Zendesk CEO on AI in Customer Experience:
What Works, What Doesn’t, and What’s Next

Introduction to Zendesk and AI in Customer Support

Michael Krigsman: Welcome to CXOTalk number 886. I'm Michael Krigsman, and we are cutting through the hype around artificial intelligence and customer experience with Tom Eggemeier, CEO of Zendesk.

Tom will share honest, real-world lessons from Zendesk's AI journey, what has worked, what has not, and how AI is reshaping customer expectations. Tom, welcome to CXOTalk. I am delighted to chat with you.

Tom Eggemeier: Great to be here, Michael, and I am delighted to be number 886.

Michael Krigsman: Everybody has received messages from Zendesk. We submit a support request, and we get something back from Zendesk. Briefly give us some background. Tell us about Zendesk.

Tom Eggemeier: Zendesk is an AI-first customer support software company. We are all about helping companies resolve their customers', their employees', or their fellow business issues.

The Realities and Challenges of AI Implementation

Michael Krigsman: What is going on with AI in customer experience today?

Tom Eggemeier: When we think about AI from a customer support perspective, there are huge gains for companies, employees, and consumers. It is not hype. It is actually real.

Michael Krigsman: When you say that AI is disrupting, can you elaborate on that?

Tom Eggemeier: AI is actually solving customers' problems. I am old enough to know that 20 years ago, we talked about how machine learning or predictive analytics were going to be able to solve customers' problems without human intervention. It has been a bit of a hype cycle for the past 20 years, but we are actually seeing that happen right now.

We have one product called AI Agents Advanced, where you put a bot into one of our customer environments. With three clicks of a mouse, taking some knowledge from the company, we can solve more than 30% of customers' interactions almost instantaneously. What is interesting is the customers are more satisfied, it costs less, it is more accurate, and you can do it with three clicks.

That is the promise of AI: always on, fantastic accuracy, and lower cost for companies so that they can put their finite resources into solving the root causes that are causing problems with their customers.

Michael Krigsman: There is so much discussion around AI, and you have just presented what could be called an idealized version. Is it totally realistic, or where is the gap?

Tom Eggemeier: The gap is a couple of things.

1. I see some people in the marketplace guaranteeing 65% to 85% of interactions can be resolved with AI within a matter of days, and that is hyping it a little bit. Some of our customers are getting 85% or 90% resolutions, but some customers with more complex use cases are getting 20% or 30%. Part of the hype is that it depends on the customer type and what kind of interactions they are having with their employees or with their customers to determine what kind of resolution rate you can get.
2. There is a little bit of hype in that, right now, you need to have a knowledge base or a bunch of structured data so that you can get resolution rates quickly. That is not always the case, and we have to be honest with customers and consumers about that.
3. There are still mistakes that AI makes, just like humans. We do not want to underestimate that, and we want to make sure that we get as close to 100% accuracy as possible.

What is interesting is we have done some A/B testing, and we see fewer hallucinations from AI than we see mistakes from humans. It is interesting that we use different nomenclature and we call AI agent mistakes "hallucinations" and human agent mistakes "mistakes."

Michael Krigsman: When things are set up in the right way, if the data is organized properly, then the AI does a great job, is essentially what you are saying.

Tom Eggemeier: Exactly. I was with a customer about a month ago in Australia, for instance. Working with them, we found out they had five different policies from different parts of the organization on what constituted a return as an e-commerce company. When you have data like that, of course the AI is not going to get it perfect because it is getting input from five different policies that are slightly different from each other. Having good policies, good knowledge, and good information is key.

Now, we can help, and other players in the market can help create that knowledge. For example, one of the things we do now is we see a bunch of the customer interactions and we see you have a gap right here. In the past, we would have said, "You have a gap. You should go solve it." Now, we say, "You have a gap. Here is a recommendation." The AI writes a recommended article or a recommended policy for you. You can accept, decline, or edit it. The world is flipping on itself right now. But as you said, data that is consistent, publishable, and readable is important in this AI journey.

Zendesk's Transformation and the Future of AI in Customer Support

Michael Krigsman: You mentioned this phrase "AI journey." I was interested. At what point did you recognize that AI in customer support was so important that you had to transform the company around it? Is that too strong a term?

Tom Eggemeier: It absolutely transformed the company. The way we look at it at Zendesk, we can be the disruptor in AI customer support, or we are going to be disrupted. We were the disruptor 17 years ago with customer support help desk, and we want to do that again.

