I think it's a choice in CX design. And it's a choice that needs to mirror brand choices, but also cost as companies think about who they want to be in the digital space from an automation point of view and how they want to present themselves to their customers. Hello and welcome to Conversations with Zendesk. I'm your host, Nicole Sanders. Today, I will be talking with a couple of Zendesk's own experts on how AI will impact customer experience as this new technology continues to mature and become ever more present. My guests today are Adrienne McDermott, Chief Technology Officer, and Teresa Hahn, Senior Director of Technology and CX. Adrienne and Teresa will also be speaking next week at Zendesk Relate in Las Vegas. So if you like what you hear, you should join us and come visit their session. We'll dive into the conversation in a moment, but I'd also like to remind you that we love hearing from our listeners. If you have feedback for us, a question that you'd like to hear answered or a topic that you'd like to hear us discuss, reach out anytime at
[email protected]. Please be sure that you're following Zendesk on LinkedIn, which is where we share the latest episodes, NIGUS, and more. Okay, onto the show. Adrienne and Teresa have welcomed the conversations with Zendesk. How are you today? We're doing great. Thank you for having us. Thank you very much. Good to see you, Nicole. Good to see you. It's so fun to have both of you and have you together in one room. I think we're going to have a great conversation today. So you are both wonderful, strategic technical experts. Would you like to share anything about your roles or your work before we dive in that would help set context for everybody? Sure. I'm the CT of Zendesk. I've been working as Zendesk since 2010 building product, thinking about products, talking to customers, immersed in CX, and Teresa here is the boss of me. I've worked with Adrienne for about five years now. We actually started working together when he was over on the product and engineering side, and I was in product marketing. So we'd worked together on a lot of product strategy, and you product keynotes, and then joined to work on thought leadership with Adrienne a few years ago. And this is actually fun, Nicole, because we're calling you from the San Francisco office. And Adrienne, I work out of this exact room every week. We have a big brainstorming collaboration session for a couple hours. We think a lot about what's happening in technology. It's happening in CX and just debate it back and forth. And then we see what new snacks are in the office. And we go eat them. That's right. I love it. What's the best snack that's there right now? Today, everyone's really excited, because we have the chocolate pandas back. They've been restocked. It's a big deal in a while. Chocolate pandas is a big deal. It's fun. Our podcast listeners will get to be a fly on the wall in that room today. It's lunch. Of course, we've been talking about AI a lot this year. We will continue to talk about AI a lot. It is not only the hot new thing. It's a really big inflection point that is going to substantially change the way that people are interacting, engaging, getting help, conducting business. To put this moment in perspective, can you walk us through some of the past inflection points that have made a big impact on CX so that we can get a sense of what we are approaching here? Definitely. It is interesting, though, to think about what is the impact of an inflection point in general? What is it technology does when it's first introduced, right? Actually, there's a quote from a technologist Tim O'Reilly that I love. He says what new technology does is create new opportunities to do a job that customers have one done. I love that quote because it's customer-centric. It also thinks about the disruption in terms of how we serve and what we can do for customers. There's also a lot of fear that comes with these technological innovations or revolutions. The best example there, I think, is the invention of the steam engine. My gritty Northern England ancestors were all working in cotton mills and wool mills and factories, and that was that gameful employment weaving by hand. Then along came steam engine, which allowed mechanization and industrial scale. There was a fear, I think, that all of the jobs would be replaced. Economists call this concept the lump of labor fallacy because, as we perhaps know from studying history, what actually happened was the industrial revolution. There wasn't just a single, small lump of labor to be done in the economy. There was actually a tremendous amount of work and creativity that could be created. Technology comes along with opportunities. It scares people. It brings risk, but it also creates a different way of being in terms of a different way of the businesses operate and different opportunities for basically serving customers. We've been looking at a few of these different inflection points historically, and we do think it's helpful to do that and look back at these big periods where technology did really shift and change the X because we can look at some of the things that really transformed in that period and then think about how that might happen again with the next inflection point. What does that look like being on this next wave of AI and how might these different pieces also shift? The first was the invention of the telephone, then came the invention of the internet, and then the later stage of the internet, what we call Web 2.0. The telephone was a really incredible invention that really started so many of the paradigms and customer support that we see today. These are paradigms that we still see in practice of things like how you staff a team, how you staff for peak periods, how you think about the procedure and the script, the right method and approach