From Reactive to Agentic: What an AI Native Contact Centre Actually Looks Like - Zendesk

In this CX Today interview, Rob Wilkinson speaks with Jonathan Barouch, GM of Contact Center at Zendesk, about what “AI-native” and “agentic” actually mean in practice, where enterprise teams get stuck (data silos, authentication, brittle infrastructure), and how to roll out automation without betting the whole operation on day one.

The conversation also challenges a common CX metric: containment. Barouch argues leaders should measure resolution, not deflection, and shares how teams can pilot agentic workflows, run A-B tests, and build a learning loop that turns real interactions into better knowledge, self-service, and agent support.

Transcript

SPEAKER_01 0:08

Hello and welcome. I'm Rob Wilkinson and today we're taking a closer look at what an AI native agentic contact centre actually looks like. So if you're challenged with moving past a basic kind of scripted bots or if you're exploring autonomous resolution as an opportunity to innovate, stay with us and you'll get a practical takeaway that you can use in your business. Because today I'm joined by Jonathan Barush, GM of Contact Centre at Zendesk. He's an expert in this space. He's got a really clear view on the decisions and the risks, all the opportunities that are out there for leaders and what they're facing every single day. So welcome Jonathan. Thanks very much for joining me.

SPEAKER_00 0:51

Thanks for having me. Wow, you set me up. Expert and the agentic contact center like rolls off the tongue. So we've got a lot of cut a lot of ground to cover.

SPEAKER_01 1:00

For sure. Yeah, and it's and these are sentences that we weren't saying many years ago, right? These kind of didn't even know what they meant, right? Exactly. But it's the space has moved so fast and now it's just like it's everyday. Um bit of maybe a bit of a buzzword. So I guess probably before we get stuck into stuff, for the audience's benefit, can can you kind of give us a bit of a background on this agentic buzzword that we're saying and why it matters so much at the moment?

SPEAKER_00 1:29

I think context editors have always been at the forefront of automation, right? Whether it was uh IVR moving into an IVA, whether it was a QA, um, you know, you had leaders in the field 10, 15 years ago that were innovating in this space. I think the the thing that really changed it was obviously the Chat GPT moment, where we saw AI agents responding in natural language, um, almost like magic. And I think because we're seeing these agentic experiences like an agent that can reason, that has sort of human-like speech, that can understand the context, and we're seeing them in our everyday lives. Contact center leaders are getting a lot of pressure from their customers and from their boards and from their execs to bring those kind of magical experiences into the contact centre. So I'd say nothing new, but the capability's just been absolutely turbocharged.

SPEAKER_01 2:20

I love that turbocharged, and you're right, it's it's not new for contact centres. We all we've always innovated, we've always been, you know, the the the the at the forefront of new technology. So I guess we need to be careful because in the past I think we've had habits of you know making mistakes when adopting new technology. So what where are with this whole new move, where are organizations feeling pressured most directly when they're still relying on the kind of the legacy and more reactive models than than their options that they do have?

SPEAKER_00 2:57

So we've just put out some research which people are free to go and download from Zendesk. We interviewed about 1,100 contact center leaders uh from our customer base just to get an opinion on that question, like what's holding them back, what are they seeing? And we had this really interesting paradox. Um, more than half the leaders were saying that volumes are increasing. Um, and 70% said at the same time their management team are pressuring them to bring cost out of the contact center. So you've got increasing cost, you've got sorry, increasing volume with the pressure to take cost out, and you've got this disruptive technology which is sort of enabling you to be able to balance the books and to be able to do both, which is kind of AI. And so there's this extreme pressure. The challenge you've got is most of these leaders go on to say that all of their data is siloed. And in fact, I was with a customer the week before last on the West Coast that said, love the technology and they've played with it, but their IT team can't even build them an API. So they can't even access the data, they can't even authenticate the customer to go and do um you know multi-step processes with an AI agent. And the best they could do is repeat sort of a dumb knowledge base. And so we sort of had to reimagine like what could you do to sort of set them on the path away from like essentially a dumb bot to an autonomous agent? And and the answer in that case was in-app calling, because within their app, the customer was already authenticated. We knew who they were, and that unlocked a lot of extra capabilities for service that wasn't available because they didn't have sort of custom APIs or the ability to authenticate in the IVR. So sometimes it's something small that can kind of trip you up from going on that journey. Um, but the pressure is there, the pressure is real. Um, and so it's how do you find those little moments to kind of drive uh the organization forward?