I remember when OpenAI launched in October 2022, there was an amazing amount of buzz about it. Thinking about the implications of it in the summer and fall of 2022, I was over in our Lisbon, Portugal office. I remember this vividly. In December 2022, I was with a small group of our lead AI data scientists and engineers, about 10 of them. I had just started that month as CEO, and I asked for our roadmap. We had a bunch of great ideas, but we did not have a roadmap.

Talking to industry experts in the summer and fall of 2022, I came to the conclusion that just like the internet, just like mobile, just like premise-to-cloud, there was going to be a major disruption. We did not have a roadmap. We did not have enough people on it. We were not thinking about it.

We spent a couple of days writing on a whiteboard in Lisbon about what kind of first product we wanted to launch. I wish I would have taken pictures of it. That was the aha moment for me. I would have thought it would have been in the Bay Area, but it was actually in Lisbon, Portugal.

Michael Krigsman: You mentioned disrupting yourself. That is a hard thing to do. Most of the time, businesses get disrupted because there is a competitor or something going on out there, and they have no choice. In this instance, you saw the tidal wave coming, so you took proactive steps and said, "Hey, we need to change." It seems like that is what you were just describing.

Tom Eggemeier: I do not think we had an option, honestly, Michael. We have a stat: we think 80% of customer interactions with companies are going to be through AI within five years, and 100% are going to involve AI in some way within five years, meaning humans will use AI. We did not have a choice, but we have a lot of great Zendeskians, employees that saw this wave coming. They saw that we were the disruptor 17 years ago when Mikkel and the whole Zendesk team disrupted the help desk and customer service, and we wanted to do it again.

We were fortuitous that this was coming and that we had some of the building blocks and pieces in the foundation. But it is also an imperative that if we did not do it, we would be left behind. I had seen it in the on-prem-to-cloud contact center world. I worked for a company called Genesys that was 100% on-prem. It is now one of the leading cloud providers of contact center. They went through that metamorphosis, where the other two leaders of on-prem contact centers did not go through it. It is easier to do as a private company, but there is an imperative to do it because there are going to be winners and losers here. I feel responsible, first to our customers to give them value, second to our employees, and finally to our shareholders, that we need to disrupt ourselves.

Michael Krigsman: I want to remind everybody that there is a tweet chat taking place right now on Twitter X, and you can ask your questions using the hashtag #cxotalk. If you're watching on LinkedIn, just pop your questions into the LinkedIn chat.

Transitioning to AI: Challenges and Reactions

Michael Krigsman: We have some questions that are coming up, but I am still interested in this transition. As you were moving from the pre-AI world into AI, did you get any resistance from stakeholders, from your board, from employees, from your customers? It is a big change, and change is always hard.

Tom Eggemeier: I do not know if there was resistance. With our board, it has actually been, "How can you go bolder? How can you go bigger?" They see the transformation. We have a great board, and they do strategic agitation. Their agitation is to go faster, think bolder, think bigger on how you can give value to your customers.

There are different levels of maturity within our customer base. Some people want to be cutting edge. Over the last couple of years, some people were saying, "Is this going to be real? Is this really going to disrupt? Are we really going to be having AI agents, copilot, and generative search for knowledge bases transform customer support?" The vast majority of customers are now saying that AI is real and AI is disrupting customer support. It has gone from a question mark two and a half years ago with our customer base to a consensus. The board was already there.

Employees ask good questions, like, "How are we going to be the disruptor and not just be disrupted? How are we going to do all the core things that we need to do on our core customer service omnichannel platform while adding AI? Am I going to be left behind if I am not part of the AI revolution?" We have coffee chats and AMAs where we talk to the employee base. If anything, the employee base is excited about being the disruptor, being number one in customer service, and serving our customers to get real, tangible value with AI.

Michael Krigsman: Now would be an excellent time to subscribe to the CXOTalk newsletter. Go to cxotalk.com. We'll keep you updated on upcoming shows.

AI in Customer Support: From Reactive to Proactive

Michael Krigsman: Tell us about the changes that you drove into the product. How is customer support rooted in AI different from what came before?

Tom Eggemeier: There are a couple of things, where we are now and where we are going in the future. Where we are going right now is we think we can automate a lot of customer interactions. For the customer, it is quicker, more accurate, and they are able to solve their problem in a good way. For companies, they are spending less on it. What has happened right now in AI is a little more reactive, and it is going to be more proactive in the future.

In the past, we had about 30,000 knowledge-based customers. Think, you go into a company's website and you type in, "How do I return these pair of shoes?" A knowledge base in the past would come up with five or 10 blue links, and you would go pick one on how to return shoes. Now, we have launched something called generative search. This is a free part of our product for knowledge bases where, like a Google Gemini search, you get the answer on how to return the shoes. It is getting even better with Agentic AI, where you might be able to do that whole process, including going back to an ERP, for instance, with all the actions through an AI agent.

