How AI will help businesses understand customers better with Zendesk’s Cristina Fonseca

But I think things become more interesting for agents that are new that need to learn. Maybe they are being trained remotely or the context of the company changes frequently. And there's new information that needs to be made available to everyone really quick. So these assistants are very powerful in those use cases. Hello and welcome to Conversations with ZenDask. Each week we speak to customer experience innovators and experts to discuss belay to trends, interesting ideas and what's new. I'm your host Nicole Saunders. Today I'm speaking with Christina Fonseca, VP of product at ZenDask, who heads up our AI division. We're going to discuss the ways that AI may upend customer support, digging into everything from what it means to leverage AI to offer more streamlined customer experience to what kinds of standards need to be put in place in our industry. And how ZenDask's new AI tools will help both customers and support agents. Christina is a great expert to talk to you about all of this. She started her professional careers in entrepreneur having co founded talk desk in 2011, which was the first cloud native call center solution and raised 500 million from top global investors. In 2018, Christina co founded cleverly AI and artificial intelligence automation startup that got acquired by ZenDask in 2021. She holds a degree in computer engineering and telecommunications from instituto superior technical and Lisbon and was appointed as a young global leader by the World Economic Forum in 2021. In addition to her leadership at ZenDask, Christina is also a startup investor focusing on technology and ocean related companies. I am so excited to have her with us today. So enjoy our conversation. Redis, hit your customer experiences to the next level, build lasting relationships, the zenest complete customer service solution so that you can exceed every customer's expectations. Sign up for a free trial at zendesk.com. Welcome to the podcast, Christina. I am so glad to have you here today, talking all about AI and we're talking about how AI is going to potentially upend and impact customer service. So it's exciting to have you here as Zendesk expert on all things AI. Thank you so much, Nicole. The interesting beat is that maybe if one year ago, it was about maybe the potential, I think like to buy it's happening. So it's way more interesting to talk about everything AI. Super glad to be here with you. Absolutely. The future is here, right? It's happening exactly. I thought a great place to start would be to hear just a little bit about your background and how you got into the AI space in the first place. So it's actually a funny story. So what happened was I am a software engineer by training. Then I became an entrepreneur. So I could find another company in the CX space. And then a company that was acquired by zendesk. So I joined zendesk through an acquisition a little bit less than two years ago. And maybe 2015 16 AI became a little bit of a trend in the start of world and everyone was pretending they had AI, but it was all fireworks. So people would just talk about AI as if it was this magic thing that would change every industry. And I struggled to connect the pieces. I thought I am a software engineer, but I don't really understand how this is happening and what it takes for real. So I did a degree on AI and then from that to applying it to customer service, which was the industry that I knew was like a very short path. So I started exploring building some models and then I created a company in the space that was eventually part by zendesk. That was how I got into AI. Very cool. I didn't realize that you can get degrees on AI already. I mean, I suppose it makes sense in the computer science world. It was actually a nano degree. Like it was almost the year of doing an upgrade on AI. And I think it started because companies were trying to find talents, but there was not enough people. So these degrees were a fast way for you to study AI, which was super cool. I imagine a lot of people are going to be getting into that space in the very near future. It's such a hot topic these days. It is for sure. So zendesk does our annual CX trends report. And this year we did a mini report, if you will, all about AI and consumer expectations around AI, just to really try to understand that space. And one of the things that we learned out of that research is how much consumers are really expecting AI to upend the customer service space. And if anything, what we're hearing is that customers want companies to move faster and faster and get it implemented. At the same time, of course, there's a lot of conversation going on around, how do we do this safely? How do we think about data privacy and all of these things within it? I'd love to hear your thoughts around that challenge for businesses of how do we move fast, but not too fast. How do we try to meet those consumer expectations while still being thoughtful about how we're implementing it? It's a very good question. And I think when I read that, I think what customers really want is get me whatever helps me get my issues resolved quickly with a good customer experience. So I think there's also some reports and some insights on customers expect AI to have empathy and to really understand the customer. The customer just wants to be understood. The customer is totally okay. We're talking to a robot very transparently, as long as the robot helps solve the problem. Now, I think if we look back 10 years, the first AI implementations were very, oh, let's go full automation because this is the potential that exists in the technology. And now you talk to my robots. And I'll make it very hard for you to talk to a human, which might have some benefits for the companies. Right, because everyone understands automation. You can say, of course, you can increase productivity that way, but from a customer experience perspective, that's not the right strategy. So what we've been advising customers to do and one of the pillars of our AI