Webinar. Beyond the Hype: GenAI in Customer Experience in 2026

In this Hello Customer webinar, CEO Bram De Vos is joined by CTO Jonas Beullens for a blunt, practical conversation about AI in customer experience, and why the loudest takes are usually the least useful.

They open with a deliberately provocative claim: AI is threatening your job. Then they quickly pop the balloon. Predictions around “AGI is near” and “whole professions will disappear” have been confidently wrong before, and the bigger point is simple: in the next three to four years, CX is far more likely to be reshaped than erased. The work shifts, but the human layer still matters because customers still prefer humans for many moments that count.

From there, Bram reframes the hype cycle: what looks like a sudden revolution is often a long evolution. LLMs feel new because they became visible to everyone at once, but the underlying trajectory has been building for years. And the people shouting “this changes everything” often have something to sell, or at least attention to win.

Then they tackle the opposite complaint: AI doesn’t know our craft. The answer is “it can, but only if you feed it properly.” Jonas breaks down how language models work, and why outcomes depend heavily on inputs. A simple story makes it stick: a friend failed a golf rules exam using ChatGPT, then passed after uploading the rulebook. Same tool, different context, completely different results.

They bring it back to CX execution with concrete examples from Hello Customer: evolving from basic categorization and sentiment to hyper personalized taxonomies, adding implicit feedback from chats, emails, and call transcripts, and using Ask Isaac to turn questions into consistent, data backed reports. A key warning shows up here too: plain ChatGPT can give confident but inconsistent answers to the same question on the same dataset, which is exactly why prompt enrichment, retrieval, and traceability matter.

On trust, they land on two pragmatic guardrails: verify sources (like Perplexity does, and like Hello Customer does by linking back to knowledge base articles or verbatims), and treat privacy as real risk, not an afterthought. They call out data leakage and model training concerns, and explain why keeping models and processing inside the EU and preventing data reuse changes the equation.

They close on the 2026 direction: the real leap is from insights to action. Automate the low risk work, keep a human in the loop for high risk moments, and aim for fewer dashboards and more guided next steps. The final advice for CX people is refreshingly unglamorous: get your hands dirty, try one small use case each week, share what works, and do not give up.

Learn how Hello Customer turns feedback into business improvements: https://www.hellocustomer.com

You can watch the full webinar and see the discussion unfold on video here: https://www.hellocustomer.com/en/resources/webinars

Transcript

SPEAKER_00 0:31

Welcome once again to all of you. Thank you for registering in such big numbers. Truly appreciate it. The webinar today is about a topic that is pretty controversial. It's the topic what happens with AI in CX? What happens with CX in AI? It's a topic that is hot. And I'm very happy to be able to talk about that with an expert. Because on behalf of hello customers, I can I can very well play the CEO, but I have experts on my side who actually know what they're talking about. And in this case, I welcome Jonas Burns, our CTO. Welcome, Jonas.

SPEAKER_01 1:07

Thank you, Ram. Thank you for having me. We have indeed a very excited and very trending topic because we got a large audience, so uh I'm very excited to get this thing going.

SPEAKER_00 1:16

All right, let's get the show on the road, as they say. Now, the way that we wanted to tackle this, Jonas, was you know, it's a broad topic. We thought we would take a couple of bold statements that we come across in our line of business and just put them forward and take them as starting points, take them as debating points, take them as conversation starters if you like. And so that's the way we're going to play this. So without further ado, bring on the first statement. AI is threatening your job, lady or gentleman listening. AI is threatening all jobs in CX in particular, from the support staff to the CX manager. It's of course something that's been running around in our industry for a number of years, even now. Like, what is going to happen? Is AI going to eat up jobs in CX? Now, if I were to put that just plainly and boldly, without any UNS to you, what would you say to that, Jonas?

SPEAKER_01 2:06

Well, indeed, it's a very bold statement, directly to begin with. And I think it's fueled by provocative statements. This is an example by Salton Altman, by Elon Musk very often as well, claiming that we're very near, we're very close to AGI, AI at a human level. And it's these kinds of statements that fuel these thoughts of maybe we're gonna be all made redundant.