SPEAKER_01 4:42

Ah, so okay, so internal capacity resource was the stumbling block rather than the technology, and I guess and we can come on to the technology in a second and we'll we'll talk around that because it's important to plan for these things and include those people in the the decision-making process, right? Um but we before we kind of go down go into that, I I just want to think about in addition to like what what the kind of organizations are feeling, I want to understand what it feels like for those teams on the front line. Um, if they're trying to still get work done using those you know, scripted bots, you know, basic uh dumb bots as you called it there, what are the consequences for the guys on that frontline team?

SPEAKER_00 5:22

One of my American colleagues joked with me today that I obviously I'm Australian, so your your listeners can understand that. Australians have between 50 and 20 different ways of saying yes. You know, we would call that 20 different utterances, which means yes. And so in a bot world, in a like uh V1 of an IVA, you would have to sort of map every one of those utterances to a specific intent. And if you missed one or two, you'd have fails, and that's where you get the press one, press two, help operator, help operator. And so over time, you've literally got human resources who are listening to call recordings or mapping unmapped intents, and that's sort of how these deterministic kind of QA bots are not very um flexible when things change, very pre-scripted, very QA based. I think what's remarkable about this new kind of LLM-based technology, um, they can understand the context really easily. So you don't have to do all of those mappings. And you can even switch language mid-sentence, and our AI agents at Zendesk can actually follow the conversation with you. Um it was interesting. We I was within the customer meeting yesterday, um, and the customer was really pushing us hard and said, Oh, what happens if I switch from English to French? So try it. And again, that the bot immediately understood and continued. And they said, Well, what if I want to change my email from customer.com. Actually, I want to change the domain to gmail.com. And same deal. The the a the agent understood the intent, which was changing the domain, but keeping the prefix of the email, and it automatically switched it. And so in a scripted bot world, that's really, really hard to do to understand that context and kind of build a procedure around it, whereas that sort of comes out of the box with like an advanced AI agent.

SPEAKER_01 7:06

Yeah, I I I'd I'd argue just from first hand experience, it's not just it's not just difficult, it's it's nigh on impossible, really, because you you can never think of everything possible connotation or combination of words that that might be said.

SPEAKER_00 7:20

So yeah, why why rely on that when it's not or have multiple multiple intents in an argument, and then a bot goes, well, oh gosh, am I resetting a password or am I issuing a refund? And then these things kind of fail. But again, they are these new agentic tools, can reason, understand first, we'll help you with a password, and then we'll move you to a procedure to execute to kind of organize the refund. And so you can actually get quite complex calls happening now in an automated fashion.

SPEAKER_01 7:49

And it's it's rapidly um evolved and got better and better in terms of this technology. So even from 12 months ago, let alone kind of two years ago, it it's it's changing all the time. It's it's so important to keep ahead of that kind of technology, and also not to go and rule it out because you maybe looked at something 12 to 18 months ago and it it didn't quite stack up to your expectations because it is it is evolving it is evolving that quickly. I I guess uh talking about the technology, we should we should jump into that as well. Um but when we talk about moving from those kind of conversational models to to an AI native platform with the kind of contextual intelligence that we're talking about here and the memory that goes with it, what what how does that play out in practice? It sounds great, but how does it what does it look like for for the actual users at the end of the day?