That is still reactive. You then go to an AI agent, maybe from the generative search, and an AI agent can act like a human agent and do all the things a human agent can do. Sometimes generative search and an AI agent are not going to solve the problem, and you go to a human agent. But that is all reactive right now.

We are flipping that right now in customer service to be proactive, where we are taking all this information from the searches you make, Michael, the interactions that you have with an AI agent and human agent, and we are saying, "These are seven root cause issues that are going on." For example, "We keep sending out black shoes when Michael wanted red shoes." We then have customer interactions. What do we need to do? We need an insight into the supply chain to see why we have a problem there, through AI. Customer support is getting flipped on its head from reactive to more proactive, from reactive to more root cause analysis, and from reactive to actually taking actions to solve root cause problems a company is having.

Michael Krigsman: I want to come back to this comment you have made a couple of times on accuracy. Can you talk about that? This is a key dimension of everything here.

Tom Eggemeier: We have done a lot of A/B testing. On the top 50 or 100 use cases, an AI agent is as accurate or more accurate than a human agent. When you get to really complex cases, maybe a customer asks three different things in a query. They say in an email or a chat or a phone interaction, "How do I return my shoes? I want a credit. I want to order a new pair of red shoes instead of my black shoes, and I want to go order a sweater." When you get three or four queries in one message, sometimes AI does not nail that. A human does better still, but it is getting better.

The other place where AI is not quite there is a very complex corner case that the data does not solve, where a human needs to get involved. On the other hand, some of the data that we have had from recent research says AI is doing just as good or better on empathy, which is surprising. I would have always thought humans are going to be more empathetic. There is an interesting research piece, and I am forgetting where I read it recently, where AI agent doctors were considered more empathetic than human doctors. We are starting to see more empathy and more caring from AI agents, which is something I never thought we would have had.

Michael Krigsman: We have come a long way from the chatbots, which were simply prescriptive and, for a lot of users, a waste of time. Things have changed. There is a very different universe now.

Tom Eggemeier: Exactly. Those chatbots that you probably have in your mind that would be frustrating for me and frustrating for you, were decision-tree based. If then, you go yes, and you go down the decision tree. A lot of times, the inquiries were more complex, and you would get stuck at part of the decision tree and then get handed over to a human. The human agent would not have the context of that chatbot.

What is different now with Agentic AI is that these AI agents are basically autonomous and can reason. That is the fundamental difference from the chatbot experience everyone had. I know you probably dealt with chatbots in the past, but if you deal with a true Agentic AI agent right now, you are going to be wowed by the experience. It is not perfect, just like human agents are not perfect, but that is the fundamental difference. You are going from a pre-baked decision tree to something that is agentic, that is reasoning.

Transparency and Use Cases in AI Decision-Making

Tom Eggemeier: What is important as part of this is that you need to be white box, not black box. If I were a company or consumer, especially a company, I would want to know how the autonomous, Agentic AI agent made decisions. What was its thought process? One of the things we do is we lay that all out for every single interaction an AI agent has with a customer, so that companies can go refine that by giving it natural language instructions to keep getting better. That is important because if you are black-boxing it as a company, you do not know why the AI agent is doing something, so it is harder to improve.

Michael Krigsman: I use large language models, multiple LLMs, every day, all the time. For the reasoning models, when they expose their logic, it is so useful.

Tom Eggemeier: I agree. I will give you an example from my personal life. My wife's grandfather was born in the US but grew up in Ireland. My wife wanted to know if she could get Irish citizenship. I took our Ancestry.com data, gave the password to a large language model, connected it to an email account that I created, and said, "Can you research and grab records to support the claim that my wife can get Irish citizenship?" It was amazing what it was able to do. It requested parochial grade school records from an Irish parish school. It got me all this information and then wrote a legal memo on why it thought my wife was able to get Irish citizenship.

The only problem, Michael, is it got the legal conclusion wrong. It said the legal conclusion was yes. It is probably a 50/50 corner case. But it exposed some of the logic and some of the reasoning, and that is how I caught that it might have been wrong on its conclusions. First of all, it is a cool business case, right? You can take your tree from Ancestry.com, create an email account, and get a legal memo with supporting documents by just giving some prompts. On the other hand, that is why it is so important to give that white box treatment to understand the logic behind it, because there are flaws and mistakes, just like human beings would make.

Michael Krigsman: When the AI makes a mistake and a customer service agent is actually working with a customer, do you have ways of trying to trap that? How is that managed? The mistakes can be insidious. The LLMs speak with the full confidence of someone who knows they are absolutely right.