strategy is to first of all, help them understand what can be automated and should be automated and distinguish between that and what needs to be escalated right away to a human. And if we can have AI that can understand what the customers want really well, then I can make that differentiation up front and not have my customers engaging terrible customer experience where I'm trying to push a box to talk to them, but I know the bot is never going to be able to result that particular issue, either because it's too complex. There's no standard way or flow that leads to a resolution or it just requires an extra degree of empathy and human connection. So to me, the first job of AI is to understand the customers well. So companies can differentiate between what can be automated and should be automated and whatever like should be escalated right away. And I think that's what customers want at the end and what those trends reveal. Got it. So I know that we recently announced a lot of things around Zendesk's AI offerings and one of the things that you really highlighted when we made those announcements is how what we're offering is really focused around customer intent and customer sentiment. That makes a lot of sense based on what you were just saying that it's understanding when the robot is going to be the right resolution path. And when it's oh, I just need to route this to somebody like this one's going to have to be handled by human or get escalated or something like that. Is there anything else that you can share about that intent and sentiment focused part of what we're releasing and how that's going to work? The main point about Zendesk AI that has these foundational elements all let me understand that customer first is exactly that one in the past 10 years. AI systems have been while let me give you a platform that you can use to train your models and then build whatever you want on top. And this is very painful for customers to implement. It takes many months of consultancy projects and consumes a lot of time and effort. And if we look at statistics most AI projects would fail because of that. I mean like it is we would ask teams that were not experts in AI to train AI systems. So at Zendesk we have a completely different approach. What we have been doing has been how do we make AI work on day one with the click of a button. So our customers can really focus in driving value and collecting like the benefits that we have been promised over the last decade. Without engaging in these super complex implementations or projects to make AI work. And when we look at let's call it the bottleneck what's needed for bots to perform well for agency's applications to have good performance and to really help agents solve customers issues faster. What are the bottlenecks in these systems from a technical perspective. That's basically one understanding what the customer wants because if we can understand what the customer wants. Within the Zendesk ecosystem where we have a suite of applications that help you deflag we have bots we have reach integrations in deep integrations with other systems. If up front I can understand very well what customers are asking with a high level of granularity, then I can just program my zen. desk to be super efficient. So we focused on that, identifying customer intent and customer sentiments. And that is really like a game changer for our customers. I still remember my startup base when we would turn these on for our customers, they would literally start dreaming. You're writing me like, now I know when customers are writing me about, oh, I just made a mistake in my order address. Can you please update that? And I can prioritize that right away, or if someone is stuck in the shopping cart and you're just telling me I cannot pay, I can proactively prioritize that. And that means extra revenue, that means a way better experience. And that means happy customers at the end. So our approach has been no, we are experts in machine learning, we're going to do the heavy lifting, we're going to train all of these models for you to understand what your customers want. With a level of granularity that allows you to personalize experiences. And based on that, then we can have bots, we can have agency's applications, and we can really drive results and impact to our customers with AI. Now it's very obvious, after the chat you PT lounge, I honestly believe the best thing chat you PT to the world was set the expectation that AI should work off the shelf. Now everyone is like, yeah, yeah, but I want chat you PT because it just works. So these notion of having AI that just works, just set the expectations that I believe our customers should have. I want to see some that I turn on, it works, it drives results. And that's the exact philosophy of Zenf SKI and what do we launched? We were talking a little bit about the ways that AI can personalize experiences and what that means. And maybe you can share some examples of an anonymized experiences of this and a personalized experiences this, right? Rather than coming and having to give them all my information, the assumption is the robot can look up my account and my history. So in terms of personalization, it can mean lots of different things, but I think in CX, what customers really want is to make sure, whoever, like it can be a human or a robot, is at the other side, understands context. I think like the worst thing that can happen is that I have to interact multiple times with a company and I have to explain over and over where I am in the buying process or who I am. Like that's just very first writing. It's so frustrating. Yes. Especially when I know the company has like the data like they have the entire history. Right. I've collected them before. So that's very frustrating. And I think the AI systems and the human interaction are going to benefit from the same type of integration in terms of data. So if I'm supporting a customer, I should have information about the backhand, the order, the CRN data, and we should look to integrate all of this. So I can really personalize interaction. I