Bit by Bit: Leadership Conversations | Paying the Service Debt

Added: 2026-05-22

[00:00:00]Does anyone have enough? Do you talk to your customers often? The answer is no.

[00:00:05]>> [music] >> Hello and [music] welcome to bit by bit leadership conversations. I am your host Carolina Milanese and today we are going to talk about paying the service debt. I have a on location which doesn't happen very often with Adrian who is the CEO of Zendesk and we are at day zero of Relate. Yeah.

[00:00:38]And I stole the title for my podcast from something that you said and I'm going to get to it in a minute. But >> You been spying on me have you?

[00:00:46]>> I have a little bit yes, stalking on LinkedIn and all sorts of things. But what I want to start with is the fact that there seems to be a disparity between is like you know walk the walk but or talk the talk but don't walk the walk. When we're talking about the fact that if I think about customer service is one of the most uh rewarding if you like part of the business when it comes with customer interactions but it's probably one of the most that are sub funded or underfunded.

[00:01:21]Um That's not how leadership talks about it in in an organization. They talk about this is high stakes but yet we're not there. Why is there this discrepancy between what the executives say they do but then how there's low recognition on the services business.

[00:01:43]Mhm. I will I think firstly if you think about the modern business organization right? It's a it exists almost to to optimize to extract value from the resources that it has. And so customer service is going to be no different. And so, we may have started with these high-valued intentions of engaging with our customers, but those the inner accountant in all of us takes over and drives efficiency, right?

[00:02:10]But if we if we go back to the value, you know, customer service is for most businesses, it is the tip of the spear. It is the place where you actually talk to your customers. I look at Zendesk's own data, you know, Zendesk on Zendesk, And I see that um maybe I I forget the exact number, but like 40-odd% of our customers contact us, right? In a given year.

[00:02:36]I think the given year was 2025.

[00:02:39]Of those, they represent 95% of the value and 94% of the growth.

[00:02:44]So, it is your most important customers that are talking to you, that are having that conversation. And I think ultimately, customer service has perhaps been constrained by the fact, and this is something I'm sure we're going to talk about, >> Yeah.

[00:02:59]that it's been a human-powered machine.

[00:03:02]And human power is expensive.

[00:03:05]And I think when we remove that constraint and the means of production becomes in whole or in part, probably not in whole, uh machine-powered and AI-powered, and we live in an era of abundance, we can think differently about customer service, and we can be unconstrained. And we can say, well, do we have enough? Does anyone have enough? Do you talk to your customers often? The answer is no.

[00:03:32]You know, I work for a software company.

[00:03:35]We can never have enough software, Right.

[00:03:38]all of the money we spend on the agentic coding tools, I think is valuable. I mean, I waste a lot of it personally, but in general, all of it is valuable, cuz we're creating more product. We're creating more ultimately, like differentiation and and, you know, value. And I think in service, we're creating more contact.

[00:03:58]We're creating more loyalty. We're creating lifetime value with our customers. But most importantly, we're removing the reasons that they would move away from us in the first place.

[00:04:08]And that's the key.

[00:04:09]But I think is interesting when you're talking about service debt. Because it feels to me that we're starting at minus one versus zero in building this relationship. Tell me more about what you actually mean by saying service debt.

[00:04:26]Well, um recently, like very recently, we I've been looking at the number of interactions and contacts that our customers have who are automating their service, right? And which you could think of as proof point or the existence proof of what happens when you automate.

[00:04:44]And um I see customers that are automating service, the data that we see is that that total interaction volume begins to grow.

[00:04:55]And so, let me let me say that again.

[00:04:57]And the higher you automate, the more it grows, by the way. Mhm. And so, my conclusion from that, right? Let's say it again. So, if I'm automating, I ultimately get more contacts from my customers. Mhm. Right? Why and you know, breaking it down, I think we think that's because it's uh a manifestation of what economists call Jevons paradox. So, Jevons was a um a British economist who during the Industrial Revolution, which took took place in, you know, uh where I'm from, in the north of England, um he looked at the consumption the the efficiency of steam engines and the consumption of coal. And the assumption was, when we got these much more efficient steam engines, right? You know, James Watt invented a more efficient steam engine, that we would consume less coal. The actual opposite was true. It opened up more uses for steam power. You know, we mechanized more of the industrial complex of England and the world and and you know, the industrial revolution happened.

