What's Up with Tech?

Automation grows, humans stay: the real shape of AI in customer experience

• Evan Kirstel

Share episode

Interested in being a guest? Email us at admin@evankirstel.com

The most natural way to ask for help is to speak—and that’s exactly why voice CX is so hard to get right. We sit down with Babelforce CEO & Co-Founder Pierce Buckley to unpack why the real challenge isn’t the AI model at the front, but the orchestration behind it: identity, contracts, logistics, field service, and payments spanning a dozen-plus systems that must work in lockstep to deliver a short, sweet resolution. From “my fridge smells burnt” to warranty confirmation and pickup scheduling, we map the unseen chain of actions that separates a great call from a frustrating loop.

Pierce shares how a composable, “Lego block” architecture lets enterprises move fast without betting the farm—swapping ASR, TTS, and LLM components as technology improves quarterly, while reusing the same building blocks for new use cases. We tackle the hype around “agentic” bots with a grounded view of production readiness, explain why many voice bots disappoint due to weak integration rather than weak intelligence, and make the case for a bring-your-own-AI strategy that meets enterprises where they already invest. You’ll also hear how Babelforce partners with Zendesk to serve complex mid-market and enterprise contact centers, and why balancing automation with human expertise remains the winning strategy as volumes grow and escalations concentrate.

Security and compliance get the attention they deserve: PCI redaction, PII controls, data residency, and the reality that LLM freedom demands stronger guardrails. We close with the talent gap—modern VUI skills that blend conversation design, AI behavior, and enterprise constraints—and practical steps to build a new class of AI-enabled practitioners who can ship outcomes, not just demos.

If you care about customer experience that actually works—fast routing, smart automation, and respectful escalations—this conversation gives you a blueprint. Subscribe, share with your team, and leave a review with the one CX workflow you most want to fix next.

Inspiring Tech Leaders - The Technology Podcast
Interviews with Tech Leaders and insights on the latest emerging technology trends.

Listen on: Apple Podcasts   Spotify

Support the show

More at https://linktr.ee/EvanKirstel

SPEAKER_01:

Hey everybody, today we're diving into the future of conversational AI and CX with an innovator in this space. Babel Force. Fierce, how are you?

SPEAKER_00:

Hi, Evan. Yeah, great. Thanks for having me.

SPEAKER_01:

Well, really intrigued by the work you're doing. I'd figure it out, uh, how it fits in the wider CX landscape and where it's headed. Uh, what was your aha moment behind joining uh Babel Force? What was the big idea at the beginning?

SPEAKER_00:

Uh yeah, actually I co-founded Babel Force 2013 uh with uh with Christian and Timo. And uh yeah, I have actually worked in CX in enterprise for enterprise companies for 27 years. So already a long time since since about 1999. And um yeah, I I mean I basically have uh visible burn marks, Evan, at that time from doing huge enterprise projects like how do you deploy a contact center, how do you make it integrate with everything else? How do you how do you make voice inbound work with all the other channels, all that kind of stuff I had done for years? Um and what I had with like the kind of aha moment was you know, in a few years, this is 2013 now, we could see in a few years with everything being virtualized and on the cloud, we'll be able to deploy these$50 million projects for$5 million, you know, or and the ones that were five million are now half a million, and so you know, and and everything will happen in weeks and hours, you know, hours, days, weeks instead of months, um, which of course is what has happened. And that's basically what we do. Um, so we help people deply all of their inbound channels, like adding we we mainly focus on voice, Evan. So voice is a very hard channel, but it's a really important one, right? It's the most natural channel in a way. And now, of course, voice is getting turbocharged with uh with the with with AI advances. Um and that's you know, that's that's that's fantastic for what we do, but that's that's the area we focus on, is making voice work really, really well. And now, of course, with AI and all the advances, it's uh it's even even more exciting space.

SPEAKER_01:

It is. I started 35 years ago with voicemail technology and IVR, which was like the big thing. And it's amazing uh how long it's taken to get to where we are, actually, uh, with automated voice interactions and blending AI, uh gentic-driven automation. What are some of the biggest hurdles or challenges that you face when deploying your technology?