SPEAKER_00 2:30

At the same time, it's fascinating and wonderful that humankind would be capable of creating something like AGI. And at the same time, it's like doomsday. It's a policy, right?

SPEAKER_01 2:38

It's provocative, right? I don't mind a provocative statement, but maybe we should tell our audience that when from now on during this webinar, when we'll talk about the future. We're talking about the foreseeable future, the next three to four years, let's say, for the simple reason that we don't know how it's gonna evolve and what the world's gonna look like right after. Fair enough. We don't know it, but we're in good company, but as we don't know, this the man on the article is Jeffrey Hinton. Jeffrey Hinton is considered one of the absolute godfathers of AI. He actually won a Nobel Prize for his work back in 2024. Or only a year ago. In 2016, we predicted that in five years' time, so by 2021, that there would not be any radiologists anymore. Because computer vision was accelerating at such a fast pace that they would be made redundant. Well, we're the end of 2025, almost 10 years later, and apparently there's more radiologists than there ever were.

SPEAKER_00 3:36

But you're also implying that perhaps if we get it wrong enough, then we might win a Nobel Prize. Seriously, 10 years ago, that's not that long. And this guy was so wrong, that means that you know a lot of the predictions that are being made, a lot of the hype that's happening is also wrong. So, with that disclaimer firmly in place, maybe you can illustrate what that could mean for business.

SPEAKER_01 3:58

Yeah. Well, the thinking was it's going to threaten our jobs. Yeah, exactly. And if I may get into that statement directly, I don't see that happening on the short term. The impact, the radical impact of AI making us redundant, of this technology getting to that level, we don't have extra value as humans. It's simply not happening at this point in time. The technology is simply not working today. And there's plenty of examples for that. 2025 was going to be, and that was a widely spread idea, the year of AI agents. Absolutely. Agents doing autonomous work and actually replacing us. And that was also what they thought, what they predicted at Clarma back in 2025. When they fired 700 people from their staff, claiming that in 2025, they would all get replaced by AI by agents. Well, as a matter of fact, during 2025, they didn't get replaced by agents, they actually started rehiring from these very same roles that they fired in.

SPEAKER_00 5:15

Do you think that that is essentially the reason for what you claim, namely the statement doesn't hold true? CX, the jobs and CX will not be threatened all of a sudden in the foreseeable future by AI. Is that because there is a human layer underneath that that cannot yet be replaced?

SPEAKER_01 5:32

Yeah, that's definitely part of the equation. There's the fact that the technology isn't there yet, isn't reliable enough to completely replace humans. There's the other part, which is put straightforward by Kleiner as well. Customers like talking to people. And in the light of customer experience, that's a very important thing to remember.

SPEAKER_00 5:48

Yeah, it's something that's also been buzzing around for a long time. And I think one of the people who's an advocate for this is Stephen von Belegham, who's already put that forward many times, one of the other world-leading experts in the field, saying that human empathy, enthusiasm, passion is so important because on the other side, there's also a human being, whether in business or without a business, it's a human being. So there is something to be said about that. On the other hand, it seems to be going very fast. I mean, we're being bombarded with new releases of new LLMs every day. I wouldn't say every day, but every week, every couple of weeks. There's, you know, Gemini 3 last week, I think. And it seems to be going so fast. Doesn't that mean that we are heading towards some kind of big, big new breakthrough again?

SPEAKER_01 6:35

Well, there's plenty of other examples of the statement that we've just made. For instance, the one we see here on the picture is Jan Le Kun. Jan Le Kun is also considered one of the godfathers of AI. Actually, until a week ago, he was also running the AI lab of Meta, they're the company behind Facebook, Instagram, and so on. And he was presenting about a month ago at a university in front of a lot of students. And he's told the crowd, he told the audience that all of the students in the audience who were now working on large language models, which is the technology behind the AI that we see today, and that they simply stop wasting their time.