to handle a customer. We also started to see so much of support really become scaled, and this was the first chance for that to really be possible, taking advantage of time zones, taking advantage of the fact that people could be in totally different locations but still be able to connect. What it wasn't really was democratized. It was a very high barrier to entry. You got to pull in multiple copper circuits, install a piece of hardware, have a bunch of people in a room, on the phone, blah, blah, blah. So then came the internet, and the internet was a disruption. It was a disruption because suddenly everyone could basically get someone to help them through a web page or go to a site where web pages could be built. Everyone could have a presence. And then pretty quickly following on from that, I think what was invented was the concept of self-service. Google taught a generation of humans along with Yahoo, I suppose, and the other search engines at the time, how to type something into a box and get back 10 blue links and find the answer. We still do that today, except probably we get 10 blue links and 13 ads, but version one of the internet basically created self-service and enabled every business to have a presence and start providing support 24/7. After that period, after the web 1.0 phase, so that was like the read only era, we moved to this web 2.0 phase, which is the read right era. And this was a really impactful inflection point for CX because we started to see things become so much more dynamic, so much more efficient. So support really became immensely better. We saw a number of new platforms emerge like Twitter, Facebook, all these new places and channels to get support. This feature I launched at Zendesk, we just support. And then you blasted a company through Twitter, I'm sure. Something happened. I love it. We have built it to just message right in how those ongoing conversations and that conversational nature of support, that really emerged during this time. And then also all of that happening, of course, on mobile, so not just a thing you had to do on your desktop, but easily on the go. So is there sort of a typical, basic process that we go through? Like as we hit each of those inflection points, are there common things that we see that happen with each of them? And Adrian touched on this a little bit in what he was saying right up front with saying, there's some fear, there's some excitement, there's some of these different things. What are the commonalities around each inflection point that we see? Well, we see when a technology is first introduced is we see an increase in volume. Why do we see an increase in volume? If you don't have to go to the store, if you can actually call someone up, it becomes easier. You can just go to the website and look, it becomes easier. Things that are easy, you use more of. And so we see this increase in quality. We also see people putting a ton of effort on both sides of the conversation into that new channel, that new technology. And so generally, this will be accompanied by an increase in quality and satisfaction around the love of the new. We all love new shiny things. And this is great. And temporarily then there's a reduction in cost going from staffing a room filled with humans who are synchronously having conversations to do support versus self service on the web, huge cost improvements. So that's the early honeymoon period of a technology shift. Generally, we see as these technologies mature in CX, the drive for optimization becomes industry wide. And as the driver optimization becomes industry wide, quality standards drop. You get used to road answers because things are becoming industrialized. Things are becoming standardized. So you're getting the same answer over and over again, you feel like the customer service agent isn't empowered anymore because you don't wing it at and making it up like you were in the early days of Web 2.0. And so this pattern repeats in each of these phases, right? Volume increase, quality increase cost reduction turns into flat volume basically or slightly increasing volume reduction in quality, massive reduction in cost. So within those dimensions and some of the phases that you described, where are we at today with AI? Because like I said, it feels like it's both a very new technology and also something we've been talking about a lot for over a year and now. What are we seeing? Where are we at with the inflection of AI? So we're actually feeling like we're in a very important point for a new technology like AI to be introduced. When we think back to those dimensions, you do see these improvements whenever a new technology is introduced and it's really interesting right now if you actually look at those three
we feel like we're back in one of those periods where those are all actually starting to trend negatively. And so this introduction of AI is this very welcome reprieve that's going to help reset these again. So looking at volume, for example, if we look at actually our Zendesk customers that have been with us for the last 10 years, so since 2014 to now, you can look at the amount of volume that they've been getting during that period. And we can actually see that it's 3x what they were getting 10 years ago. So this huge surge in volume that businesses are having to deal with. On the quality side of things we saw with the American customer satisfaction index is very significant steady trend of increasing satisfaction with the introduction of the internet. In the last few years, we're starting to see that quality is declining. And it's not clear exactly what is causing this, but you can see a steady decline in these last few years. And our speculation is it goes back to a little bit of what Adrian was saying that there is just that natural human tendency to get used to things at a certain point, a