can prioritize things that need prioritization. Algorithms are the same. If I really want my bots to understand a customer and put the customer on the right path and drive a quick resolution, I need to maybe integrate with my backhand system to understand like who that person is, what's the plan, is it a VIP, what was the last order. So getting these extra insights and integrating them into the AI systems is crucial to make sure like the experiences stop not. But it's not that different as in the manual side of things. Right. And I think that's also very interesting lesson. When we talk AI, as you develop AI products, which could be a completely different conversation we were having, it's very important to ask, okay, how would I do this task manually? What would you say to perform a task I'm trying to automate manually? And usually if the answer is, okay, I need to get information that leaves in other systems to do a good job as a human. If I'm trying to train a model or a machine learning algorithm to do the same, I need to do the same. I need that model to drink from the ecosystem of data. That's not maybe just that single interaction. I love what you just said about how you have to think about how you would do this in the current world and then how you would set that up with AI. I was just listening to an NPR report yesterday that was talking about a study done where a big customer service organization had implemented chat GPT to help their agents with support. And what they found was that it really helped the newer agents who had less knowledge and less familiarity level up to be more efficient to offer better service, but the really experienced agents, it was kind of a nice to have. It maybe made them a little more efficient. And so I think that really grounds that point that at the end of the day, where AI can really help from the agent side of things is bringing everybody up to the level of the best manual person who can already do all of that processing maybe in their head or based on their past experience or something. I think another place where customers are hoping that AI will really help. And I'd love to hear your perspective on this is with what we call in the industry and omnichannel experience, right? That's where I go in and first I'm talking to a bot and then the bot hands me off to an agent and I'm on a Zoom call with an agent. And my expectation as the consumer is that the agent knows everything that I just put into the chat bot, right? I'm getting a good hand off that information is getting transferred, but currently that's not always the case. That's not always the experience. I think an experience that a lot of people are frustrated by is when they move between channels and it feels like they're having the same conversation over and over, kind of like you were saying in your last example. Talk to me a little bit about how AI might help smooth that experience out and fill some of those gaps in that omnichannel space. I'm not sure AI is the answer to that. I think the answer to that is goods, system integrations and good workflows and good software that can help you transition from one channel to the other and so on and so forth. And I think your question is super interesting because in this new era where we have AI, it was promised to solve a lot of different problems, people turned to AI as, hey, dear AI, can you help me like solve this problem off? No one understands me, bots don't understand me, I'm transferred to another channel like I need to start all over again. So I think AI has been these umbrella term that people used to associate with every single thing that needs to be fixed in CX. It's that old saying when you have a hammer, everything looks like a nail, right? Everything having like lots of conversations with lots of different stakeholders, especially as we approach relate and we launch all of these new capabilities. And I think going back lots of people's like moving journally sensor one have asked me how as your strategy changed in the past six months with open AI. And the right answer to that is wait, the problems we are solving have not changed in the last six months. And the reason why AI has really a lot of potential in CX is because the amount of manual work that's performed and that can be automated. It's just huge. These agents spend time and time again like writing the same thing over and over looking for the same information labeling tickets so we can have root cause analysis and insights and reports. This is wrong, this is fundamentally wrong. So the potential of AI and CX was there years ago. It was one of the first industries that the analysts like, predicted, okay, like AI is going to change this. Chachi media came out again, same predictions. But it's not because of the technology, it's because of the nature of the work. That didn't change in the last six months. Just the technology became more powerful. And suddenly we can leverage these off the shelf models that can retrain that are very good at finding information at modifying text. So suddenly maybe an agent can just write three bullet points and Chachi PT will craft a perfect reply. And that's a big productivity boost. But the challenges of AI are still there. One, implementing AI, finding the time, giving it enough time and making and engaging in implementation projects to have it properly implemented. I think one of the biggest challenges in the next couple of years is going to be building trust in AI systems. And one of the points you made before in regards to agents, if you get an assistance to agents within a CX operation, the ones that are there and our experts already will say, I can do better than this. I'm an expert, I can do better than this. And that's correct. But if you ask the same thing to new agents who need to learn on the job, don't know where information is, they will depend on that thing forever. We could all use assistance to do the things we do on a daily basis. And like I'm a strong believer that in the next couple of years, we will all adopt a number of assistance for work and personal life and the expectation. But