Tom Eggemeier: Totally, all the time. One of the things we think we differentiate on and we think is important for any player in customer service is we have something called quality assurance. Quality assurance uses a different algorithm than a large language model, and we score every single AI agent or human agent interaction for accuracy, for tone, for resolution. Through that, we find where humans are making mistakes or AI agents are making mistakes, and we surface that.

You are totally right. One of the flaws right now is an AI agent will say with absolute certainty that my wife can get Irish citizenship, where it is a little murky, but be 100% confident. We have this strong point of view that, as a company, you should be doing quality assurance not on a sample size. Right now, most companies are analyzing about 1% of interactions. We say to analyze 100% of interactions. You can get summaries of that now based upon AI, and that is how you are going to spot issues, whether with humans or AI agents.

What is key about that is sometimes you are going to see the AI agent is handling a question better than the human agent. But sometimes, like with my 80-year-old parents—my mom will be upset she is not quite 80 yet, but my father is 80—you are going to want to have them talk to humans. They are not going to want to talk to an AI agent. You want to get that AI agent accuracy into the humans. We think quality assurance, done systematically, is so important in the AI era.

Michael Krigsman: The AI is also enabling you to have a far more granular type of ongoing QA than was possible in the past.

Tom Eggemeier: It is allowing us to be more granular and more ongoing. The great thing is it surfaces those insights where there are mistakes and proposes actions, whether that is to go build this policy, change conflicting policies, or create a new policy or piece of knowledge to solve it. It has more granular insights, it is more personalized, and with Agentic AI, it can propose and make actions to hit the root causes of customer dissatisfaction.

AI Adoption and Industry-Specific Applications

Michael Krigsman: Let's jump to some questions. The first question is from Greg Walters. He asks a couple, but I am going to go to his second one. He says, "Regarding disruption, AI eliminates the middle person. Do you see a time where Zendesk is more than a component of and a tool for customer service, becoming customer support and service, the department?"

Tom Eggemeier: Some companies, BPOs, will say, "You can outsource a lot of your customer interactions to us, and we will be in charge of the technology and the people to solve that." In the future, our platform needs to be reactive for our companies, but also proactive. It is going to change the dynamic. Right now, we are all about reactive customer issues.

Greg, this is a great question. In the future, we are going to be solving problems before they even occur with our companies. I do not know if we are ever going to be the customer service department itself. I think we are going to be a platform to help customer service resolve issues before they happen, and when they do happen, solve them. But we have debates. I tell everyone this every week: customer service is massively changing. Every week there are new AI innovations, and we have had discussions on how our platform evolves and if we are going to be taking more and more of the load from customer service departments in the future. It is a really insightful question.

Michael Krigsman: Another question from LinkedIn, and we'll jump to those questions on Twitter shortly. This is from Brandon Bird. Brandon says, "Are there specific verticals where Zendesk plans to double down with AI to gain a first-mover advantage?"

Tom Eggemeier: Over the next five years, the deeper that you can get into vertical industries, into tasks, procedures, and expertise, the better you are going to be able to solve your customers' employees' or your customers' customers' issues. Right now, we have a strong foothold in e-commerce, FinTech, manufacturing, and tech. We are going deeper and deeper and taking our learnings from those verticals and applying them to our customer service, AI-driven resolution platform.

That is going to be one of the secret sauces, where companies help their customers get more value by differentiating on a vertical or industry basis. It is already happening now. We are in the early innings, but you are going to see over the next five years a lot more industry-specific solutions, industry-specific knowledge, and industry-specific AI.

Michael Krigsman: We have a question from Firestar who says, "It feels like we're finally past the 'AI is coming' phase and into actually making it work for customers."

Tom Eggemeier: The technology works, but where we are in the next evolution is getting it adopted. It is interesting. We see customers may not want to adopt it on one end. One of the things we tell our customers to do, and this is learning from other customers, is you might want to say, "I have an AI agent available right now for you, or do you want to wait five minutes for a human agent?" A lot of people will test the AI agent because they have had the bad experience, Michael, that you had with chatbots in the past. You want to nudge them to go try the AI agent.

We also have human agents using something called Copilot. They have been using other tools, so how do you nudge them to adopt it? Once they see the power of Copilot, they will start using it more and more. Then there is just implementing with the customer to make sure that you are getting the data right, structuring the policies right, and really taking agentic AI and starting to run it through your business processes and workflows and creating new ones with AI.

There are still some adoption challenges in the marketplace. The fact that you can do a lot with three clicks, but it might take you a couple of weeks or months to get the full value stack or be on that journey, is another problem with adoption right now. We have gone from, "Does the technology work?" to "How do we get employees, customers, and autonomous AI agents all really embracing the technology?" The implementation and adoption curve is the next frontier that Zendesk and other companies are dealing with.

