The Agentic Train Has Arrived! AI Agents and their Role in CX - Content Guru

The agentic AI era has arrived, and customer experience is leading the charge. But what does that actually mean for enterprises trying to plan, prioritise, and deploy responsibly?

In this interview, CX Today's Nicole Willing speaks with Martin Taylor, Co-Founder and Deputy CEO of Content Guru, about the real shift happening right now in AI-powered CX, and what leaders need to do about it.

The conversation covers: 

What the transition from GenAI to agentic AI actually looks like in practice, which CX tasks AI agents handle reliably today, and which ones still need human involvement, and how to build on existing automation investments rather than starting from scratch. You’ll also learn why the MIT finding that 95% of GenAI projects failed should inform your agentic strategy, a practical framework for scoping your first pilots, and why CX is the natural home for agentic AI adoption.

Taylor also addresses a question many leaders are sitting with right now: is it better to wait and let others test the water first? His answer is clear, but nuanced, and worth hearing in full.

Transcript

SPEAKER_01 0:08

Welcome everyone. I'm Nicole Willing, and today we're talking about something that's no longer just a conversation, which is AI agents. You know, they've arrived, they're starting to transform how enterprises think about customer experience, and we're going to explore what's changed, what's possible today, and how CX leaders can start planning to leverage this technology effectively. And to do that, I'm joined by Martin Taylor, who's co-founder and deputy CEO at Content Guru, someone with a sharp view of the decisions, risks, and opportunities that CX leaders are navigating as a genetic AI moves from conference talk to context center reality. Welcome, Martin.

SPEAKER_00 0:43

Good to be here.

SPEAKER_01 0:44

Thanks for joining us. So let's start with the big picture. You know, AI agents have obviously been talked about for a while, but what's changed in the last 12 to 18 months that really makes this moment different?

SPEAKER_00 0:56

Yes, well, we've been in the transition period from the Jane AI age into now the agentic age. So I think many people hadn't quite settled in the Jane AI age, but it seems that that was almost a short transitional period. So what's happened, I think, is we've seen increasing levels of automation in the past 12 to 18 months. There was some Opus research back in 2024, uh, and it showed that only 2.4% of contact centre spend was on technology, and 97.6% was actually on labor. And they predicted that by 2032, 58% of that labour would have been automated. So that looked like a massive prize to play for. And their updated research a year later uh showed that that 2.4% had become 4.6% of uh of 2024 spend. So therefore, I think we can see that in that last 12 to 18 months, there's already been a significant increase in the share of spend on CX that is getting into technology, and a lot of that has been around the AI space, where there is a good return on investment case from eating into some of that labor cost in order to deliver better service with intelligent automation.

SPEAKER_01 2:22

Yeah, so then when um what are your customers exploring then when it comes to AI agents and how is this supported by your brain machine agent?

SPEAKER_00 2:31

Well, our customers have always been very keen on responsible automation. And uh, we actually, as an organization, come from automation. Long before we ever were involved in contact centers, we used to do the TV voting. So an Anton Deck would put the number on the screen or whatever, uh, and then the the rush of calls would come in for who should leave the house or who should go to the next stage in the singing contest. That was really what we were all about. So we've been very comfortable in that automation space, and it's been about at what level can we responsibly meter that out into the customer experience space. So we've been fortunate in having a number of our customers who have been in a position to intelligently automate. Um, so one I would highlight would be UK Power Networks, where we're automating currently 94% of all of their customer inquiries to full resolution. So that's complete containment. Um, and yeah, that's been a steadily rising percentage uh over the years, and that's actually gone hand in hand with a steadily rising CSAP score for that organization. So it shows that if it's done well, uh then automating can really satisfy the customer. They're getting things done more quickly, after all, uh, they're getting to resolution faster, customer effort score should be kept to a minimum. And I think that's probably one of the contrasts with the earlier automation era. Uh at the beginning it was seen that the the saving was going to be in agent time, uh, and sometimes the cost would be in customer time. So customer effort score wasn't always a metric that uh organizations prioritize. Uh, but actually, you know, no surprise, customers hate filling in loads of forms or repeating themselves answering lots of questions over and over again, uh, just in order to save the organization a bit of time at their end. Uh, that trade-off hasn't worked so well. So I think it's now really about low customer effort score, high levels of automation, uh, and moving on from that IVR age through the Gen AI period and into full resolution and containment using uh new uh agentic AI agents. Really, if you think about it, it's building on the last period of AI, which concentrated on before, during, and after the interaction. So before it could be gathering information, maybe getting the name and address. It could be about intelligent routing rather than IVR, so capturing the intent and taking you to the best qualified human or to the best queue, or possibly to an automated route through which your inquiry could be resolved there and then. Then during the interaction, it's about that second pair of ears, listening in, prompting the human agent with what to say from approved knowledge sources, transcribing the call so they don't need to type so much, summarizing it at the end of the call, and then after the call as well, getting into that automated quality management so that the auditor's time, very expensive time, is well utilized in improving the quality, not looking for call recordings. So if we think about what's agentic doing, it's really building up that before part. So the before can become increasingly the whole thing. So more and more use cases can be fully automated and in the word contained, so containment is is the new buzzword, replacing the previous era of deflection. No one wants to be deflected, do they?