SPEAKER_00 7:12

That's not talking about the bold statement. Yeah, you have one there because the whole planet is currently revolving around LLMs. It's LLMs all over the place. And then this guy says, well, it's it's on the screen there. It's quite, you know, for a man of his stature to say this, yeah. Yeah. Do you agree with him?

SPEAKER_01 7:31

Well, he did not imply that he thought it was impossible for us to one day succeed in building uh AGI, but that it was simply not going to be with the technology as we know it today and with the incremental improvements on that technology.

SPEAKER_00 7:45

You call the improvements incremental. I mean, that kind of strikes me. Like when you're on social media and you hear the comments of the influencers in the field, it seems like every new release is, you know, this is it, now finally everything's changing. And you call the changes, the iterations incremental.

SPEAKER_01 8:01

Yeah, I think that's one of the main messages that I want to convey. I noticed that for people for whose AI started three years ago, that it all seems like it was a very radical revolution. And of course, there's been made progress, there was significant progress, but it doesn't come out of nowhere. I really would like to advocate that there's somehow kind of a continuum over the past decades where the same technology has been evolving, where we, as humans, as users of the internet, we definitely got in touch with it without maybe realizing it. We already were using it 10, 15 years ago on the internet when we got product recommendations, chatbots we were working with back then as well. But of course, there's moments that there's significant breakthrough. OpenAI took the very successful gamble, along with some other companies, to scale existing technology to huge numbers, huge investments. They saw that when they scaled these existing algorithms in S.

SPEAKER_00 8:55

These existing algorithms, you stressed that.

SPEAKER_01 9:08

Before, and maybe one or two breakthroughs, before we would reach that level of human AI that many people are now claiming to be very close to.

SPEAKER_00 9:17

So if we go back to the state one more time, jobs and CX will probably not be threatened as much as some would like us to believe. And the reason for that is that we are not yet capable with the current technology, and especially everybody's talking about working within these LLMs, that will not enable us to take that quantum leap forward to make it really human. And the point that Jan de Kuhn, hey, you're in good company, you and Jan Lakun say the same thing, you want us to meet. That's that's not that's not bad to start with. But the thing is, the thing is that you call these changes incremental as you go through time. And that's what when I look at this slide, for me, it is an important slide or a remarkable slide, because I too, I admit, we've been working in Hell Customer with AI for 10 years. But it is true that on a massive scale, on a worldwide scale, it's about three years ago, that AI suddenly dams burst. It was all over the place. When you look at this slide, right at the top there, you see these golden ages alternating with dark ages, golden age, dark age. And now we are in a golden age again, fair enough. But there's no reason to believe that there won't be a dark age just around the corner again. And I think that really is something, if you zoom out and you look at history from a little bit of a distance, you get out of the yeah of the rabbit hole that sometimes we're all in, you see that. And for me, having discussed this with you in the past weeks, Jonas, something that struck me as well was that a lot of the great advocates of the new releases, the new iterations, changes that you call incremental are people who have a commercial interest in the talks. I mean, if you are from one of those models, obviously you're gonna hype it. But then there's also all of these social media influences who want the eyeballs, our eyeballs. They want our attention. It's an attention industry. So, of course, they're gonna hype the next new thing because they want you to listen to them. And so I guess that subliminally speaking, the message I'm trying to convey here, one of the things I would like to say to the audience is be wary, be careful of who you listen to. If the person has got something to sell, even if that person is just trying to grab your attention, yeah, put a bit of nuance on what they are saying. Notwithstanding that, it's impressive what LLMs do.

SPEAKER_01 11:34

But we started the talk with a disclaimer saying that it's actually impossible to know what's going to happen in three to four years' time. That's true for most of the experts out there, probably. So those making bold statements, there might be a reason why they're making these bold statements.