SPEAKER_00 8:44

So look, I'm relatively new to Zendesk, and what struck me was um at Zendesk we have 20,000 customers globally using um our AI capabilities. And for so it's one of the largest AI for service businesses on the planet. And what that I think gives us and gives me is this view incredible view into what's working and what's not, and how customers are adopting, and we're seeing some customers moving from single-digit resolution rates through to 80% resolution using these kind of um agentic tools. So we're seeing um what good and what great really looks like. And in our business at Zendes, we sort of have AI in two halves. Half are the AI agents that we've just spoken about that can reason, that can operate within guardrails, that can execute procedures on behalf of the end customer. Um, they can even start to create new procedures or build new AI agents based on the learnings that they're getting. And so we've just bought a company called Forethought, and the whole value promise there is they can actually create new procedures to solve new problems as it's actually seeing interactions with customers. The other half of our story is Copilot, which is where it's sort of the AI that's pointed to the human, the human in the loop, to make our human associates kind of even stronger and stronger. And there as well, we sort of talk about this resolution learning loop. How does the system get smarter every time there's an interaction with a customer? So, one example I give in my business, in the contact center business, is at Zendesk 100% of the calls can be transcribed. They can be put into the cases, and then we can have AI interrogate the knowledge base to say, hey, there's some gaps. Hey, Jonathan, there's a gap. We've just seen this new kind of uh problem come up. We've seen 10 agents resolve it in a certain way. Let's draft an article, and the AI drafts the article, it's human reviewed, and then that article goes back in to help with self-serve or to help it with co-pilot with the next agent. So you get this kind of resolution learning loop where the system is self-healing, it gets smarter and smarter the more it more it gets used. And for me, that's the promise of AI unlocked in the contact center. We've been talking about it for a long time. Um, but to actually see it come live at this kind of scale um is pretty magical.

SPEAKER_01 10:56

It it's it's it's a super exciting time, and um I think in contact centers we've kind of we've had the ability to record uh every conversation for like ever, it feels like 20 years of time.

SPEAKER_00 11:09

But you've done nothing with the data, right? We've just sat on it.

SPEAKER_01 11:12

Yeah, and there's just this light bulb moment, I think, in so many contact centres now. That the reason we're telling customers that we've been recording these calls, actually, we now have a really great reason, and actually we're doing something with that information across every single conversation, not just a tiny sample size that someone manages to get around to checking, right?

SPEAKER_00 11:30

So there's well, I think some of us are, and I think that that's the big challenge for contact center managers to unpack the hype from the reality. And I mean, uh the reason I joined Zendesk is I'm excited because we are bringing that promise of AI to 20,000 customers and growing, not in a lab, but actually in the wild helping end customers. And I think not all CX platforms are born equal, and I think um we're obviously pretty proud of the scale that we have and the fact that we're helping real customers with real problems.

SPEAKER_01 12:01

And that's the the the that's ultimately the the only goal, isn't it? So um I guess um you mentioned uh a containment rate shift, sorry, not containment rate, uh a resolution or success rate shifting from single digits to 80% there. Um it's important to clarify what what that metric is, what that number actually means, because there's lots of new metrics out there with this new kind of way of doing things. Is that resolution or because there's a big difference between resolution and containment, right? And I think it's important we clarify that for the audience because they're very different things and they don't always mean the best thing for the customer today.

SPEAKER_00 12:36

Yeah, correct. I mean, containment is almost like how do we stop them hitting a human? And I think that's the wrong thing to focus on. We're sort of singularly obsessed at Zendesk about the resolution, like actually driving an outcome based on what the customer intent was, making sure we've solved the problem. In fact, we're we have such strong conviction that that's the right metric, not AHT, not first time uh first call resolution necessarily, um, not um average speed to answer. I mean, what does that even mean in a world with AI? Our metric really is resolution. Did we resolve the issue for the customer that they set out to call or to email about? And we have such strong conviction we actually have even started charging based on that metric. So we will only charge our customers with some of our AI products if they've solved the problem. And so our economic interests then are like totally aligned with our customer that we want to make sure that the end customer is satisfied and our technology is sort of in pursuit of that end goal.

SPEAKER_01 13:34

So you beat me to it because when you mentioned the resolution element being such a fundamental thing, I was going to ask you, are you putting your money where your mouth is and you've got there before me? So good and good work.

SPEAKER_00 13:44

We're starting to, and more and more of our customers are moving to that model, and we sort of are pioneering that model, I think. We see a lot of the rest of the market sort of waking up to that and kind of almost following in our tracks because for us ultimately, you know, you're only as good as the the the resolution, the outcome for the customer, and therefore we should we've orientated the whole business towards that outcome.

SPEAKER_01 14:06

It really is um the future if we're talking about that's being actually deployed, so that's fantastic. Um and I think it yeah, you've got really got skin in the game there, haven't you? So yeah, I think that's awesome. Um and actually it builds nicely on on uh a recent piece we did um with uh with Zendesk, and it was all we're talking about, um evolving from cues to conversations. Um you've said yourself you've got quite a great vantage point and lots of customers to draw kind of information from. What what evidence would you say you can sell, you know, early signs that you you might be able to share that that might convince um even like the most cautious buyer that these autonomous resolutions are ready for the real world and then it's not too soon to look into this further?