I think things become more interesting for agents that are new that need to learn maybe they are being trained remotely or the context of the company changes frequently and there's new information that needs to be made available to everyone really quick. I understand that maybe for like the expert, they will say, well, I can do better. And then there's a very important aspect that we really take into account as we develop the systems which is make the confidence level of our predictions available and transparent to everyone. Because AI is not always right and there's a difference between, if for example, I'm trying to recommend a reply. to be sentenced to a customer to answer a particular case. Sometimes I'm highly confident that's the reply. Sometimes I don't have a good suggestion, but I'm going to try. And I'm going to give you a suggestion that I know it's maybe only 30% or 40% probability of being right. If I don't make a difference between, hey, I'm 95% confident this is it versus I'm just 20% confident and I'm trying. And I'm asking you for feedback. And maybe next time I'll do better, because this is the way algorithms work. It's very easy to point that, oh no, the algorithm is getting it like 20% of the time is it's wrong. Yeah, but maybe it's because it doesn't have a good reply. It's a new case. It's a new situation. The machine didn't have enough examples to learn how to handle that situation. So, we build products with the right mechanisms so people can leverage AI while understanding it has limitations, but building trust in the systems. That is such an interesting point. As you know, I run our user community. And if you think about how people respond to each other and they're answering questions, they give that kind of validity score all the time. They're like, well, I've never done this myself, but here's how I would try it. Or I've done this before, but you should check these things because sometimes this is what breaks or something like that. And so I think that is a really good point because of course, one of the concerns with things like chat GPT is that people will be like, well, the computer said it. So it must be right. And without something saying, we're 80% confidence in this or something like that, people are going to assume that that is the right answer. Most of us lost a little bit of that tendency to go validate information. And so I love that idea. What do you think are some other things that companies can think about or do to help build trust in their AI implementations and to help build some transparency into it? I would say the Confederation Level Trick is actually one that I believe transparency builds trust. That's one of our principles as we design AIML systems. That's one of our principles in the community as well. Various lines. It seems we work at the same company. Almost as though. So that's one because you see so new, there's no playbook to implement AI. There's no playbook. Some companies are okay to risk a little bit more that also depends on the industry you are in. Some companies really have, especially more established brands a little bit more of a resistance to automates. Thanks. Oh, like automation. No, that's very dangerous. I'm afraid the system will do something that I don't think is right. One, humans are not perfect. So I think it's totally fine. I think if we assume and we make it transparent that it's a robot answering customers, that's my preferred way of deploying AI. Hey, I'm gonna tell customers this is a robot. The robot is not gonna be perfect at all times. But if you just tell us like the robot did a crappy job and then I prioritize that. So I have a perfect combination of automation will do the first try when that doesn't work. There will be a human ready to assist. I think this is the best combination. But these are just like small tricks we can use to still leverage automation. Select the benefits of AI. Risk a little bit. I think these days both companies and then users are okay with a little bit of experimentation. I would be very careful in trying to automate everything or having generative AI answer customers. I still don't think technology is there. Still some risks that need to be taken into account even in terms of cybersecurity and so on and so forth. So the potential is there, but the technology is still not mature enough to be deployed broadly at scale. And there's such big interests in everything generative AI and myus. Okay, but I think companies should work with experts and people that have expertise in implementing these systems at scale because they might not be seeing their all the risks in implementing these kinds of technologies. So partnering is always like a very good idea and we are making lots of work trying to engage with every single person and spending a lot of time in effort making sure we provide the right tools for our customers to implement AI. That makes a lot of sense. And to your point, I think trying to have awareness about the potential risks and working with people who are experts in this space is going to be really important. Obviously there's been a lot of talk lately around AI in terms of what kinds of regulation and management should we have of that? We heard a lot about Sam Altman going to Capitol Hill last week and saying, "Please regulate us," which there's a whole world of commentary about what that was all about that we're not going to dig into here. If somebody came to you and said, we're going to put out a couple of industry-wide standards for implementing AI, what kinds of things would you like to see different companies and organizations that are developing in this space agreeing on or holding themselves to? So first of all, you touched on a point that I believe is essentially the right way, which is doing industry-by-industry. I think there's lots of, especially in Europe, I'm sitting here in Europe. So we are the masters of bureaucracy and regulation. And I think there has been a lot of debate around, "Oh, we need to regulate AI." Of course, we can't use a technology for certain things to do harm and to have negative