SPEAKER_00 11:50

But in any case, and that's probably not so bold, but your point of view, if I may summarize, is that what we're looking at is not a revolution, but an evolution. Rather, and what I would like to do, break that down and come back to whole base, if you like, to hello customer, and to make an attempt to translate what you've just said into concrete examples. Because what we are doing, we are in the CX business, and you know, for those people in the audience who wouldn't know, we are a software as a service. Hello customer is a software as a service, a voice of the customer platform. So, in other words, what we do is we help organizations to understand what the experiences that their customers go through, or listen to the voice of the customer, hence the term of the industry, help them to understand what a customer thinks and feels when he's interacting with the organization. So we do that. And if I may, Jonas, I'm gonna give you the floor in a second to tell you about how we're evolving and how that then implicates CX people using us or any other platform. But in order to lay the field out, what we do, four steps. We gather and centralized feedback. So we do surveys, customer surveys, we welcome surveys done by other parties, third parties, we gather public reviews more and more increasingly. We are now also processing feedback that is what we call implicit feedback, feedback hidden within chat conversations, email conversations, transcriptions of conversations of all sorts. We then run that through our AI engine called Isaac, that we started building 10 years ago now, that has constantly been improved. From that come insights in many shapes, sizes, and forms. I won't go into that, it's it's on the slide. And we also offer the capabilities to CX people to close the loop with a respondent who's given feedback. That essentially is what we have been doing or what we are doing with our platform. Now, if you look at that, this is the basis. How is that evolving as you talk about not evolution but revolution? How is that evolving by grace of AI?

SPEAKER_01 13:56

Well, it's important to first stress that our platform at the heart, that the essence, it's always been an AI platform. The company has been around for a decade. We celebrated 10 years a few weeks ago, and it's always been the essence the heart of what we're doing. Now, what we did with AI, we classified feedback. So we give every piece of feedback that comes in, we give it a category and we give it a sentiment. Now, with the use of new artificial intelligence and both large language models, where before we're using language models, it actually shows a bit on the continuum already. And we were always capable of saying, okay, so now you know 14% of your people are talking about online and digital, 5% about price, and so on. That's what we did before large language models. Now, today, with the power of large language models, we're capable of making such a taxonomy tree, such a categorization that is really hyper-personalized for every organization, for every client, and making this kind of an analysis much more powerful, much more relevant for every customer specifically. And on the other hand, we're also capable now of not only analyzing feedback surveys, but as Bram mentioned, to also analyze what we call implicit or indirect feedback, feedback hidden in voice calls, feedback had in hidden in chat messages, feedback hidden in the case. Volumes are fast.

SPEAKER_00 15:18

Volumes are gigantic.

SPEAKER_01 15:19

And together bigger volumes, a much more personalized category tree, it makes the insights so much more relevant, so much more personalized to your use case. That's one thing. Another thing that I like to show is what we call Ask Isaac. Ask Isaac is the chat bot that really automates data analysis for you. So if there you have a particular question, you don't need to crunch the numbers anymore or look in the dashboards, but you can ask a question. One of my favorite questions actually is one like here to ask immediately for a direct report. Behind the scenes, the chatbot is actually querying all of the data that we have, all of the KPIs and so on, taking all of the relevant data, and it's creating such a written report with numbers all very personal to your organization, to your feedback, using all of the categories and classifications that we've seen earlier as well. Now, this is a written report, Bro. I know that you think that it looks a bit boring. That's what I've been saying. Yes. So we use the representation of this, and now thanks to AI, we can just within a click go from the text that we see there that we got via the simple question to a presentation that looks like this. It's the same contents, but it's formatted as a report.

SPEAKER_00 16:29

Hell of a lot less boring. Yeah. For me, having been in this business for 15 years now and in that whole CX space, this kind of creating this is a pain, right? Not just the number crunching and checking and checking again, but also going from that raw data that you have then crunched into a nice looking presentation, it's a pain. It costs you so much time. That's the kind of heavy lifting I would rather not do. I would rather look at the results and see what I can then do with it. And AI is enabling us much faster to do that. So, in that respect, going back once more to that initial question, will AI eat up the jobs in CX? You would make the whole thing more relevant. You would make it more accessible, you would make it more concrete, more tangible. So, in that respect, with what we have been going through, the evolution that we've gone through embracing the new models, it's made me very enthusiastic, especially because I think we can do so much more with CX and therefore for CX people. So that's definitely a good thing. If we wrap this up, can I say that all things considered we shouldn't be scared? We won't be out of the job.