SPEAKER_00:

Um well, actually, the main ones are not around AI, you know, which is often surprising to people, right? So um, because the AI advances relatively quickly. I mean, essentially, you can probably wait two quarters now if you're unhappy with the way the model works at the moment, and it's probably going to be solved, right? Or maybe three quarters, but you know, very, very, not very long. The problem is in orchestration and integrating all the stuff. Like a, I mean, you will know it, Evan. We could probably exchange a few war stories if you you were 35 years ago active with IVR, but in the CX space, like customer, customer experience, customer service, sales, engagement, everything that every kind of customer interaction that happens at scale, or you're doing millions of them or hundreds of thousands of them a month, um, you often have 20 systems at play. And I'm not, and it's no exaggeration. The average number of systems integrated in our solutions is 12. That's the average, right? Um, and that doesn't mean that these companies are really weird and you know, it's it's just because you've got CRM on the one end and logistics systems on the other, and payments in between and ERP, and you know, the list goes on and on. Um, and so getting a voice interaction, like getting a short, sweet dialogue, which is really what the user wants, right? The user just has a problem, they want it solved, they want it solved without friction. So short and sweet is really what they're after. And getting that to work actually takes a huge amount of data orchestration, right? Because what the user thinks of as a simple query, like I, you know, I just my fridge smells burnt, and I want to figure out if it's still under warranty and who the hell will pick it up, right? That's basically what they want to have in their mind. But can you imagine, you know, yeah, like it might not be obvious to all listeners here, but if you've worked in space, you know, oh my God, like that's a warranty, you know, there's a contract system. The contract may not even be in the hands of the agent who could be sitting in the Philippines or in India or in or in Texas, right? Um how do you find out what whether the person bought the product and what where they bought it and does the warranty still is it still valid? Is is there a field service system, you know, that needs to be notified and then somebody's going to pick it up? Like the chain that happens behind that really simple, I just want the I just want my fridge to work, you know, is pretty big and it's mainly data. It's not, it's mainly data and orchestration across different systems, not necessarily the AI at the front.

SPEAKER_01:

Got it. And you position your platform as a composable contact center, sort of what plus voice bot AI. Uh maybe talk about the architecture, the design philosophy behind Babelforce.

SPEAKER_00:

Yeah, and I guess from that aha moment, the the thing we realized that would really move the dial on how to how to be able to manage this complexity is Lego blocks. And you just mentioned composable. So um if you're going to need to design a concept and then implement that concept uh using any kind of complicated technology or you know, rare skill sets, and and and I know it sounds seems like there's a lot of developers in the world, but actually, right, there's not enough, and there never will be, it seems like. Um so if you're going to rely on that, it's gonna be difficult. So what we essentially do is try to turn enterprise needs into Lego blocks that can be put together. Um, not not the little, not the big Lego blocks for the toddlers, you know, the duple Lego blocks, but the complex ones. Let's say closer to Lego Technic, where you know you can also put a robot arm uh into it or a robot or a robot uh engine or whatever. And and what we try to do is make everything composable because no enterprise is like another, right? They they will have grown through mergers and acquisitions, they have probably different variants of different systems and different business units. Um that come so you never find a greenfield site in a middle-sized company or an enterprise organization. Um, so what we are saying is, okay, you come along with your brown field site with all of its complexity, and then our job is to provide Lego blocks at all kinds of levels of abstraction that allow you to deal with that complexity and still and still be able to very quickly deploy a thing like a voice bot at the front end, or very quickly deploy some new routing engine that's going to make your contact center way more effective. That's so that that that's essentially essentially our approach to it.

SPEAKER_01:

Fantastic. And there's this balance or tension between you know, sort of human first CX and automation first, and almost every nuance in between. How do you see that balance today and how's it how's it shifting over time?

SPEAKER_00:

Um so for us, we're pretty much agnostic about that actually. Because uh like even if you look at our revenue streams, we're about 50-50 between licenses that are purely for automation and the those that are user seats. Um and there's no difference in in the way we uh in the in the way we count licenses. So if an organization started with a thousand agents and they end up with 500 agents and you know, and several hundred licenses for automation, it's all the same to us. It doesn't make a difference. It's the same Lego blocks, it's the same components, the same licensing. So actually, that's a really nice thing, good question, Evan, because we from our point of view, we can kind of act as an independent broker in that sense. Because if you come onto our platform and you start an automation and then you realize, oh, I need an escalation channel that's that just needs humans, then yeah, that's no problem. Then you just start deploying human licenses or the other way around, which is of course more common nowadays, that you start with human and you're moving over to automation. Although both are occurring, interestingly, as you deply AI, um the escalation need is often increasing because the number of interactions you're doing and you're tending to push the more complex things to humans, and there tend to be just a lot of complex things left to do. So at the moment, at the moment, it's kind of still balancing out quite a lot where human agents are really, really important. Although, of course, the growth in completely automated bot-handled interactions is just, you know, is incredible, right? I mean, the growth rate month over month or year over year is astonishing. Um, but both are still completely necessary. That's what we're seeing. And people are getting more nuanced, I think, and more sophisticated about thinking about it and starting to think of use cases that are for automation and use cases that are for humans because the brand loyalty matters or because a human happens to be more effective at that particular touch point. So I I'm we're seeing much more sophistication now than um just after the launch of ChatGPT or the launch of the first AgenTech uh agents.