implications in society. We all agree AI can be used in that regard. But I think you can just regulate AI broadly. I think industry-by-industry, it's possible to have certain guidelines that everyone in the industry follows in banking. That's very different from customer service. That's very different from retail. So I feel industry-by-industry, we will eventually have energy regulators, but more like associations. And I think like we are building some guidelines of how to properly implement AI within a particular industry. What are the safeguards we need to put in place, and so on and so forth. I have a hard time seeing broad regulation be implemented, because I think that's a very, very tricky subject. As any other piece of technology, it can be used very well, but if it's used wrong, it can really cause harm and damage. And based on what I've seen in CX, I mean, we really want to do this right. Every time we talk with our partners, open AI included everyone is trying to figure out how to do these things properly. I think these technologies companies have the best interest in designing systems that are safe, secure on the legal side, like they meet all the regulations. People are genuinely well-intentioned. I think of course, as any other technology, there's potential to go off track. I don't know if more broad regulation is going to solve that. I don't think so. I think it will be very hard to implement that. Certainly we'll be interesting to see how different countries approach that. I'm certainly hopeful that we will start to see some standards and best practices, maybe even before anything legislative comes to light, official. But things where we're all going to hold hands and agree, we shouldn't use it this way, or when you're doing this kind of practice, you should make sure you've got these things in place. It will be fascinating to see where it goes. Of course, if there's one thing I've learned a lot of the time, it's that the rules get made after somebody does something and you go, "Oh, we shouldn't allow that to happen again." Let's go make a guideline around that kind of thing. We're about at our time to wrap up. This has been a fascinating conversation. I feel like I could chat with you all day about this. But I always like to end on a fun note. What is your wildest idea of how AI might change customer service? What's the blue sky? If you can see it, do anything. What would really be exciting for you? What came to mind is like so wild, but I think it summarizes very well. So, customer service is an industry that's fundamentally broken, I would say. So, no-goal center agent is super happy about that particular job. And no customer likes to come back to customer service. So, that's why with AI in the middle, we really have the potential to make the world a better place for both sides. So, I look forward to a future where automation can really take care of all the manual work, the mundane tasks. And you ones are left to go to the beach and do better things and engaging deep conversations that are not just like a transactional. Hey, I want to refund these mundane tasks are just nonsense to me. So, I hope AI can make the industry a better place overall. Oh, what a fantastic answer. Making the world a better place, making jobs better for agents, making customer experience better, making work more meaningful. Like, yes, there's lots of also debate around are we eliminating jobs and so on? And based on implementing AI with some of our customers, like these people end up being repurposed to have way more fulfilling jobs. And I think that's also something we should strive for. So, looking forward to the future. I really appreciate that you raised that point. And we could probably do an entire episode about the potential impact of AI on jobs. And that I think is the elephant in the room for a lot of people. They're really excited about it. But it's kind of like, AI is so cool. It might eliminate my job. Nobody wants to think about that. But I think my hope, just like yours, is that in the long run, it means that, improves the experience for everybody and makes life better. Yeah, and again, it could be a longer conversation, but we can't forget that this is probably an industry with one of the highest turnovers. So the cost of hiring people, training people, replacing people from an Asian perspective, like if you change jobs every year because your job is just very repetitive, not interesting, that's also not desirable. So again, too many inefficiencies, it's a job that can partially be performed by machines and humans can really go to more fulfilling things. Love it. Well, thank you so much, Christina. This has been such a wonderful conversation. I'm sure we will have you back again as we continue to dig into this AI world, but have a wonderful day and thank you so much for joining me. It has been a pleasure. Always happy to come back. Thanks so much for listening today. I hope you'll join me next time for a great conversation I'm going to have with Deepa Misra, senior advisor to startups for AWS. We're going to talk about the ways of a variety of startups are developing for AI, how to cut through the AI hype into what is truly exciting about this space and what practitioners should do right now to make the most of the moment. In the meantime, I hope to see you around the Zendes community. Please be sure to like and subscribe to the podcast and tell your friends. If you're interested in learning more about Zendesk AI, we recently had a fantastic event with some of our product managers who shared more about the capabilities and answered a number of users' questions. You can find a link to the recording of that in the show notes. Until next time, I'm Nicole Saunders for Zendesk, the intelligent heart of customer experience. Ready to explore how AI can help your business provide richer customer experiences? Zendesk enables you to instantly harness the power of AI to deliver scalable, world-class customer experiences. In the era of intelligence, the X, Zendesk is here to help you find your way. Sign up for our free trial at Zendesk.com.