SPEAKER_01 17:34

Well, we claimed, we stated that there's a continuum, that there's an evolution, but it's a fast-paced evolution. It's a fast-paced evolution that allows us to do more with less effort. Yes. So if you can do more with less effort, then you don't need the same amount of people to do the same thing. I think this graph, this graph is very well explaining this. What it shows is many of the big companies, and of course, these are special cases, the online digital companies, over the past years. How many employees did they require in order to get it to 100 million euros annual turnover recurring revenue? Well, for LinkedIn Airbnb, not that long ago, it took them about 1,000 employees. In the past few years, we saw a few companies getting to 100 million euros in annual recurring revenue with 100 or even less than 50 people.

SPEAKER_00 18:26

So that does that does show that you know there was new ones in there. Yeah, let's continue with our next statement. The second statement. If you can bring that up, does everybody see this? AI doesn't know our craft. Second statement, bold statement again, polar opposite of the first one, but one that we have come across quite often.

SPEAKER_01 18:46

Absolutely.

SPEAKER_00 18:47

As in, people have jumped onto the bandwagon of AI three years ago with great enthusiasm. Companies, and they have tried it, and they felt that, well, it's not really doing the trick. When we really wanted to do something for us, AI, it doesn't work. This is something we've heard. And what we hear is that companies, people say, AI doesn't understand our industry truly and really. It doesn't understand the customs that we used to, it doesn't understand the jargon that we used to. Therefore, it may be good for some personal use, fine, but to really make a difference, really move the needle in our business, in our industry, it's not doing the jig.

SPEAKER_01 19:25

It's so remarkable that it indeed, on the one hand, we have the those who claim that AGI is very near, and we we've discussed this, we see that in reality. On the other hand, we also have those who tried out AI, who saw it really wasn't working for their use case, and then they just abandon it altogether. Now, in reality, the truth is somewhere in the middle of those two ends. But what we do know is that you and I, and everyone in the audience, we have an impact, we can influence how well these models and how accurate these models are. How can we do that? In order to do that, let me very briefly try to explain from a high level how these models actually work, these language models. Go ahead. They are basically black boxes, the models themselves. Now, and that's a simplification, of course, but believe me, for many people working on these actual models, they're also, to a high extent, black boxes filled with mathematical functions that are a bit random at the beginning. So you have these models. Now you give these models inputs. The input that you give to these models is text and is normally a question, the question we ask to chat GPT. Based on that input, the model is predicting the next piece of text. You'll say a word. So it's predicting the first words that will come that will answer your question. Then it will take your original question and that word, give that back as input to the model, and predict the second word of your answer. And it will do that all over and over again.

SPEAKER_00 20:56

But I thought it predicted words on the basis of the words that it had just predicted. What you're saying is it just it doesn't just do that, it actually ingests the original question, the original prompt again.

SPEAKER_01 21:09

Every time and again it takes the original text that you've given it and the words that it already predicted, and it iterates and predicts word by word. Okay, we look at the system, then there's actually two things that we could optimize. The first one is the black box itself. If we want better results, we can have better models. The second thing is the inputs. So there's two degrees of freedom, so to say, to improve a model like this. Now, the first one is clearly out of scope. Improving the model, that's what OpenAI does, what Google does, and so on, and tropic. That's cost billions, that's something that we can't do. But the reality is that improving the input, that's something that we all can do. That's also what 99.9% of all the AI companies that we see popping up everywhere is actually.

SPEAKER_00 21:55

That's the prompt engineering, right?

SPEAKER_01 21:56

Is that the prompt engineering at play? Essentially, that's the prompt engineering. That's also what we do at Hello Customer. And to show you how, or to illustrate how effective it can be, there's a story. This was for the Belgians in the audience, they know this. This was a very big controversy.

SPEAKER_02 22:13

Yeah.