Michael Krigsman: It's a great point. Across different industries and different companies, as you mentioned earlier, there are varying degrees of maturity, which has a direct impact on that adoption capability.

Tom Eggemeier: Definitely. It is interesting. When you talk about a small customer, they are looking for instantaneous impact and the three-click solutions. When you talk to an enterprise customer, they are looking for embedding AI deeply in tasks and procedures, changing their workflows, inventing new workflows, and having actions that go across the ERP and supply chain to interconnect it. Different customers are asking for different things.

I look at it as a maturity curve for customers. There is a segment, whether you are SMB, mid-market, or enterprise, and there are also geographic differences. People in the Bay Area, for instance, are ready to embrace it. I am originally from Ohio and my parents are from Kentucky. When I talk to customers in Ohio and Kentucky, they might be where the Bay Area was a year ago. You have this Rubik's Cube based on geographic region, segment, and the customer's maturity on where they are ready to go. Largely, we are getting over the "Is this hype or real?" question, and people are accepting that it is real.

Michael Krigsman: I have a lot of questions of my own, but I like to prioritize the questions from the audience. You guys are really smart. We have a question on Twitter, X, and we are going to jump to LinkedIn in a moment.

Balancing AI and Human Touch in Customer Service

Michael Krigsman: A question from Lisbeth Shaw on Twitter who says, "As AI-enabled customer support functions increase, especially with AI agents and then agentic AI, how important will the human touch be, and when will human touch be the right thing?" So, how important is the human touch, and when do you need to loop in the human from the electronic agent?

Tom Eggemeier: We still think 10% or 20% of interactions are going to be human-led. They are going to be the most complex interactions. They are going to be for segments of your customer base, maybe your top customers, and there could be people that just want to talk to a human. There are going to be two or three areas where humans are still the central focus point of a resolution. Sometimes, even though I said the AI agents are sometimes giving more empathy, if a customer is really upset, a human being can often handle that well. We believe humans are still going to be in the loop.

The second thing that is exciting is I know there has been a concern that people are just going to start cutting human agents. What we have seen is something different. There has been a service deficit up until now. Most people would say when they interact with brands, whether as a business or a consumer, they might not get the best experience.

We think that AI is going to provide a service dividend, where humans will be freed up to spend more time on the root cause of some of these bigger issues. They are going to be able to manage the AI agents. You might have a human agent managing a team of AI agents to ensure accuracy, using QA and other tools. We think the service dividend is just going to increase customer satisfaction, whether that is a B2B use case or a B2C use case, because you are going to be dealing with more of the root cause and preventing things before they happen. You are going to get more personalized.

There are going to be a couple of reactive use cases. We believe people need to remember that there is always a human on the other side. In 12 to 18 months, you are probably going to be using your own bot to talk to a company's bot. At the end of the day, there is still that human on the other side of the interaction, and we need to remember that.

Michael Krigsman: This is from Preethi Narayan, and she says, "As AI reshapes customer experience, how do you see industry verticals like retail, healthcare, and financial services evolving their customer engagement models? Are there industries where AI-driven CX has become a strategic differentiator rather than just an operational tool? And how is Zendesk positioning itself to lead in those transformative shifts?" I am just going to make one request to people asking questions. Keep your questions simple because I'm reading your questions, and I need to make sure I get it.

AI as a Strategic Differentiator in Customer Experience

Michael Krigsman: So, AI-driven CX becoming a strategic differentiator rather than just an operational tool, and how you are positioning to lead in these transformative shifts.

Tom Eggemeier: It is interesting. We are seeing some customers see it as a transformational tool that can differentiate their business. I will give you an example. We have more and more customers that do not want to be customer references because they do not want to let their competitors know that they are getting 70% or 80% of their interactions solved by AI agents and are having this service dividend that they can go reinvest. Some of our best customers do not want their name out and do not want the results out. To me, that is the proof in the pudding that they are getting amazing automated resolution rates. The AI is solving the problem before it goes to a human, and they see that as a competitive advantage.

The way we are trying to do that is we focus on what we call the resolution platform, or automated resolutions, and the whole system of it. We are looking with our customers on a day-to-day, week-by-week basis at what percentage of your interactions you have resolved and automated. As we talked about earlier, we are getting deeper from a vertical perspective on that. When we get those high automation rates, you can go into that service dividend land. That is how we are positioning ourselves with our customers. We are not perfect, but we are trying to drive our customers to higher and higher automation rates.