[00:05:58]So, when you make something more efficient, you use more of it. That's Jevons paradox.

[00:06:03]I think as we make service more efficient, we can see it in this data, right? We'll use more of it and that is the service debt cuz we are taking a step back.

[00:06:12]What were those people doing before?

[00:06:15]What this this 2/3 of customers that were not contacting you or these 2/3 2/3 of contacts you were not getting, where were they going? Probably going somewhere.

[00:06:23]>> They were never they were never coming to you in the first place, right? Those are the customers that gave up on hold.

[00:06:28]Those are the customers that were like, "Oh, this is too much trouble." You know, those are the customers that were like, "I don't want to fill out form."

[00:06:33]Those are the customers that were like, "He was rude." Um and they went away.

[00:06:38]And so, that is our debt, right? Until we see until until people like, "Okay, I don't want to talk to you anymore." or like until, you know, you've hit sufficiency and maxed out, then I think you have a service debt and I think you can, you know, as the economics the unit economics of customer support change, at least you lose that reason for why you can't scale your service and do more.

[00:07:03]You're making me uh laugh a little cuz when you say, "Where are those customers going?" When I first moved to the UK, a two uh call center was in Scotland. Mhm. I could not understand these people. I moved from Italy. I was like, and the the amount of times that I would call and hang up because I couldn't even go past "This is James."

[00:07:25]>> Oh, you've used all your minutes.

[00:07:28]I love that idea.

[00:07:30]A call center a contact center staffed by the cast of Trainspotting. That's that's good.

[00:07:34]>> [laughter] >> That would keep people engaged.

[00:07:37]>> I think so. I think so.

[00:07:39]So, let me dig into some of the numbers that you said earlier because your number was a cumulative number, but then if you look at how you break that down. So, 40-something, 45, 47% of the people that um interact with you. Of that, about 25, 30% are people that have like a routine low-value problem. And then you have another 25% that is the very high-value problem. So, you probably want the humans to concentrate on the high-value interaction. Mhm. But we have a system that is being set up from a workflow perspective >> Yeah. in those tiers. And if you change that, what are the people going to be doing going forward? How are you thinking about that?

[00:08:31]Um yeah, it's it's a it's a great question. So, we took um I think about 15 million conversations, you know, tickets, as it were.

[00:08:44]Um and we every ticket that comes into Zendesk, we actually look we use AI Mhm. to figure out what is the intent of this user. And we actually have a taxonomy of about 24, 2,500 intents.

[00:08:58]Across, you know, retail return, product question, you know, uh missing missing item in order, you know, so on and so forth, right? There's these It turns out there about 2,500 reasons people call customer service on average.

[00:09:11]And then there's a long tail of things that are just specific to businesses.

[00:09:15]So, we if we break down those intents, we actually ran them through a foundational model.

[00:09:21]Um and we said, "Okay, which of these are people calling because something was broken?" Like your company, Zendesk, company X, Acme Corp, whatever it is.

[00:09:34]>> Mhm.

[00:09:35]You you lost something, broke something, your product doesn't work, you disappointed the user, blah blah blah.

[00:09:41]And actually that sort of break fix motion is kind of why customer service exists for you. But that that represents about 47% from memory of inquiries. On average. Varies by industry. Software is about 50% of 50% Software is 50% failure. I've spent 30 years of my life in software. I could This seems accurate. Um but uh those those contact reasons actually represent the purpose of customer service cuz ultimately our businesses should go fix all that stuff, right? We shouldn't be using customer service and bots and tools and blah blah blah. We should be like, "Yeah, maybe we should not leave the spring rolls out of the bag before it leaves the restaurant, right?" Just to you know, tie back to my own personal problems.

[00:10:27]Um maybe the trousers should be sized correctly. You know, or maybe Adrian should be in a one of by the right size.

[00:10:33]Uh but whatever it is, like you can fix a lot of the stuff in this business. But I'll you know, we in in engineering we talk about runtime environments. What's the runtime environment for this product for this product, right? The runtime environment for me as a CTO is probably PowerPoint, not code right now. The runtime environment for fixing the problems in your company for 40 years has been the humans of customer support. If you start doing something wrong in a new way, you write a little procedure or a script and you hand it to Carolina on the phone and she's like, "Uh yes." Or maybe if you're working for a company you'll be like, "Oh yes, this is the way we go." You know, however it is, right? But that's the way you solve problems. That's the runtime environment. Uh Amazon don't do that as an example, right? They have no customer service number because they just go fix the problem. Yeah. Right? If you call and say, "Where's my order?" they add a where's my order button and you just press the button. Now, you may not find the button and that's a different question, but basically fixing problems upstream is what we want to do.