SPEAKER_01:

Fantastic. And there's so much in the press now about CX and AI in general, of course, but lots of mistrocessions and um and misunderstand uh standing around conversational AI and CX, driven by some very high-profile headlines that you see in like the Wall Street Journal every now and then or other tech uh press. Anything there you'd like to debunk about conversational AI and CX? What what what's being uh misunderstood about what's really happening on the ground?

SPEAKER_00:

Yeah, well, I think um I mean the hype the hype cycle is always ahead of what's happening on the ground, right? So like agentic was fairly uncommon a year and a year, a year and a half ago, a year ago even, and now it's in, you know, kind of everywhere. Yet the number of real agentic, and by real agentic I mean, you know, where the where the bot has agency and is deciding for itself to some extent, right, on what's going on, the ext the number of production systems that are really uh autonomous in some sense, sorry, in any real sense, is relatively small, actually. Um yet it's everywhere from a hype cycle point of view. So I I I think I think the I think the market in terms of adoption is very different from what you see from a trend point of view. Uh I'll give you an example. So a lot of companies wish to deply uh voice bots and they've picked maybe five use cases that they want to focus on. Often what they realize is two of those, no problem. They can actually do uh something close to an agentic bot for that and do it very quickly. Um and in the next three use cases they go, oh, we never, you know, we we haven't invested in the underlying infrastructure, so we don't have the right APIs, the data's not clean, it comes from different sources, it's different in one case and another. And the bot, although it would be deployable, would be terrible, not because of the AI, but because of the data. And so I think when the rubber hits the road, I think there's a lot of work to do in the background orchestration and preparation. And I guess that's the space we're in. Like we're saying, okay, we we are not trying to shy away from your complexity. We get quite excited about complexity, Evan, basically. When a customer comes to us and they go, you know, I don't know if you can solve this because we hit a rope bit of a roadblock on the data and our APIs and everything, we're like, oh, well, we're rubbing our hands together, you know, like this sounds like a job for us, uh, because that's what we're trying to basically prepare an organization step by step to manage this complexity and to manage the level of change that's going to be hitting them. Like if if LLMs or if the underlying models for speech speech models, for example, if they're getting better every six months or every few months, then you need to plan for change, essentially. Like you change has to become the new normal, um, which means you've got to do things in the right order and prepare yourself for that. And that's I think that's a I think that's the biggest I think it's the biggest area where I I think we, and I think in general, the technical people and the vendors need to help organizations because I think people are really struggling to get to grips with that.

SPEAKER_01:

Fantastic insight. And speaking of helping people, you've aligned up with Zendesk and integrated very tightly with them and helping customers uh talk about that partnership and what what it means exactly.

SPEAKER_00:

Yeah, so I've worked with and around Zendesk uh for 14 years now. Uh 11 years ago, I think we were one of the first companies in the world to build a contact center integration, what used to be called CTI, computer telephony integration. We were the first, just after Zendesk launched their very, very, very first version of that, which of course is very different from what is now depleted, of course. So um, and we've done a lot of we've done a lot of projects with Zendesk over those years. Um so we uh we work very closely with Zendesk on deplying uh mid-market, so middle-sized and enterprise contact centers. Um and yeah, that's a fantastic collaboration. I mean, we we Zendesk is the market leader in from it when it comes to ticketing and CX, uh CX digital channels. Um, they also have invested more and more in voice and are getting more and more known as a contact center product, although in general they're not really thought of as contact center or or CKS, but that's becoming more common. Um, yeah, and we're really happy to be one of the main players in there in terms of supplying the uh the complex contact center needs. Fantastic.

SPEAKER_01:

Uh let's talk about the market landscape. Um, the latest CX uh landscape uh had thousands of AI voice bot companies. It's it's almost beyond imagining uh how many there are popping up everywhere. Where do you see yourself on the landscape? And um there's a lot of MA and acquisitions happening, um which is a sign of how vibrant the space is. Where do you think things headed?