SPEAKER_00 22:13

Yeah. So basically, what happens was you know, people go to the medical exam, they need that, you need to pass the exam to be able to go to medical school. And there's hundreds of young people doing that. And apparently some had tried to use or had used allegedly Chat GPT. To uh fill out the answers. That that's what that's the news story that broke in July of of this year. And it was, you know, it was it was mayhem because yeah, it's cheating, it feels unfair, etc. etc. So that's that's the story. But your friend is not uh in that no, but it the story did inspire him, actually.

SPEAKER_01 22:49

The story inspires you didn't buy him. He wanted to start playing golf, and apparently, if you want to start playing golf, you need to do a theoretical exam. The theoretical exam is an open book exam, you can bring whatever you like, it's on the computer. So he thought, No need to study, I'll just use Chat GPT, like maybe some of the medical students. Now he did this, went to the exam, walked out, and he failed the exam miserably. You really used Chat GPT, failed the exam. I have I'm surprised. I am surprised, yeah. So am I. Now I told him, okay, no, don't worry, you still won't need to study. Go back to the exam, but work on the input. Okay. So now he took his rule book, the rules of gold, with him, he uploaded them in chat GPT, so improving the input, only then afterwards copied the question from the exam, and he got much better results, much better output. The LLM suddenly started predicting words in a better way because it had more input to base itself on and did a lot better this time.

SPEAKER_00 23:53

Okay, I see what you mean. And that's kind of like you know, this way of improving the outcomes because you can't you cannot improve the black box, but you can improve the prompts. That's basically what we are doing as well. That's one of the evolutions also that we are going through.

SPEAKER_01 24:08

That's what most of the AI companies are doing. That's what an agent is actually. And an agent is a AI system that first tries to find all the relevant information that is out there, then adds that to your initial question, and only then sends a request or a message to ChatGPT. We looked at how Sky's chat bot before to give you an idea of what it does as an illustration. You ask it a question. First, it's gonna determine is this a question about the platform itself, how to use a platform, or is this a question about the actual feedback data? Yes. That's probably the most interesting track. It's a question about the actual data. Then, based on your question, it's gonna start looking for relevant context to add to the prompt. First, it's gonna check, okay, I need to add verbatims, maybe only for 2023. I wanted to know what customers were taking in 2024. So that's the time context. Maybe I want it only for some touch points. Maybe I only want it for some metadata filters. I know you want to know what people in Brussels who bought product A, what they are thinking about my company. Maybe you want to know it only for people who talked about a specific subject, a specific Isaac category, the waiting lines in the store, the quality of the products. Those verbitams you want to retrieve, only those that are relevant to your question. Then the system is gonna look in all of our databases to have all of the right KPIs, the calculation, so forth, to retrieve all of those. All of that useful information is gonna actually add to our question and then gonna send something to ChatGPT as we would do. And that brings an answer that is not only much more reliable and interesting, but that's also much more consistent.

SPEAKER_00 25:45

Yeah, that consistency is an issue. I'm doing if you explain it this way, I understand why the consistency is such a problem. But of course, if you can improve on this and keep this consistent or keep this rich enough, yeah, then you increase the probability that the outcome will be trustworthy, will be consistent in itself, right? That's the whole thing.

SPEAKER_01 26:05

Yeah, this is an example that we run into ourselves. So here I have a short screenshot showing a conversation that we have with our own chatbot. So we ask it what are the top five complaints that we're seeing popping up.

SPEAKER_00 26:18

This is a very good illustration of what AI can do for CX. I mean, it's an actual question. You don't do the number projecting, you just ask it. Okay, and then what happens?

