarly, in the opposite direction, the AI in return can speak something, send you a textual message or send you visual content.

9:17Adrian Swinscoe:And I think what we'll, and so that first of all, will allow us to use the right modality. And so the example I like to give is you could be speaking with an agent, you know, and let's assume you're doing it like in a little web widget in the web page, right? You're speaking to it through your computer and it says, oh, what's your account ID? Instead of now doing the annoying thing, which no one likes of like speaking it out, you can go just copy paste it and text it and or you could speak it. And so that's something that doesn't have to be pre-programmed either. Like the user can be flexible about how they present data.

9:53Adrian Swinscoe:And then in return, certain information, Adrian, you said this were important, is better represented visually. And my favorite example of this is a seat picker on a flight. I guarantee you, you will never get an experience just with voice doing a good job at that. It'll be terrible. You need to look at where the seats are and is there someone in the middle? So you can think of that as a form of visual content, like a widget that the AI will be able to show to you that sort of starts to blend together the conversational experience of a chatbot and the traditional web experience too, so that the visual elements are integrated when they're appropriate.

10:34Adrian Swinscoe:And then you'll click on the seat clicker to pick your seat, and then the AI will say, got it, Jonathan, you picked seat 5B. So I think that's where it will go, and that will solve this problem you just raised, Adrian, which is that we can use the best modality for collecting and exchanging the piece of information that needs to be exchanged

10:53Jonathan Rosenberg:yeah because it strikes me that you know the all of this all of these systems are um contingent on their success it's contingent on the data that they get and it's always obviously if you get garbage in you get garbage out and in many and many and so the more that you can control the front end and the quality of the data that's going on in the front end and the more you can do to enable people to give you accurate information like i sort of imagine a time when the system says oh we're seeing a lot of uh people reaching out to us today about x y and z and that automatically populates into interfaces that are you calling about x y and z can we fast track you into kind of like those sort of those sort of solutions because it's then it just makes oh you know what i'm talking about i don't have to explain myself this yeah this is kind of fine and that translates into ease and convenience and so on and so forth with the customer.

11:48Adrian Swinscoe:Absolutely. And that's sort of this personalization dimension where, again, I think generative AI is so good at this, where you can just give it all this context about the user, about the company, what's going on right now. Just give it that context at the beginning of the call and it can have this incredible personalized response because it sort of remembers all this information. And it's not that you couldn't do that before. We've been building AI systems that have had that kind of experience for a long time. It just required an immense amount of coding and configuration and having to predetermine all these conversation paths.

12:28Adrian Swinscoe:And that's now fixed by generative AI. And so that's what I mean by this tipping point. It's see this massive personalization in our consumer experiences customers will demand it from their businesses and we now have a tool to generate it and that's really exciting yeah excellent i mean

12:46Jonathan Rosenberg:but then many of our listeners might be kind of like thinking you know this is all great it sounds really kind of like exciting and pretty easy to say but probably quite a lot harder to do yes and there's a whole bunch of research out there i mean crumbs there's the stuff from mit and there's a stuff from which is on a small sample size but a more robust sample size i think there's research coming out of ibm and accenture which is saying many of these organizations are struggling to implement things at scale they're struggling to generate an ri from their investments and they're still they're still caution about some of the risks i mean thinking about that i mean what are you seeing from the people that you're working with at five nine from companies and what are they doing well and or right in order to harness their potential with this new technology?

13:40Jonathan Rosenberg:And also, what are they not doing?

13:41Adrian Swinscoe:What are they avoiding? Yeah. In some ways, this is the stuff that hasn't changed and shouldn't change. And I think this is forgotten in our enthusiasm and exuberance about AI, which is, listen, AI still makes mistakes. It's inaccurate. It's got, in some sense, got a little worse with generative AI, with this hallucination. They can make just make stuff up. But we've always had accuracy problems. And the way you combat that is with sort of a rigorous process. Right. And so even we have a name for this at Five Nile, like the AI blueprint process that we take our customers through to help make sure they deliver success.

14:19Adrian Swinscoe:And it's things that make sense. Identify the use cases you actually want to solve. Don't do all of them. Focus on the lowest complexity, highest return on investment cases. And those don't have to be this science fictiony, amazing stuff we just discussed. You know, Adrian, the amount of people who are still doing password reset by like an old, like by calling up and talking to a live human, like if you're still one of those people, that's a good start, you know? And you don't even need you to do that. You'll get a big, so take these cases and identify what they are, quantify how much success you'll get from them to determine your metric.

14:58Adrian Swinscoe:Then the next thing you have to do is identify what data sources you need, what are the use cases and the types of conversations you have. You have to plug in those data sources. I'll tell you, this is another thing often overlooked.