SPEAKER_00:

Well, I mean, uh so we are we are more of a a uh of a CCAS player than a voice bot player, first of all. And also, I think maybe I didn't I should explain just to as background to answer your question, Evan. We also are agnostic about the AI that's used. Um as we are in the CCAS space in general, about the integration. So I I think we have like, I don't know, if you look in our platform, just um maybe 80, 90 integrations. And you would know you would know most of those, right? And all the big players in CRM, ticketing, but also players that uh you know that do some other important function, uh completely different products, everything from identity management um to products that help you with data qual uh cleaning, for example. Um so you'll find all those on the platform, and we and that's just for CKS. In the space of AI, we have, I think we've uh over 600 AI voices available, which are across, which are across three major providers, and that number of providers is also growing, you know, like there's some of the every year, every half year, there's more added. Um and the same is the case for speech to text, so uh right, or what used with speech transcription, um, and also LLMs for dialogue management. So you have well the way we think about it is in sort of a if we break down what people really need, what enterprises need, there is not an enterprise in the world that is not doing multiple AI projects. So in fact, I went into an insurance company just a few weeks ago, Evan, just or it's a few months ago now, um, and they had they they told me they had 78 AI projects active. They had counted in. So somebody said, somebody said, probably the CFO's office, right? said, hey, we need to do a bit of an audit here. Like this is crazy. 78 projects. That's probably okay. It's a very big global insurance company. But, anyways, that's what's happening inside an enterprise. That means that a lot of enterprises are getting savvy and realizing, hey, we need a reuse strategy here. Um, so and I I talk about this explicitly because it comes back then to answering your question. When we go into that organization, we don't ask them to buy another AI from Babel Force. We say, we we say, just like every other component you have, you have your complex situation, what AI are you using? And if they if they tell us we're already building a brilliant um internal AI, maybe it's on OpenAI's platform, maybe it's Lama, whatever, it's going to be hosted here, maybe it's inside Microsoft Azure, maybe it's directly from OpenAI, whatever. It might be a mixture, usually of LLMs as well. And if they're working on that, maybe with another vendor or with their internal organization, and if it's for email and for other text channels, it's very likely going to be a foundational model for the voice channel. So we'll say, okay, look, nope, let's reuse that. Here's how you reuse it. That's our bring your own AI. And of course, for smaller companies, or which is at the lower end of the mid-market from our perspective, they will often need the models ready as Lego blocks. So we provide that as well. So you can go whichever, you can start from whichever part of that spectrum that you want, you know, and you can evolve as you go. So you could be even a mixer using some bring your own AI and some ready-made components. And we and again, we're agnostic about it. So I don't even regard us as one of the, what did you say, Evan? Thousands or thousands of thousands, yeah. Yeah, I mean, we we spend tens and tens of millions on our composability for enterprise. Um, there's other people spending tens of millions and hundreds of millions, and as we know, even billions, right, on the uh on the AI. So, you know, we're not building, we're not building another AI. And also really another answer to it, uh, you might remember about 2012 or 2013, I think there was the first wave of bots, of chatbots. And I think somebody, a VC, uh, a venture capital, um famous venture capital guy had published on LinkedIn that he that they had done an audit and they had found 2,000-something bot providers. Uh and you asked about consolidation. A few years later, there were a lot less, a few hundred maybe, and then there were 50. Um, so I think we'll probably see the same.

SPEAKER_01:

Mostly pretty bad. That was not a uh a successful experiment in those days, I think.

SPEAKER_00:

Mostly pretty bad. And you know, that's one you did ask me to debunk earlier or maybe talk about what's really happening. Actually, to be honest, a lot of bad bots are being deployed now, too.

SPEAKER_01:

Um that's a great insight.

SPEAKER_00:

Especially because of data orchestration. So so for example, I often say to people when they say, Oh, they tried out a they they got a bot that's from a diff from a vendor or and you know, they don't know is it good or bad, what's the benchmark on it. Um, or if if a vendor tells me about the way their bot works and I realize they don't have an orchestration layer, then I say, Well, send me 10 telephone numbers where the bot is deployed, like really productive, right? And then let's call them up and then see what it's like. And you know, like it's not that great, you know, to be honest. It's like um, so I I think I I think we're seeing another wave of some, but this time much better. Like the ones that are good this time are much better than the last time. But I I mean I would there's a reason why the market is become why the buyer is becoming a little bit more um a little bit more weary, right? There there is a reason because there are a lot of AI projects that are not achieving the goals that they are that they're looking for. Um and I think there's way too many, uh Evan, way, way too many already of the uh of voice bot and chatbot providers. Like it's it's it's not going to go well for a lot of people. But then again, that's what we expect, right? It's creative destruction is often what is necessary for the next wave of innovation.