SPEAKER_01 6:39

Exactly. Exactly. So then when we talk about AI agents handling CX tasks, I think a lot of people in our audience they have a rough idea of what A agents are supposed to do, but what can they reliably do in today and where do they still struggle?

SPEAKER_00 6:56

Yeah, so AI agents today can resolve increasingly complex cases because we've moved on from perhaps the early era where you'd speak and then that speech would be translated into text, the text would be processed, uh, and then it would might go back into speech again or into a written form. Now the agentic agent is able to actually work with the language just like a human does. We don't write something down and then read it and then decide what to do about it. We actually process it in our brain in the moment, and then you know we're we're able to actually get straight on and get into the resolution phase. So that's really what the agentic agent is all about.

SPEAKER_01 7:44

Yeah. And then how did these agents build on that existing um you know investment into automation that's already happened rather than starting from scratch?

SPEAKER_00 7:54

Yeah, as I said, this is an iterative period. So we're building on the 40 years of uh increasing complexity of the call center into the contact center, into the intelligent automation age, the AI age, and now this. So over that time, we've got a lot of very good uh BAU data. We know exactly what it takes to uh in terms of average handle time, in terms of first contact resolution for lots of different use cases in lots of different verticals. And the best measured of those are the ones that we can best apply agentic AI methodology to, because we know the start point, there we've got the BAU well mapped out, and we know the cost to serve very, very well, and then when we apply uh the AI agent over that very good solid, reliable data, then you know we're able to prove that ROI. And that's where a lot of Gen AI went wrong. We had the MIT report in late 2025 showing that 95% of Gen AI projects have failed mainly because of not demonstrating return on investment. Uh, now you know in the agentic era, we can correct a lot of those early enthusiasm type uh mistakes by a much more methodical approach, working off the best documented cases. One good example I could cite, it's a city council uh in the UK, one of our customers, and we were having a discussion about where to employ agentic resolution first. And firstly, they thought, well, let's put it in our new food waste service. Uh, when we send a letter out, a lot of people pick up the phone uh so we could agentically automate that whole thing. And so we worked out how we would do that. In fact, we demonstrated how we would do that uh using the data that they had available, would have worked very well. But the decision was then made to no, let's not do that, let's use the switchboard because we know how many calls we get per day there, we know the where they route to, we've got the average handle time, we know how many are resolved in that first contact and how many result in a repeat call. So it was a decision to you know go to a something that was less exciting uh than just automating a completely new service. But in terms of being a pilot, it's great because you know we'll know exactly how successful that's been because we've got very good solid measurement, because that same service has been running in our platform as a switchboard for a couple of years now.

SPEAKER_01 10:48

It's interesting because you you mentioned that MIT report, and I think you know, some a lot of people they responded to that by kind of stepping back and saying, well, let's wait and see what's going to happen with you know this transition, especially from Gen A to agentic. But why is that a risk for enterprises?

SPEAKER_00 11:05

Yes, well, people often talk of the risk of action, don't they? Um but I suppose the other ROI is the risk of inaction. So what happens when you don't do these things? So if you don't move forward, A, other people will. So you know everyone works in a competitive environment. So the aim, of course, is to increase your uh customer satisfaction, a better CX, better competitive differentiation. But the beauty of these agentic methodologies is that they also promise uh an improvement in operational efficiency. So every one of those paths that you can fully automate and contain is one where you you potentially won't need to deploy your people on. And those people, those human resources, your most expensive resources, remember it was 97.6% only a couple of years ago of that spend, can be used for those complex, urgent, and emotive inquiries, or for helping your vulnerable users or your high net worth users, or whoever you wish to prioritize. Uh, and then let's uh just drain off the easy stuff into this automated route. Uh, so if you don't do it, uh, you're going to be handling a lot of that traffic that you could avoid handling whilst others in your space aren't doing that. Uh and so you'll have not just potentially customer satisfaction problems, but uh you'll also have a harder press workforce. So where agents, human agents are supported by drawing off a lot of this uh routine work, then you know they are able to function better, be more satisfied, reduces churn, the direct costs of churn, reduces agent training time, because they're not needing to find out where all the tabs go and what what forms to fill in which different cases, because that's all being done for them. So there are myriad reasons to go ahead, but to proceed with caution and only for those use cases that have been fully scoped out.

SPEAKER_01 13:28

Yeah, yeah, exactly. But then even if a company isn't ready to deploy agents today, what should they start planning for?