[00:11:29]We know most businesses won't or not able to fix everything and the real world is complicated and messy, so it's kind of difficult, right? We're not all operating at the scale where we can software code it.

[00:11:41]>> Yeah. And so those are going to exist.

[00:11:43]And I think the goal of service is to automate with as low friction and as much pleasure as possible.

[00:11:51]Automate as many of those as you can, right? Because they're kind of low value. The second category is explicitly low value. What is low value? That's the people who hit zero to get a different answer, even though the answer they got is the only answer. That's the people who are saying, "Where's my order?"

[00:12:08]looking at the "Where's my order?"

[00:12:09]button, but they're going to call anyway, right? That's the humans of support. I mean, I am a perennial "Can I speak to your supervisor?" person if I don't get the answer that I want. I don't know why. Yeah, I don't do that anymore cuz I don't want to be called a Karen, so >> [laughter] >> There is no What is the male equivalent of Karen? I believe it's Adrian.

[00:12:29]>> [laughter] >> We're working on it. Who knows? Who knows? It's a good question. Um but that's what that is. Think about low value queries, right? Which is about quarter. Yeah. Is They don't go anywhere.

[00:12:41]People will, you know, we're all we're human. Sometimes we just need to talk to someone cuz we think we really need to talk to someone and that's okay. It's just feeling hurts.

[00:12:52]>> You might You might be a brand that doesn't want to talk to people and that's okay, too. You make choices, whatever.

[00:12:57]But ultimately, I think for most customer service organizations, they're going to still exist 20, 25%.

[00:13:03]And then we we get to the high value, which is the conversations you should be having, the conversations that are high value. The opportunity to expand someone's product use, the opportunity to basically enrich someone's perception of your brand or legitimately make their experience better.

[00:13:19]And I think there to me, right? Are humans in customer support um need to move away from the that can be automated because the human in the loop doesn't add any value Mhm.

[00:13:33]to, you know, spending more time on the inquiries where the human in the loop does add value.

[00:13:40]And in an ideal world, a world that I think we'd all love, we could convince ourselves that customer support is a growth mechanism because you talk to the customers that are the most likely to grow with you.

[00:13:52]That you're adding an enormous amount of value. And that you should be taking sort of what you were spending in humans on tier zero and tier one, the things that you've automated, even the high value that you've automated in some cases, and moving it to those value conversations, right? That's what I believe, tying it back to the service that should be happening.

[00:14:13]It isn't necessarily 100% what's going to happen, but thankfully, >> [clears throat] >> I believe sort of this initial rush of excitement that um, we had as existence in a capitalist business society that we can automate everything and save a whole bunch of money. I think some of that has gone away, right? One of my favorite examples of that, by the way, some, you know, uh, an example I love that I heard recently is from someone who said, "Yeah, he's like, you know, the newspaper industry, when uh, which when the internet came along, you know, cuz newspapers are effect, were effectively printing presses with trucks that distributed pieces of paper, right? They did content aggregation and local advertising, but that's what they were.

[00:14:56]The internet appeared and they were like, "Sweet. We don't need the printing presses anymore." It didn't work out like that for them, right? That was not a good interpretation of the business model change that was happening. Google it, local advertising and content aggregation when the means of production is free, you know, just became a thing and, you know, we can think about aggregation theory and so on and so forth.

[00:15:19]I think in customer service, we have to then go, you know, we as the business model changes of the of service itself, I think we have to go back to first principles and say, "What are we here for? Who are we serving? How are we trying to engage them? And how does this grow the top line of the business, not just impact the bottom line as a cost?"

[00:15:38]Yeah, which I think is critical.

[00:15:41]I like what you said earlier about with AI, we are no longer constrained.

[00:15:47]And I think that the people that think that way, they're really reimagining customer service.

[00:15:54]They're not just trying to get rid of, you know, tier zero and one, but they're really trying to deliver on the promise that you were just talking about.