SPEAKER_01 26:28

So you get an answer, you get an answer with numbers and so on, and it's all based on the data, the classifications that we have in the platform. If you ask this question twice, you will get the same numbers phrased differently, the same number. As you would expect, as it should be. Absolutely. Now we tried the same thing, just with plain simple chat GPT. So we uploaded all of the feedback, we asked exactly the same question, and we got this answer. And it's a convincing answer. It looks very beautiful, it's well structured, but we immediately saw okay, now it says that the most frequent concern are app issues and website issues, not the same as what we saw before. So we actually decided let's ask exactly the same question or exactly the same data again to Chat GPT, and we get a convincing answer again. But now it says frequent price increases and extra cost confusing billing is again different. And we asked it again, and now it's extra charges, customer service issues, and ATM access, and we asked it again, and it was again something different. You get the point. So it's yeah, the issue is because it doesn't have this extra prompt engineering. Yeah, um, it's inventing stuff, it's taking different approaches to ask to the same question. And if you ask the same question today and again in a month, you have no idea if the differences are because the issue is because it doesn't have this extra prompt engineering, yeah. It's inventing stuff, it's taking different approaches to ask to the same question. And if you ask the same question today and again in a month, you have no idea if the differences are because the experience of your customers have changed over time, or if it's just because ChatGPT took a different approach.

SPEAKER_00 28:04

So going back to the statements, AI doesn't know our crafts. Yes, it can if you prep it well, and if you use information that is properly prepared for the LLM to do its magic, keeping in mind that it's just an LLM, it's just a black box predicting things. So therefore, it kind of makes sense. So the answer is no, it can very much understand your industry if you're treating it well, if you're prompting it well, and prompting it is an art in itself. Now, I'm gonna move on to the next statement, which of course is very much connected to that. People sometimes say you just cannot trust AI. And I think one of the elements in that is the inconsistency of the answers, which we have just covered. I think what is important there is that we used to see computers as something that was perfect in predicting things, it was much better in predicting things that than a human being. You put something in, you get something out consistently. We are going to have to be we are learning to live with computers that do things inconsistently if you don't treat them right. But I think this graph or this picture, which is probably AI generated, right? Explains that, doesn't it?

SPEAKER_01 29:17

Yeah, it does. A calculator is not as powerful as an LLM, right? But if you put something in a calculator, you're 100% reassured that what you get out of it, that is correct. Large language models, if you configure them right, everything that we discussed in statement two, actually, they can get a high level of accuracy, fidelity. But even if you get it to 80, 85, 90, 95% of accuracy, that still leaves you at least 5% of the time that the answer that you're getting is not correct or not completely correct. Now, the question that you need to ask yourself is do you feel comfortable of taking business critical decisions when 5% of the time the information that you get is not correct, and that's a tough one. And what's your answer to that? I mean, how how do we how do we tackle that? Well, that's a widespread problem, and there's many companies that have been trying to find ways around this. One of the famous examples is perplexity. Perplexity was actually founded by PhD students at the University of Stanford who wanted to use Chat GPT in its early days in order to do their PhD, but run into the same problem. How do we actually enable our readers to verify if what we're writing is actually correct? So they do that by adding the pages of sources at the ends of their thesis. Now they say, what if we now copy that idea and we add that to large language models? So they make perplexity that works just like ChatGPT. You can ask a question, but at the bottom, you actually get a list of all of their sources from which the language model actually got those answers, giving you the ability to verify if the And that's what we do. That's what we do as well. So if I go to our own chatbot, what you see here, I asked a question about our platform. Tell me something about how you set up the email server. It gives you the tutorial, but then the end also gives you the articles on our knowledge base linked, where you can click on, and then you'll see them where you can actually verify or read further if you want more details. We do the same thing. If you ask a question about verbatims, you can click through and read through the verbatims if you want to actually use it.

SPEAKER_00 31:21

So you know where it's coming from. Yeah, so you you're kind of stepping away from the black box just a little bit. Now, there's of course another thing. I mean, you say you can't trust AI when that is being put forward, is the whole privacy issue, isn't it? It's uh I always I always find it remarkable that we we live in an age where, I mean, we've all I mean, the whole of the audience, we ourselves, Jonas, we've been confronted with with all things GDPR in recent recent years. And uh we all know how how delicate, how how brittle our privacy is. And yet at the same time, we're all enthusiastically throwing all kinds of things on any kind of LM that that comes that walks through our door. Yeah. And there's there's a paradox there. You know, the way that we are very careful and very very aware of risks, privacy is everything, and then on the other hand, well, just you forget about it, throw everything at it. Where's the truth? Where's the real risk? How do you see that?