SPEAKER_00 14:54

I don't think it has to be binary, and I don't think we're trying to convince that naysay that they have to go all in. So that would be my first point. We have lots of customers that would have no automation in some really uh extremely high-stakes uh lines of business. It just goes through a very basic IVR or straight from agent. They have some that is deterministic, like just a pure QA thor, um, you know, perhaps areas where you don't you don't there's like just no um variability, it's very black and white. And then introducing sort of agentic workflows, you know, Q by Q or intent by intent to start testing and learning. And I think actually that's the advice I would give. Um it doesn't have to be all or all or nothing. Figure out a few intents, figure out where you have the API, where you've got the cleanest data, where you've got the ability to execute, and maybe do A-B tests and kind of start there and grow. And I think the customers that have been the most successful uh have started that way, like taking division by division or use case by use case or brand by brand. And when they see the resolution, they see the outcomes that it drives, they see in some cases um some of these AI resolutions are stronger, faster, better than actually putting the customer through a convoluted process that hits a human agent, and that builds the confidence to go back to the executives and the board and say, we've got the right guardrails, we've got the right system, let's go. And obviously, choose a partner um that it wants to go on the journey with you, not just a vendor, a partner that's doing it at scale, um, that understands your business and that will be there in the trenches with you. And I think there are many great platforms, and obviously I'm biased. I think Zendesk is one of those that um you know folks at home should consider.

SPEAKER_01 16:33

Yeah, it's it's a really good point about kind of you know starting in a sensible way, in a almost a de-risked way with conversations that don't open you up to potential problems and make things straight more straightforward uh for the testing. Um I think that's a fantastic tip. Um so we're running out of time. I could talk to you about this all day, but we're running out of time. So I'd like I'd like to kind of build on that tip uh and just maybe if you could only give like one piece of advice to our audience who's watching this um around maturing the current CX model um you know in in line with all this new potential that exists, what would that what would that one thing be that they should go for and do this next quarter?

SPEAKER_00 17:18

And look, I think I'm gonna answer it in a cheeky way because there's no one size fits all. But I would say um what I tend to do with customers when I'm on site is try and ask them what their biggest problems are and actually get them to show me the customer journey. We we call into the call center together, we open the app together, and sometimes that's sort of the nugget or that's the thing that gets the customer to go, oh, that's the thing I want to automate first. So the authentication story that I gave you earlier with with the customer, we discovered that because I actually we actually called through the contact center and kind of then actually saw what that experience was like, and it wasn't great. And we tried to kind of riff off each other to figure out what would a great experience look like, and we couldn't get there straight away because of the APIs, and we found a midground on how we could automate, introduce AI, and sort of solve the problem. So I'd say get a whiteboard, get a texter, uh, get a few folks who know your business or call in and find that one pain point that you can improve tomorrow.

SPEAKER_01 18:15

Yeah, I think that's so important. Um, and all of those approaches, it ultimately what you're doing is putting yourself in the customer's shoe, right? So you can never go wrong when you're walking the customer journey yourself. So I think that's an up tip. Okay, so unfortunately that is everything that we do have time for, Jonathan. Thanks again uh for joining me, answering all these questions.

SPEAKER_00 18:35

Um next time I'll send my AI to interact to interact with you, I think. The IVA, yeah.

SPEAKER_01 18:41

And mine can welcome it, and yeah, well it's safe. When when we're on the opposite ends of the planet, it might make sense to do that one day. Um, but that's a whole nother conversation. Uh before we do close up, just for anyone watching uh who wants to kind of explore this in more detail, what what's the best way then for for them to kind of find out more about Zendesk and and and maybe to get in touch with someone?

SPEAKER_00 19:03

Yeah, I think follow us on LinkedIn or head over to our website. And as I said, we've just put out some great research on the Agenti Contact Centre. All of our research, uh at no cost, download it. It's a great report, and I think it will help spark some conversations within your businesses.

SPEAKER_01 19:16

Amazing. Um I I'm a I'm a bit of a geek, so I'll be reusing that myself. Um, and just for our audience as well, don't forget, um, you can also find a wealth of related resources and stories, videos just like this one at cxtoday.com. So check those out too. Um and that wraps things up. I I'm Rob Wilkinson from CX Today. Um, thanks very much for joining us.