[00:16:03]How do we help your customers think about not just doing the bare minimum, which is zero taking care of zero and one, but really get to that high-value interaction. What does that level of customer service look like in your mind?

[00:16:21]Well, I think, um, you know, you can't [clears throat] manage and you can't change what you don't measure. Mhm. And so it it begins in my mind with really understanding what are the what are the intents that your customers are coming to you to you with? What are the sentiments? And then, what is the quality of the service that you deliver, right? And at Zendesk, we measure all of those things on every interaction for our companies' customers. And we just launched a new quality score that we put in every ticket, which is a sense of how much effort did you put in and how much satisfaction would you have from the answer that you got.

[00:16:53]Um, and the answer would be, "Yes, Karen, press zero." And she got to speak to the supervisor.

[00:16:58]Um, uh, maybe not. Maybe not. And I think that, um, measurement then tells you a little bit about, sort of, you know, what I like to think of as the the ticket estate. You have to look at all of the journeys together and think about it. And then, be a little deliberate in where you want to take your service interactions, right?

[00:17:19]>> [snorts] >> I think it it's natural, you know, that you're going to focus perhaps a little bit on value creation and service, right? And how can I how can [snorts] I move conversations towards um towards these sort of product introduction, product expansion ideas, product utilization or adoption ideas.

[00:17:37]And so I think our customers at the moment are really starting to think about measurement and understanding of categorization. And then honestly, like it's incumbent upon us as a vendor to do the things that I think we're doing, which is basically have this idea of a of a our resolution learning loop where we're making suggestions of like, you know, um our admin co-pilot will tell you it's like, you know, I see you apply this macro, this canned response a lot, but agents then have to change it or it would do or results in like lower customer satisfaction or customers seem to have a longer handle time when going through this AI agent procedure, you know, the procedure for returns or whatever it is, versus this other one for a product substitution. And so you can learn how to fix also what you have or improve and get this sort of continuous improvement, this idea of Kaizen within the system going and running, which I think is fundamentally important. And I do think, you know, um it isn't all about what our customers do. We as technologists have a responsibility to deliver tools that make their lives easier and automate so much of it. As we should because we also get to use an abundance of reasoning capability and read every ticket and understand every interaction and have access to all of the data.

[00:18:56]>> Yeah.

[00:18:57]And obviously, without data, AI is pretty damn stupid.

[00:19:01]Yeah, I mean, great AI implementations are driven by knowledge, you know, uh data assist and system integrations, right? To get you and you need to go.

[00:19:11]You're talking about a service architect. Mhm.

[00:19:15]Something that doesn't exist, but should.

[00:19:18]Mhm.

[00:19:20]What role does the service architect play in a team that is going to look more and more different than the way the team what teams were organized up until before AI? Mhm.

[00:19:35]I mean, we can think about the career ladder or the this the stack rank of jobs in customer service, right? As they as they existed, you know, today, right? And so, the rituals of customer service, right?

[00:19:47]Were established in large, fluorescently lit rooms in basements in Delaware, you know, doing phone service for enabled the credit card business to exist, frankly.

[00:19:59]And in those rooms, we created queues and holds and wait times. And is there anything else I can help you with today?

[00:20:06]And you know, all of those things like what what we call the rituals of customer service.

[00:20:10]The job families were created as well.

[00:20:12]Tier zero, tier one, triage. And then you had, you know, this all-important role of the the manager, the agent manager, the tier one first-level manager, right?

[00:20:23]Critical, critical. And then, you have these specialized roles like knowledge, workforce management, QA, so on and so forth.

[00:20:30]In an AI world, right? We've talked about the fact that the low-value, repetitive, easily automated conversations of tier zero and tier one, most of those are going to go away. And that's that's okay, right? Cuz you you can do those better with AI. You know, to a certain extent, I think if you if you talk to business process out so source operators, right?

[00:20:52]They actually function as large-scale recruiting organizations, basically just churning people through that layer of the organization, hopefully moving some of them out, but we just with generally with enormous attrition, right? I'm I'm sure that's your lived experience, too.

[00:21:06]Um and I think it's good to be able to automate some of that and put it then put people into these higher quality tier three, tier four jobs. I think that the first level manager role changes dramatically because you're obviously you're still recruiting, training, and career managing this group of actual humans who work for you, right? This team of agent in the of human in the loop agents. But you're also doing the same thing while you're training and understanding the workload and figuring out routing changes and behavior changes for the AI agents that are involved in the service delivery. A huge upskilling in that role.