15:10Jonathan Rosenberg:Yeah.

15:11Adrian Swinscoe:For success is a lot of the time we spend building AI for customers is like plugging the API into the thing. And it turns out, oh, we don't have an API or the API doesn't work. or, oh, you know, to get a credential on the API system, it takes, you know, you have to go through infrasec approval. Like all these are real problems. Like, and you can rework those. You have to get, you have to focus on that, make sure you understand your data sources, make sure they're available. You can connect to them, define what success is, have a good test suite, you know, a good rigorous process of acceptance test, incremental deployment, measure your results, see if they are there, adjust your performance of the system and tune the models and the ai to get better results and then have this continuous loop of monitor improve deploy monitor improve deploy running yeah and and that's what you need and everything i just said is as true 10 years ago when we started building ai systems as it is in the era of generative ai in fact it's more important in the era of generative AI.

16:16Adrian Swinscoe:I think that's absolutely right. I mean, it's like, I think it was a systems

16:19Jonathan Rosenberg:thinker gentleman called, I think he's Robert Checkland or something. And he had a mantra was like, check Blandu, check Blandu, check Blandu, check Blandu. And that's the thing, because we are, we have to think about these things systematically. And so I wanted to actually go back to the kind of the AI kind of blueprint sort of approach. I mean, you articulated the different approach, but I also wanted to sort of dig into the bit that I know that you do at the front end which is sort of rather than actually going here's a spanner which is a particular kind of like type of tool for a particular use case and go and then try and plug it in but as part of your blueprinting you actually don't go tool first as it were you almost going to look for patterns kind of like first and that tells you kind of how to do it because actually this is the the trick i think is it leverages the power of generative ai and those things to actually find the patterns.

17:15Jonathan Rosenberg:Yes. And then you then know where to start thinking about deploying the tools. And that feels like a thing that many people don't get right. They almost say, find a spanner and then go look for a knot to tighten rather than stepping back and going, okay, what's the landscape and what's the patterns?

17:30Adrian Swinscoe:Exactly right. So we have a product even that we use as part of the blueprint process called AI Insights.

17:35Jonathan Rosenberg:Okay.

17:35Adrian Swinscoe:It uses generative AI. It analyzes these calls and conversations and identifies what are the call drivers, clusters them together. correlates those with handle time and costs, and then allows you to deep dive in, you know, also understand other metrics like customer satisfaction, initial resolution rates, so that you can zero in on exactly what are the things that I should go and automate, you know, on this platform. So this data-driven approach that helps you quantify it is really, I think, the right formula for doing this.

18:09Jonathan Rosenberg:Yeah, I mean, I've been thinking about it or almost like characterizing it with uh do you know the remember the old fable of the back the tortoise and the hair of course yeah and i think it's it's it's it's beautifully illustrative of what's going on right now you've got everybody that's trying to be a hair yeah so but it's but it's the tortoise the tortoises that are winning because they're just being very deliberate very kind of like you know considerate about what they do and they don't know and ultimately they win the race yeah yeah i think

18:39Adrian Swinscoe:That's a great analogy. Let me hearken it back to a tortoise, a tortoise cautionary tale, a hare cautionary tale, I should say, which is we had this financial services company that I won't name them, but you probably know the story where they made some big, bold statement that said, we're going to get rid of all of our agents. Oh, yes. I guess it is. Yeah. We're going to file them all because A is going to do everything. And they reportedly did. and then some months later they came back and said oh you know that thing that was that thing that we said yeah you know we brought the agents back because we think we're going to focus on delivering our great cx so that's now i don't know whether that whole thing was just a pr stunt or what if it did was brilliant but um the uh it wouldn't surprise me if they did in fact try and fail on this and i think that's sort of this cautionary tale of getting the exuberance ahead yourself.

19:36Adrian Swinscoe:And I think we are moving faster and this technology is allowing us to move faster, but you still have to do all the stuff I just described. And it's still, even with this new generative AI, people are taking, it still takes time, usually weeks, months sometimes to get these implementations and going and go through these processes to make sure you're getting the results you need. And you can't forget about that. And we're seeing it because I'll tell you what I'm seeing another thing i'm seeing agents these punks these punks and this in this case not a good way these punks that go and they they find some little startup that's got a website with we can get your customer support agents built today and they type a few words into the instruction editor you know and they call it up and it talks to them and it sounds great and it does something like oh my gosh this is amazing they go to their boss like boss we can fire all the agents i found it it's this thing over there and we're we're seeing that we're seeing like these sort of like you know and then kudos to them trying to use this new technology to disrupt but they get they get it's so easy to get a good demo out of this stuff it takes like nothing to get a great demo yeah and but but all the rest of it gets you know you still have to go do that and that's actually what matters