That is why we have come out with our new pricing model called Automated Resolutions. We only get paid if we resolve a customer's problem. If it goes back to a human, we do not get paid for that interaction, even though it costs us money. That is a way for us to lead the way in customer service and put our operating model in line with our customers, because they care about driving up the automated resolution rate, and we are in the exact same boat because of how we price.

Michael Krigsman: I like that, putting skin in the game. Historically, software companies, going back to on-prem, sold the product, threw it over the wall, and if the customer used it or not, they still paid. Now, you are going to the opposite extreme where Zendesk has a stake in the customer outcome as opposed to the process of getting to that outcome.

Tom Eggemeier: We talked about transformation in culture a little before with employees. This is a big transformation for our customers, for our employees, and for our shareholders because of how we look at our operating model. In the past, it was seats. We still have some seats. Some companies made the switch to interactions. Now, the next evolution is to outcome-based pricing or really aligning with your customers.

I will give you an example. I was on an airline recently, and I noticed that I was going to miss my connection. I got on their app. I think they had a beta going on with AI agents, and I got about 90% done solving my problem of rebooking my connection. Unfortunately, the plane took off, I turned off my phone, and there was no WiFi. When I landed, the AI agent had not solved my problem. I am confident that the airline would have been charged for an interaction, even though I was more dissatisfied. I then called a human agent, and the human agent did not have any of the context from the AI agent. I felt like I was repeating myself. I had to wait eight hours at the airport because I missed all the connections as they got booked in the meantime.

That is an example where, if you are on a seat-based or interaction-based model, the software provider probably got more money from my interaction with that AI agent, even though I was more dissatisfied. That is not what we want to do as a company. We want to align our customers' outcomes with our outcomes.

Michael Krigsman: Here's an interesting one from Ryan Smith.

Evolving Metrics and Challenges in AI-Driven CX

Michael Krigsman: This is really nuts and bolts. He says, "As AI begins to resolve tickets faster and automate more interactions, how should CX leaders rethink success metrics? Do legacy KPIs, like customer satisfaction and FCR (first-call resolution), still matter, or do we need a new framework?"

Tom Eggemeier: I am going to give you an example. There are a lot of companies right now that use average handling time, which is how long a human agent stays on a phone, an email, a chat, or a web form to resolve the problem. You want lower average handling time. One of the things that we tell our customers is once you implement automation, automation takes all the easy and medium-hard interactions away, and the humans are left with the most difficult ones. We tell CX leaders, "Your average handling time is going to go way up." They might ask, "Why did this happen? I just implemented AI agents and automation, and then I gave my human agents Copilot, which should make them more efficient, but their average handling time is going up." We are starting to develop more frameworks with our customers so they are not surprised.

When you explain that, a CX leader is like, "Of course." They are getting the 10% or 20% of interactions that are really difficult and take a long time. We are encouraging more A/B testing and different frameworks. I do think still NPS or CSAT matters. I think there are going to be some new frameworks and new metrics, like how much you can reduce interactions—not for the sake of reducing interactions, but because you are solving the root cause. Over time, how quickly you reduce the interactions that keep coming up is going to be a metric in the future. Average handle time is going to be less of a metric. I still think first contact resolution is important.

If you look at the psyche of customers, they would rather wait longer to interact with an AI agent or human agent—wait on hold for five minutes and get it solved immediately—rather than being moved around two or three times between different humans and AI agents. I think first contact resolution is still key. But Ryan, your proposition that we are going to have an evolution in metrics is absolutely going to happen with CX leaders.

Michael Krigsman: It's so interesting how AI is driving change in so many different directions. The tentacles are far-reaching and the implications are very profound. Just look at what you just described with your pricing policy. That is such a dramatic shift.

Tom Eggemeier: I do not think we know all the implications of AI, Michael. People that are certain on how it is going to go over the next 15, 20, or 25 years are probably wrong.

AI's Broader Implications and the Future of Work

Tom Eggemeier: But it is profound. One of the things I do is every two weeks, I do a deep dive on what has changed in AI in the last two weeks, how it could impact our customers, and how we can provide more value to our customers. If you were in the customer service industry from 2009 to 2019 like I was at another company, the changes you would have in years are about the same pace that we now have in weeks. There are so many profound implications on the future of work. I have a 22-year-old daughter and a 20-year-old son, and they ask me, "With AI, what should I do?" That is an existential question about what skills and capabilities they should learn because the nature of work is going to massively shift over the next 10, 15, or 20 years.

Michael Krigsman: As CEO of Zendesk, what does this rapid change mean? It's extraordinary that you have to look every two weeks to see what has changed in this AI world that could have some potentially profound impact on the market and on your company.