[00:21:44]And then all those specialist roles, those probably weren't invented immediately in the base in the fluorescent lit basement in Wilmington, Delaware. Those came a little bit later as as a as a function matures, right? Over time, you develop new categories of job.

[00:22:02]I saw study recently, I think it's Goldman Sachs data or actually, and they showed all like employment today versus employment in 1940 and how many people were employed in jobs that did not exist in 1940.

[00:22:16]And yeah, you know, shocker, it was most of them, right?

[00:22:19]We didn't need podcasters in 1940. Um, you know, you would have been a radio presenter.

[00:22:26]>> [laughter] >> Probably better for everybody, but yeah.

[00:22:31]>> I like to say I've always been first made for radio.

[00:22:33]Um, but um, I think that those very specific jobs, workforce manager, knowledge manager, QA manager, so on and so forth, right? Operations manager, support operations, etc. I think that in a period of disruption, little chaos, Mhm. and change, um, you don't get to be that specific.

[00:22:55]What you get to be is more agile, more like uh, development, you know, I often tie jobs that look like that back to um, you know, product owner in a scrum in a scrum development organization or a product manager in classical software development.

[00:23:14]And for me, you know, I think it's it's hard to be that specific in a time of great change. And so I think at that point you need athletes and you need generalists and you need people with confidence to make changes who understand, you know, this idea of design, build, test, measure, fail, do it again. Cycle, iterate, and learn, right? And that that to me is the AI service architect who can look that job is, you know, in steady state and run rate is look at this intent that is coming in that seems to be spiking like maybe there's a drop in CSAT, maybe there's a there's a there's a blowout in time handled or like in spillover into escalations, whatever it is. And say, "Hmm, I'm going to break that down. I'm going to find out what I'm going to look at the data. I'm going to look at actual individual examples.

[00:24:14]Maybe I'm going to sit on some of the escalation calls. I'm going to figure it out and then I'm going to design a new AI process. I'm going to add new co-pilot. I'm going to go figure out the back office system integration that's needed. I'm going to build it in an AI agent builder. I'm going to build a custom agent that goes off and runs a bunch of work and does these things, right?

[00:24:31]You're going to go pick up the new tool chain.

[00:24:34]And you're going to use that tool chain and you're going to innovate. I think that requires um a bold and big skill set. Eventually, maybe you know, we'll go back to having, you know, integration engineers and AI designers and blah blah blah, right?

[00:24:50]Maybe we'll invent those new titles that came like podcasters since 1940. Um but I think that's going to take a bit of time and it's um little over ambitious for us to start building job architectures from that at the moment.

[00:25:05]I have a double whammy cuz I'm actually an an industry analyst and I am a podcaster. Two things that people have no clue what it means. So it's Um What you're talking about is fascinating to me because from from this idea of having people with knowledge because right now we're at a time where a lot of organizations are to your point about cost cutting. They're thinking, "Ooh, I have AI. I can get rid of X percent of people."

[00:25:35]With those people there's a lot of knowledge that they acquired over time. How do you suggest people go about making the right choice and basically not shooting themselves in the foot in getting rid of people that have so much knowledge not just about the company and how the company works but the products and the customers. Cuz as you were talking, I was thinking about how customer service has this huge complex hurdle than a lot of other areas that we can automate don't have, which is us. So a lot of interactions are not the same because the human is not the same, right? So the problem might be the same but the way that you escalate it versus not changes. So you know, that's why I think the architect is such a brilliant idea of wanting to then come in, sitting on a call, listen to what was handled differently. What happened?

[00:26:35]>> shadowing?

[00:26:37]Oh, maybe. Have you agent shadowed today? Uh that I but I think that that's where then the knowledge comes in, right? Of all of that. So how do you advise people to to balance that, to not getting just excited about again the cost saving I'm going to get rid of X amount of people.

[00:26:58]Yeah, I the I think it was a recent MIT study that showed that you know, it got a lot of headlines, right? Cuz it 80% of AI projects fail.

[00:27:07]If you read the fine print, 80% of of AI projects focused exclusively on efficiency fail. Yeah. Right? Um and I think that is you have to work back from a service design customer experience perspective, right? Remember, we're in we're in this case, we're in the customer experience business, right?