SPEAKER_01:

So, that's so well said. Uh, there's also a lot of uh tension and concern around innovation versus data privacy and data protection, uh, security, uh their data breaches every day. How do you balance that tension with getting uh innovative new services into production while staying uh compliant and true to customers' desires there?

SPEAKER_00:

Yeah, I think I mean I think the the the obviously with LLMs, um, and I've listened to a few of your of your um interviewees, your your guests here on the that who are more in the security area, and like obviously these kinds of bots are more susceptible to attack than other systems that we've had and need new kinds of protections. Um on the that's one thing. Then guardrailing is quite difficult. The guard, you know, like if we're really honest with guardrailing is really difficult because the bot can do anything. So the like ultimately the code or AI or mixture of that's going to try and protect it from you know from damage is going to be more complicated than it needed to be as well. Like you can't increase the complexity and the freedom of movement, let's say, of the one and not up the complexity of the other. Um, so I think it's a hard area. Luckily, some of your great guests and some other people in the market are actually looking at that enterprise-wide. And that's that's the kind of expertise that we actually uh also recommend to our customers, right? Like is that you you don't go about now protecting your uh your company from the potential downsides of AI abuse by just looking channel by channel. You know, now there are channel differences, but if you're going to protect your email, uh, for example, PCI compliance, for example, with the right, right? Redaction and cleaning and ensuring that a credit card number doesn't end up being put in ChatGPT, it's the same problem as whether it ends up being put into a ticket in Zendesk or into a field in an object in Salesforce, right? It's not different in from a you know from a pure component point of view or a processing point of view. So we just we we actually it's a big emphasis of our projects to make sure that that kind of thing is done. And of course, we're working in all these complex industries like utilities and banking and and and so on. So we get exposed to all every single possible variant of uh what can go wrong and what needs to be protected against. And that's a big yeah, it's a big topic for us.

SPEAKER_01:

I bet. Um so what's next for your next phase of growth? Um and what's the biggest obstacle obstacle that you're facing to get there?

SPEAKER_00:

Um uh actually for us, I think AI expertise. I mean, when I say us, I mean our customers, actually, right? Because our platform is provided and we're not all we're we're rarely the ones doing the work. Because if you take the Lego blocks, they have to be put together, and then we have a whole load of partners in the world to put them together. But I think the number of people who have deep AI expertise, and then if we look at the sub-niche within it, who have deep AI and usability expertise combined for voice, is relatively small. And you might remember then, Evan, from years ago in A in IVR, it was also very constrained back in the day. It used to be called VUI, right? Vice user user interface design. Um, and there weren't enough then, and there's definitely not enough now. Um, so I think that's an obstacle for both the buying organization that would be buying our product, like the people we serve, the customers we serve, and it's an obstacle, of course, for um for for us and our other delivery partners as well. So I think a lot of training, a lot of engagement with the community is our solution to that. It's like, you know, right, trying to build, trying to build a a new type of citizen developer or a new a new type of AI-enabled expert who's able to guide customers in this. Because I think there's a lot of guidance necessary, right? The complexity, I I see while while performance is increasing of most of the models, um complexity is not going down. I think getting it applied in the enterprise means that actually complexity is going up currently. I mean, there might be later in the wave, there'll be more comfort functions and right and so on, but it will take a few years before this stuff climbs the engineer's ladder um and gets all its usability tweaks, you know, where where it where it becomes uh where the upholstery is better, shall we say. Um, that's gonna take a while. So in the meantime, we're gonna need we're gonna need to really engage and train a lot of people, um, I think.

SPEAKER_01:

Well, so much work to be done. Uh, congratulations on all the success, onwards and upwards.

SPEAKER_00:

Thank you, Evan.

SPEAKER_01:

Or appreciate your time and uh this amazing journey. And thanks everyone for listening, watching, sharing the episode. And be sure to check out our TV show on Fox Business and Bloomberg TV at techimpact.tv. Thanks, everyone. Thanks, Pierce.

SPEAKER_00:

Thank you, Evan. Lovely being here.

SPEAKER_01:

Take care.