Tom Eggemeier: It means you need to make sure you are getting a lot of signals from customers, in particular, about when we are doing well and when we are failing. Signals from the industry at large are also important. I encourage employees to be learning employees. You have to learn AI. You have to embed it into your daily workflows. You have to stay on top of it. That is a skill, and it is going to differentiate employees going forward, including CEOs: who understands AI, who is embedding it in their daily life, who is using it strategically and tactically. I give advice to people, "You have to always be learning, and you have to be always learning about AI." It is going to impact your job in some shape, way, or form over the next five years, no matter what job you are doing right now.

Michael Krigsman: This is from Arsalan Khan, who is a regular listener. He asks great questions. He says, "When will we be able to replace humans with AI? What happens to people? Will manager AIs be able to write performance reports for other AIs and humans?"

Tom Eggemeier: I am an optimist about this point. I think AI is going to enhance people's lives. We have seen with every major revolution in the world that it creates more jobs, but it creates different jobs. I am still an optimist here on what AI is going to do.

It is an insightful question about performance reviews. Right now, AI can give summaries. If you get a lot of 360 feedback, it can provide summaries and insights for performance reviews. It is probably going to happen in 10 or 15 years, maybe even sooner, that you will have performance reviews written by AI and edited by a human. I have talked to some startups that are doing that right now.

We have to understand that at the other end of that, there is a human getting a performance review. You have to be confident in the technology that it is getting all the data sources and all the signals in a good, honest, and transparent way. Again, it comes back to that white box theory, Michael, on what it got and how it came to its reasoned conclusions about your performance.

We are already there with quality assurance, where we are evaluating 100% of interactions, whether it is a human agent or an AI agent, and scoring them. It is not a formal performance review, but it provides indicators of success. There are some profound implications for human society over the next 5, 10, 15, or 20 years about how jobs are going to evolve. Your longtime listener alluded to it when he said you are going to have different jobs. We think there are going to be human managers managing teams of AI agents going forward. That is going to be one of the new jobs created from this AI revolution.

Michael Krigsman: Arsalan Khan has been waiting a long time, so I'm going to pop another question from him.

The Role of Data in AI Development

Michael Krigsman: He says on Twitter, "Can you talk about the role of data in your product and AI?"

Tom Eggemeier: One of the advantages we have as a company is we do 4.6 billion resolutions a year. There is a huge dataset that can help post-train our models, so they are not just large language models off the shelf. You post-train those models to get more accurate. It is based on that 4.6 billion interaction dataset. It is based upon, to the earlier questions, deeper vertical information. We are able now with knowledge bases and pulling things from Confluence or SharePoint or others to get even more knowledge and more data. But your models and your answers are only as good as your data.

He is hitting on something strong: you want to get as much data ingested, but you want clean, scrubbed data that has some effectiveness in it, even if it is unstructured. We are getting to a world where the more structured and unstructured data that you have that is scrubbed, the more precise with post-training you can be to give answers more accurately to customers, employees, or other businesses. Data is really important. We have gone through a whole process internally at Zendesk over the last two and a half years that was not sexy, but we spent a lot of time with our internal data, scrubbing it and putting it into the data lake so that we could use it in a bunch of different AI applications. Until we did that, we had a bottleneck on using our data. I am not talking about our customer data; I am talking about internal data to run the business.

Michael Krigsman: I think smart companies have taken this on as a project because it is not fun, it is not sexy, but it is going to have a dramatic impact on how a business operates in the coming years. It is the foundation.

Tom Eggemeier: It is the foundation, and there are some cool use cases right now. I have been talking to a bunch of startups on reporting and analytics. We just acquired a company that does some cool stuff in reporting and analytics where you are getting to the point that instead of getting a report, you can use a natural language query: "What is the year-over-year increase in automated resolutions in Malta for our customer base?" You can go get that data with natural language queries that can give you insights. You can go prompt reporting and analytics to give you more forward-thinking things: "What should I be thinking about to run my business? I am concerned about Malta, and I am concerned about how our customers are getting value from our product." You can get prompts and ideas back. That is shifting from old static reports to real insights. You can do that through natural language.

AI's Impact on Jobs and Customer Experience

Michael Krigsman: Brandon Bird on LinkedIn says, "When you increase self-service levels, how do you address agent burnout?" I'm going to ask you to start answering these, Tom, relatively quickly because I want to get to everybody. We have a lot of questions, and I have my own questions.

Tom Eggemeier: With self-service, you are answering the easy questions. Human agent burnout is real; it is one of the most stressful jobs possible. You can help them get better at their job through Copilot. You are serving them the answers, which can help with burnout because they are getting more accurate, showing more empathy, and getting it right with customers.