thing that wants really pleased you mean you're very satisfied by something then just becomes expectation. So that's what it feels like we're out again today. I've really not reached that point where the chocolate pandas snacks. So I just not possible. I'm sure that that concept doesn't apply. We'll find let's experiment. We'll find it. Eat enough and try. Anyway, when thinking about the cost to mention, I think this is where real tension is at right now in my mind because during the pandemic, right. Every company had to become a digital business within Zendesk data. We saw the huge increase in volumes for so many of our customers. And that affects hasn't dimmed too much during that time period. Those businesses were growing very fast. Right, they were going through many transitions at the same time as we've stabilized now at the same level of support cost. We've needed to keep that constant higher level of humans, right. Rough numbers, but we estimate that spend on the humans of customer service, the humans of CX is somewhere between 500 billion and a trillion dollars. I think the spend on software is probably in the 50 to 100 billion dollar range based on the market side. And so there is potential here to help alleviate some of that human cost pressure for businesses. But I think it's going to come at a rebalancing where obviously we're spending a bit more on software and technology to take some of the load of those inquiries. Where is the industry at in general with implementing AI? Are we seeing a lot of companies using it? Is it starting to mature? What's that look like? I think we're very early days, right. Most people have had experiences of using chat GPT and generating amusing, useful or amazing results. We're not really harnessing that power yet in business, right. We're still kind of hardy tricks, not platform stage of a technology where it's look at what it can do. Now look at what it's doing for me every single day and every single conversation. I think we have to think about the skill set of a foundational or frontier large language model. And how we're using that skill set. My model is to think about three core skills of an LLM. The first is they have this ability to create. We're all familiar with it, right. Write me an answer to this customer service response as I IQ in finish. They would easily be able to do that. Not totally useful, but then they can create. And I think they create reasonably well. The second skill is the ability to leverage their amazing world knowledge. And this is extremely important, right. Because to generate and give a good answer, you kind of have to understand certain things about the world and concepts. Now, large language model is a friend that is standing next to you that has read the entire internet. And it's a perfect recall in any piece of it, which is extraordinary. And I think we're just beginning to harness that power a little bit. The third skill that we're barely using is the ability to reason and plan and make decisions where they can actually understand what is needed. And make a decision and set a plan and make it happen. The more we use those skills, the closer we'll be to having autonomous AI agents that can work for us in customer support. And of course, the available 24/7 generally be consistent and give good answers and hopefully start giving good experiences low cost high quality, you know, infinite volume. There's a quote that we've been saying a bunch related to this, but I think Adrian's getting a little sick of because we've been saying it so much. But this is known as a Mars law. So Mars law is one that probably many have heard, but it is we think very applicable here. And this is a great example of a Mars law, which is we tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run. I think part of what's happening right is we see this amazing technology that's been developed recently with Jon and the AI. And immediately everybody gets so excited has one pickup with some bad hallucination and then somehow feels like this thing is letting them down and not going to be as impactful. But that's the side of overestimating in the short term and not thinking about with just a little bit more time how far this development can go and how significant these improvements can be. So that's underestimating the long term that we're really seeing now. But we look at a typical support experience today. I think we see that very few interactions at the moment can be fully automated. Right. Maybe about 20% were 80% is still going to require human intervention. And we've been thinking about what are the reasons for that human intervention. But as long as we require 80% human intervention, we're going to be limited by human capacity and CX is only going to scale with humans. The purpose of AI is to sort of alleviate some of those human scaling concerns. So going back to Amara's Law, your favorite, if we think about how quickly this might change, we're actually talking about maybe just in five years, six years by 2030, we're thinking that equation actually flips. So we start to see that 80% of our actions are actually solved by AI completely without a human being involved in that actual interaction with the customer. Humans might still be involved, of course, in supervising, helping to train that AI, get that to a good spot, but they're not involved actively like they are today. And then there's 20% of issues that fine are still being escalated to humans. Maybe they're very private matters or risky situations that still go on to humans are complex enough. But AI is still involved in those interactions too. It's just more of a co-pilot. Can I double-clific into that real quick? You said that right now what we're seeing, you're looking at a lot of the reasons that humans need to intervene. What are some of those reasons and what are some of the things