[00:27:28]That is our job. CX is our job. Good experiences are our job. Ultimately, it's about you know, um you want people sitting there not worrying about the 42 tickets they have to handle today. Just worrying about this ticket. Just worrying about this interaction.

[00:27:46]And so, one of the challenges in customer support is it used to be done in these big I said I'm going to say it again.

[00:27:54]Strip fluorescent strip light lit rooms and then the knowledge actually it was an oral tradition. It was passed from person to person, person to supervisor.

[00:28:03]And we used whiteboard cave paintings around there and there was lots of post-it notes. Have you ever visited work in contact center?

[00:28:10]Um I remember it was when I started at Zendesk the first one I went to.

[00:28:15]And uh there was a sticker on the desk.

[00:28:18]And it said, "This desk is used by Carolina between in the morning and Adrian in the afternoon and the evening shift was Bob, right? Or whatever." And I was like, "Whoa." Like, three people sit at this desk and answer customer inquiries in different ways. And then there was something on the on the on it which was like, "What Don't What are my plan and like don't know my Don't know my stuff over whatever." But, um cuz it was a human factory. Yeah. And I think the pandemic actually um which a real struggle for many uh, service organizations because so much of the knowledge wasn't captured. Right?

[00:28:55]Um, there's this convenience of you know, there's a joke in AI, right?

[00:29:00]How do you solve a problem in an AI system? The answer is with more AI.

[00:29:03]Right? Uh, so, you know, how do I test my AI system? Well, use another AI system to test your AI system. How do I test that one? With another AI system.

[00:29:11]And so on and so forth. I think in customer service, how do you solve a problem in customer service? With another human process. Right? Um, maybe you wrote it down or maybe you didn't.

[00:29:20]And so, what we need to be doing in these service transformations is A, working back from the customer experience.

[00:29:27]B, thinking about how am I capturing and understanding what good looks like and what best practices and all these other things because your AI agents do not read the whiteboard, right? They do not go, um, to the espresso machine, right? They they don't get to partake in any of these social activities and the the frameworks that bond us and bind us.

[00:29:51]They exist in a hermetically sealed vacuum of of a machine model. And so, we have to put it all in there in basically written and verbal form and have it have it captured. And I think that is what this end to end tool chain is all about and that is what like the process transformations that we take on are all about. But, it's ultimately work back from value, don't focus on efficiency only and, you know, try and create great experiences for your customers.

[00:30:21]Last question and it's not a technology question.

[00:30:24]>> Woah.

[00:30:25]You were talking about AI Does anyone want to answer it?

[00:30:27]>> If >> [snorts] >> AI >> [laughter] >> failures, but if you're looking at the biggest hurdle today in in AI uptake is actually us, human beings. And a lot of time actually leadership within an organization.

[00:30:44]How you dealing with this within Zendesk and then with your customers? Because there's so much change, the way that you're going to sell today for your salespeople is different than what you did 2 years ago.

[00:30:59]How do you make sure that your people are coming on this journey with you and then helping your customers adapt to this change, but really from a from a human perspective?

[00:31:11]Yeah, it's it's not easy, right? And I think it's good to acknowledge that it's not easy, right? I think we all have a little those little moments where we see someone flexing on what they're doing with Open Clawe or what all the agents do, right? You know, I'm I'm definitely liv- I'm living in the AI bubble. You know, I follow I listen to too many podcasts, I follow too many influencers and so on and so forth. Um and that sometimes gives me the sense of like, "Oh my god, I'm so far behind.

[00:31:44]Like what am I doing?" And then you start, you know, overreacting and everything else. But it's actually the tools are there to help us. The tools are there to to improve what we do, improve workflow. And I think the first part is beginning to understand um what we can each do to create better work product and be more efficient. And then ultimately I think um not worry about I I always counsel people, don't worry about efficiency, worry about creativity.

[00:32:18]Right?

[00:32:19]Um In service in in service design, right?

[00:32:24]Um that you know, there's a famous cliché in product management, right?

[00:32:28]Which is um someone, you know, Henry Ford Ford Model T um said that if I'd asked customers what they wanted, they'd have said faster horses.