SPEAKER_00 13:36

Well, as I say, uh failing to prepare is preparing to fail. So uh a lot of the gen AI things that went wrong uh were were were where they prepared, didn't prepare properly. So start scoping out the cases that you might want to automate. Um you're not necessarily going to do them today or even this year or next year, but have a descending order list. And I think once you've gone over all of your high volume use cases, uh you'll soon identify those that have sufficient lack of complexity to be very safe to automate. And you'll also see the ones that you might not want to do first. So I think the goal is to arrive at a hit list. Uh high volume, low complexity, let's start there and work our way kind of up almost like a Maslow pyramid. Um, and you know, let's get the humans working at the top of that pyramid where there's the the most valuable work to be done.

SPEAKER_01 14:49

Yeah, absolutely. And it's that interaction, isn't it, between the humans and the agents. I mean, when you think about CX, it seems to be really leading this adoption of a gentic AI. And what's the reason for that?

SPEAKER_00 15:02

Well, CX, as I say, builds on 40 years of process evolution. So these are ultimately business processes uh that have been honed and evolved since the 1980s when the first call centers were introduced, uh, right through the contact center era, the online era. So it's using the best adapted business processes and then applying automation to those. So really it's about adapting the best and most suited business use cases, uh, the ones that have been best evolved and that have the best data available. I always look to any AI project as data discovery, validation, and then deployment. So if a data isn't right, you're not in a position to start. Um, whereas CX has lots and lots of great data cases that have been scoped out in lots of different verticals for lots of different departments in lots of different organizations. That's absolutely why it's the place to start.

SPEAKER_01 16:12

Yeah, exactly. So then finally, for our audience of leaders, what's the biggest mind ship mindset shift that they need to make to fully leverage agentic AI?

SPEAKER_00 16:23

Yeah, well, I think open minds are required uh to approach agentic, but also a kind of cold realism. So you're not going to be able to automate everything, so don't attempt to. Um, find those use cases that are going to be low risk. Remember, the customer is very fickle, one bad experience, and many of them will walk. So not really the ideal place to carry out risky experiments. And yeah, depending what type of organization you are, you may have other regulatory matters to consider, perhaps medical risks or uh governmental regulatory risks or financial risks. So it's I think looking at a in a risk-based way, but not, I suppose, giving too much sway to uh those slower moving parts of the organization who only see risk. So they themselves have to evolve. So I'm looking at the the office of the CISO, for example. Uh information security, vitally important. No one wants to be on the end of a ransomware attack or or even a DDoS attack. Um, however, some of these, I suppose, regulatory-driven departments tend to catastrophise. We only see the downside risk. So I think uh organizations and and people should be looking at if this goes wrong, then hopefully the the condition we default to is the pre-automation condition and not total service failure. Uh so I think once you've made that risk analysis and you've looked at the best adapted cases with the strongest BAU metrics so that you know what the ROI is going to be, because return on investment is the delta between how it was done before and the new way, then you'll have found your use cases for your first pilots. And of course, work with your technology vendor. So as an end user organization, it you don't need to be some sort of AI experts. There are those exist out there in industry, so make full use of them and seek their advice. They in turn should be horizon scanning and hopefully working with lots and lots of vendors, large and small, who are developing these models. And of course, the role of the CX specialist in technology is to then look at how best to apply those different models for the needs and objectives of the customer in delivering this great CX.

SPEAKER_01 19:12

Yeah, as you mentioned, it's not an all-on-off thing. They can roll back if it doesn't work out how they expect it.

SPEAKER_00 19:18

Yeah, it's digital, but it's not one or zero.

SPEAKER_01 19:22

Yeah, that's a good analogy. So before we wrap up, where should people go to learn more about Content Guru or take a sensible first step?

SPEAKER_00 19:30

Yeah, well, obviously you can go to contentguru.com. You can come to one of our Brain 2050 events. Um, yeah, we have those at intervals throughout the year where we're looking for the latest advances in AI. Uh, and we don't just talk amongst ourselves, you'll also hear from foremost experts in the field because that's the advice we also seek. Um come to contentguiru.com, uh, come and see us in any of our offices around the world. We're very much a people business. You can chat to us. Bring us your business problems. You know, we like to have this consultative approach. Uh, we won't apply a solution unless it is actually going to fix your problem for you.

SPEAKER_01 20:22

That's that's uh very reassuring, I'm sure. And it sounds like a good place to start. So, you know, it's clear that the C the AI agent trainers arrived, and the clear takeaway is that um, you know, companies need to start planning now, focus on CX outcomes and build on what's already in place. Um, and you know, the leaders who act today are going to be the ones who set themselves apart. So thank you, Martin, for your insights.

SPEAKER_00 20:45

Absolute pleasure.

SPEAKER_01 20:47

Thank you. And to keep exploring, um, our audience can find related stories and resources at cxtay.com and also uh join our LinkedIn community. Make sure you subscribe to our newsletter and uh you know watch consume more of our video content as well. So thanks for watching, and I'll see you next time.