that need to shift with the AI to be able to start to flip the mass on that equation? The easiest way to think about this, I think, is to classify the way humans resolve issues right now. There's this class, the 20% that you can resolve, right? Those kind of mirror to traditional what a customer service is called one touch, one station, one shot tickets where you go get the knowledge you impart the knowledge to the customer. The more complicated class is multi-touch where I can't resolve your question. You want to return the red trousers until I figure out what it, you know, you tell me you want to do a return after you figure out what your orders are, is it within bounds? I have to ask you multiple questions. We have to have a back and forth. That requires that skill of planning, right? That skill of making a plan and understanding what script is being followed here and following that script with a user. And LLM needs to learn to do that. So that's when you enable that workflow, you enable multi-touch. And we have integrations because the data doesn't just live in isolation. It lives in other systems. And so AI needs to be able to go look around and find the answer somewhere else. And then finally, I think there's the last mile, which is a combination of all of these areas, right? And that's really where I think the work of CX design and journey design is done. When you figure out what it is these humans are doing or what the next most important thing to automate is, when you go attack that problem one by one. When we move up that scale then, so when we think more about that multi-touch, this is where generative AI is unlocking so much potential. So we're already seen that even with early generative AI, but obviously so much more is coming. And part of that is because one when someone reaches out, the ability to understand language so much better is really helping us to truly understand someone's intent. And what we really need to do to help solve their problems. So that understanding of the problem is so much better. There's another piece that I'm which Adrian mentioned going back to that ability to have the world's knowledge just right there. At your fingertips, there's a way to bring together so many more sources of knowledge to provide an answer to a customer with generative AI versus previously being more forced to use the static forms that existed. And if an article existed, that was a close fit. That was the one that got sent versus bringing together pieces across different sources of information to then give the right answer. And then the third part where generative AI really makes a difference with some of these multi-touch issues is there is a bit of that back and forth. There is a bit of that conversation that has to happen that people are used to having with a human and sometimes that does involve clarifying a bit of the question asking for a little bit more detail. And so generative AI is helping to make that possible to solve a little bit more of these complex questions that require just a little bit more back and forth. And that back and forth is probably going to require AI to look things up in other systems, right? It may be that AI automation and conversational automation is a great forcing function to simplify systems and put these things together which can be extremely powerful. And then you can apply that to your customer service channels. Right now it's really interesting we were looking at this recently at Zendesk and looking at Zendesk channel usage across our customers. And of course everybody is eager for messaging doing things more conversation.
There's absolutely still trend there, but the trend is actually happening a lot more slowly than we all predicted. So there is a shift, but people are still using some of these older channels quite heavily, like phone, like email. So we're finding that what's really interesting is today we're really only seeing AI applied and some of those more conversational channels. You typically think of it more that messaging sort of interaction. But if you start to apply AI across all these different channels, then that starts to really open up more possibilities. And maybe previously some of these channels like phone, like email, were very high touch. And there was a desire for businesses to also get their customers off these channels with AI coming about in those different domains. Maybe that becomes less of an issue. So there's less pressure to actually have to move people to a certain type of new channel and instead be able to use all of them because AI can handle those various channels very effectively. There's a lot more AI innovation to come. And it's coming very rapidly. If you listen to the goals of the frontier model companies, anthropocop, an AI, Mr. Alex, et cetera, one of the things they're reaching towards in the very near term, I think, is what they describe as "agentic behavior in their models." So AI that has agency to plan and make decisions, create answers. And I think one of the core use cases that these companies are thinking about is how to make that AI-based digital autonomous customer service agent that we can all deploy. And so the less mile that Theresa is talking about of deployment of AI to build automation, I think the technology will improve to bring about so much more of this. And that is certainly the goal of technologists in Sandesk, but also of technologists at the frontier AI companies who are building GPT-12, or maybe just GPT-5 right now, and clawed for as they work on the next generations of their models. You could automate your business up to 100%, but it ultimately comes down to what is right for each business. And so thinking about those unique aspects of your business that should require humans to still be involved at that stage and figuring out what that balance looks like versus just going