Michael Krigsman: Here's a question from Preethi Narayan, who says in her region there is a lack of computer literacy. What can a company like Zendesk do to make those computer interactions easier for folks who are not computer experts?

Tom Eggemeier: We have a foundation and a program. We have trained over 10,000 people who might not have the same kind of access as there is in other parts of the world to become customer service agents, giving them computer skills and customer skills. That is one of the things we are doing at Zendesk. Doing right is doing good. More and more companies are going to have to figure out how to help people become more computer literate, more customer service literate, and become great customer service agents.

Michael Krigsman: Tom, how has this AI trajectory affected how Zendesk thinks about talent, team composition, and jobs?

Tom Eggemeier: We have not gone as radical as other people who have said, "You need to justify any job requisition with why AI cannot do it." But when we talk to talent now, one of the first questions I ask, even for C-suite employees, is, "How are you using AI in your everyday life, professionally and personally?" In this AI era, if you are not using the tools, if you are not experimenting, and you are not getting better, I do not think you are going to have the skills and capabilities to succeed. That is one of the things that we have talked about.

Preparing for the AI Era and Data Privacy

Tom Eggemeier: We are going to start a big push. We have had a bottoms-up approach up until now about how our employees use AI, and we have opened up an enterprise license for ChatGPT. I use it every day. You are going to see more top-down pushes, encouraging employees to use these tools more because it is good for the company and it is good for their career.

Michael Krigsman: Can you talk about the future, and where is AI customer experience headed?

Tom Eggemeier: We believe at Zendesk that 80% of interactions will be through automated resolutions in five years, so they will not involve a human in any way, shape, or form. We think 100% of interactions will involve AI in some way, shape, or form within five years. That means that last 20% that humans are solving, they are going to do that with the assistance of AI.

We do think there is another trend, though. Once you start getting quick satisfaction, which is the opposite of dealing with that chatbot you and I talked about earlier, consumers, employees, and other businesses are going to interact with those companies more and more and have more interactions. We think interactions are actually going to explode, particularly when you start using your own AI agent and tasking it with doing something. Those are some of the trends that we are going to see in AI over the next five years.

Michael Krigsman: What about the impact of AI on jobs in general across the economy? Is that something that you have thought about at all? Where is this going from a jobs perspective? What jobs will be affected and what will not?

Tom Eggemeier: I have definitely thought about it, and I will give more of a personal story. I have a 22-year-old daughter who just recently graduated from college and a 20-year-old son who is in college. They have asked me, with the advent of AI, what job they should be doing. An answer 15 years ago would have been to go into coding; software engineering was a great place to go. I am telling them a few things now.

1. Be knowledgeable about AI. Use it in your personal and professional life. That is going to distinguish you from other people.
2. I tell them it is going to go back to the basics. If you can read well, write well, and analyze well, you are going to be fine because no one can predict how the job market is going to evolve over the next 40 to 50 years. That is their career trajectory. We have seen what has happened in the last three years with AI. I really encourage people to get those basic skills.
3. I encourage them to learn how to prompt, which is tied to the first two points. I have experienced this over the last two or three years, and my prompting is getting better and better. That skill is going to be great. It is interesting; I think some of the liberal arts education, where you learn to write well, analyze well, and read well, is going to be more and more important in prompting.

Michael Krigsman: On the prompting topic, for folks listening, we did a show specifically on prompting for business leaders. Check out cxotalk.com and go back a few episodes to find that show. It was really great. Very quickly, data privacy and security of your data. Can you talk about that? It is an extremely important topic.

Tom Eggemeier: We take data privacy really seriously, and it is part of that white box approach. We are out in the forefront right now on trust. The number of certifications that we have to ensure we can protect our customers' data is paramount to us. Second, we want to make sure that we can personalize things for you but not in a creepy way with companies, and we work with companies on that.

We are going to be in an era where that white box approach versus the black box approach, and exposing how people are using your personally identifiable information and how they are securing it, is going to be more and more of a topic over the next five to ten years. It is absolutely at the forefront of our AI approach.

Michael Krigsman: With that, I'm afraid we're out of time. Tom, thank you so much for being a guest on CXOTalk. I'm so grateful to you and for your team at Zendesk.

Tom Eggemeier: Great questions, great flow. I really appreciate your time, Michael, and I appreciate the opportunity to be on with you today.

Michael Krigsman: Folks, before you go, now would be an excellent time to subscribe to the CXOTalk newsletter. Go to cxotalk.com. We'll keep you updated on upcoming shows. Have a great day, everybody, and we'll see you soon. Take care.