[00:32:38]>> Faster horses. Right. And we are pretty clearly sometimes in the faster horses moment in AI deployment, right? Which is like if I And this is where the efficiency thing comes from. Oh, I can I can fire 70% of my customer service reps if I can automate 70% of my inquiries, right? Those that isn't true.

[00:32:56]It's just not going to work I don't think that's going to work well for you.

[00:33:00]And we have proof from uh quite famous and publicly shared companies that it didn't work for them in that way, right?

[00:33:06]They needed Moments that matter require humans. It's simple as that.

[00:33:10]>> Yeah.

[00:33:10]But I think uh my counsel to people is really like think about um what the non-faster horses version of what you can do is, right? You know, product an example and manifestation is you know, you couldn't you couldn't read every ticket for a given high-volume implementation of Zendesk, every customer conversation, to see if you know, any users were um >> [clears throat] >> They say you're a you're a you're a gambling company. See if any users were underage. You couldn't read any ticket and say, "I think he inferred that he was in school." You know, I think uh what you know, I got these signals from it. Or any other manifestation of human behavior that you may have regulatory reasons or just quality reasons or you want to know or find everyone who's going to churn. You can't pay someone to do that and read every single one, but it requires human intelligence. It's a human up until now it's been a human interactive task.

[00:34:08]But if you're not building faster horses and you're thinking what can I do with this technology? It's like, well, I can apply this intelligence given these guardrails to every single conversation I can have and I can do something that wasn't possible before, right? Which is extraordinary.

[00:34:22]And I I think, you know, that we're in this situation where we can have non-faster horses conversations about what we can do.

[00:34:32]But let's be honest, the you know, the models as they exist today we're in a very early stage.

[00:34:37]>> Yeah. Um if you go you know, one of the examples comparisons like if you go to the chat GPT prompt, right? And you go use it, which is something we all did and we had a moment and we're like, "Oh, this is cool." Right?

[00:34:51]It it's a blinking cursor on a blank box. Right? And it it's sort of like um um you're much too young to remember this, but when you got your first MS-DOS computer and it said C: greater than and then there was a flashing cursor.

[00:35:05]>> Yeah. And you're like, "Oh, I've got I've got a I've got a computer. I don't know what do I do now?" You know, and then you write a simple basic program that says, you know, print Adrian 10 print Adrian 20 go to 10 you know, goes on the screen. Like that's what everyone did. But it was really like, "What am I going to do with this computer thing?"

[00:35:21]And then we got apps, you know, that did one thing and one thing only that took over the computer and then we got Windows and suddenly it was democratized and we could do it and so on and so forth. And we grew and eventually we ended up with SaaS customer service companies, which is obviously the pinnacle of technology. But you know, we we first of all went through all of that.

[00:35:38]We're still at the blinking cursor saying, you know, um so do you really, you know, want to know about disestablishmentarianism in Krakatoa Krakatoa East of Java or is there some other thing that you're looking for, right? You know, that can answer any question in the human knowledge.

[00:35:53]And it is we have to be forgiving to ourselves, right? It is going to take time to do that. Our technology vendors and the people that we work with and the people around us within that problem spaces should be manifesting uses of this technology and I think as CTO of Zendesk that is our responsibility as Zendesk >> to begin manifesting really useful uses so that you don't have to do all of that work and that research and you know, I would love to hear the ideas of our customers and we do trust me, we do and we listen. But I think um we're we're I think the Zendesk product set that we're launching uh today at our Relate conference in Denver, I think that to me those are beginning to be examples of non-faster horses thinking in AI and customer service. And I think for every organization it's like think about the better experience you're trying to build, don't focus too much on efficiency, and then don't worry about the faster horses.

[00:36:52]I I appreciate that because it's a bit different than what I've been hearing lately cuz the faster horses, the you know, cutting time is how we measure people, how we measure our degree of success with AI. And I think if we continue down that road, we not only going to find ourselves in trouble, but we're also undercutting the real opportunity that AI brings. So, I'm an AI optimist and a customer service pessimist.

[00:37:25]>> [laughter] >> So, I'm hoping that now realist, maybe that's true.

[00:37:29]I'm hoping that really this will change things for a lot of organizations out there and that this will empower people to really put their value where it's going to be seen not as a cost saving, but as a added value to the success of the company. Well, let's drink to that.

[00:37:50]Thank you so much for your time today.

[00:37:51]Thank you, Carolyn. Thank >> you all for listening.