full automation. I mean, we often refer to great service as personal service. That requires a person sometimes, right, which is okay. So for some businesses, the number may well be 100%. Low cost, high volume, all about transaction. For some businesses, it may be 10%. I may just be triage that happens with the AI and then a human gets involved because they want that concierge level personal interaction. It's a choice in CX design. That is a really interesting point. I know that we've talked on several episodes about how do you maintain that balance of human touch of being highly empathetic with this new technology? And it's been really interesting to talk to you a lot of different businesses and they do have varying needs. And so to your point, Adrienne, I love that. It's the CX choice, how much you automate. And you know, how much do you use AI to interact with your customers versus as that sidekick, as that partner helping you out? So we've given a lot of context to where we are in terms of AI development, the inflection point. I know this has gotten me re-energized and excited yet to get out about AI. But I can see how could be overwhelming for business leaders that are trying to prepare for thinking about this and implementing it. What are the things that business leaders need to be thinking about in order to reap the value that AI potentially can bring them? There's definitely that feeling of overwhelm. There's a lot of pressure on CX leaders to start using AI and figuring out the impact very quickly. And sometimes we've seen customers are seeing it a little bit black and white in terms of either it's all in on AI and it's customer facing. It's immediately engaging with my customers or I'm not using it. And we really encourage businesses to not quite think about it in that way that there's more of a slow and steady progression they can take in adding AI to their business so they can start small finding an area in particularly internal that is working with their Asian teams as like a co-pilot alongside their agents. So giving recommendations, AI is then sharing maybe a confidence level or some indicator of how strong a recommendation is with humans in the loop to help supervise give guidance. And then as soon as they're starting to see that AI is really effective in an area and maybe getting as good a satisfaction if not better than certain humans for certain types of increase they can flip that area to then fall automation move on to the next area. Yeah, I think it's becoming a patent in AI deployment to think about what is the humans in the loop solution? Like how can I turbocharge human ability but also leverage the power of CX, right? So that's where we get the co-pilot patterns that we have or in Zendesk we have some agent assist tools to help with content creation will be on to I think those can be extremely powerful and then even from a bot automation point of view back to our return example that you want to do return. I probably have a very well defined flow for returns and there's perhaps some understanding what cognition work that AI can do in that flow to help communicate with the user or ask the right questions but we don't want is creativity that is actually not moment be creative that's a moment to follow a flow chart and rules and standard operating procedure and process and so thinking about what's the humans in the middle approach what's the hybrid approach that follows rules and deploying in that way is extremely important that's the way we think about it is undesque which in some ways is the most important meta point about using artificial intelligence. So we've talked about evolution of the industry we've been to the past we've talked about how people should be preparing for the future to ground this in today. Do you have any examples that you'd like to share of companies that have done something particularly interesting or innovative with AI so far? There's a bunch of really innovative sort of song music apps that just generate sound. Refugian is one and I like it because not to be super nerdy they actually take a spectrograph of the music recording and then they dicks using kind of stable diffusion methods what the next bar will look like based on that graph not based on the audio at all they don't analyze audio at all you can do really fun things with that I think that's pretty cool. That is pretty yeah. I actually made a song just for you Adrian right before we got started as Adrian was sharing that he was thinking this would be a great example to mention here I figured why not try to make a song just for Adrian so I love it a little prompt for us and let's see if I can find exactly what that prompt. I'm salty was the prop Theresa it was slightly insulting okay that's me but it was honest it was honest so the prompt said please make a song for my boss Adrian who is a big AI fan but a bit of a grumpy old man. And this is what we found. Adrian you're a grumpy old man always found in making demands. It's a bomb man this is it. I took a lyrics and I put them in refuge in my favorite app and I asked it to make a 90 synth version and then so this is what we have here. That has an asubasified. I like it. I like it. Well this has been a wonderful conversation I'm sure we could keep going but we should close it out here so thank you both and I would love to have it back again in the future to talk about more things and I look forward to seeing you both at Relate next week that'll be really fun. Thank you so much. Great listeners as a reminder you can always join us on LinkedIn to find out the latest and as news get updates on the podcast and we hope that we'll see you at Relate next week. If we don't we'll be back in your feed in two weeks with another episode. Until then I'm Nicole Saunders for Send Us, the Intelligent Heart of Customer Experience.
PodcastTranscript AI