SPEAKER_01 32:15

This is a true concern. I think we all need to be aware of it. We are all aware of it to some extent, but still not that long ago, we were in an AI session with business people who were explaining how they used tools like ChatGPT in order to make their work more efficient. And actually, during the session, we realized that they were throwing highly confidential information, numbers, financial numbers of the companies in ChatGPT. Now, this exposes you to two risks. One, you're sending everything to the US, to another company who could potentially read all of that data, who store all of that data. But the second thing is maybe even more important and often overseen. They use all the data that we enter in those systems to improve their systems. Now that means that if you give it some information, that information may become exposed to a competitor or someone else who's asking the same. But there's a true risk of data leakage where your confidential information actually becomes available to a competitor.

SPEAKER_00 33:20

We don't want to be a part of, and we have defended us against that. But it's a subtle thing. It's not just the first thing you say is you give information, everybody can understand that. The second thing is far more subtle, far more intricate, a little bit more complex, if you like. Because it's you're helping to improve a model, and the benefactor or the one who profits from it could be your competitor. Surely that's not what companies want to do. If you put it very bluntly, if you go even further to the extreme, you could say that you're eroding your own competitive advantage when you're using an LM in an unprotected way. So it is true to some extent. No, you cannot trust AI, therefore, you have to be vigilant and you have to work with partners that really know what they're talking about. Yeah.

SPEAKER_01 34:08

Yeah. And the good thing is, of course, we can exclude the risk that we just talked about. We at Hello Customer, for instance, we have our large language models running inside of the EU with the guarantee that the data that we send to them is never used to optimize to improve those models, isn't stored at their site. So that excludes the two risks concerning Prime Act.

SPEAKER_00 34:33

Absolutely vital. Okay, thank you. I think looking at the clock there, and we we have time for one of the uh the last one more statement, I think. I'm just gonna read it out. AI's true contribution to CX is going to be transforming insight into action. If I throw this at you, do you think that that is something we can expect in 2026 that is going to help CXs, CX people to do that, to take that beautiful leap that we've all been dreaming about for so long from insights into action, which actually creates impact?

SPEAKER_01 35:06

Yeah, I think that's really the next frontier for AI and customer experience. The next frontier. Yeah. Where we weren't until now, the industry has been focusing on insights. The next step is to have the AI also taking actions where it can take actions or suggesting actions where it's not completely sure what to do. So that means that the human will remain in the driver's seat, should remain in the driver's seat in this context, but the AI system can retrieve all the right information and propose an action that is the most relevant to the But you think that the human will remain in the driving seat.

SPEAKER_00 35:43

I mean, that's what you're that's what you're aiming at.

SPEAKER_01 35:45

Yeah, well, let's have the example again of in customer experience. You often receive feedback from customers and you want to close a loop. We see that very often in our platform. Up until recently, that was a very manual task. Companies who are more than 50 employees are day in, day out replying to the people who gave the feedback in the first place. Now, that can be automated thanks to AI. What we see customers doing is that actually they want to create more personalized messages still. AI write personal messages to the people who gave the feedback. And whenever someone gave positive feedback, for instance, they say, okay, this is not as risky. This can just completely be automated from start to finish. We don't need a human in the loop reviewing these messages. The score or the feedback, the initial feedback was negative, then the AI can still do a very good job in retrieving all the relevant information about this customer's use case and so on, and draft the message. But then they like to have the human in the loop revising the message that was drafted by the AI, maybe making some adjustments before it's actually answered. So that's a very pragmatic approach where the low risk cases you have AI controlling those, but the higher risk cases there you the AI make the human more efficient.

SPEAKER_00 36:59

It's a hybrid approach, right? It reminds me of something that I did a couple of months ago. A couple of months ago, I had to bring or I had to send, I wanted to send a birthday gift to a niece of mine. Well, there she is. She's on the bottom right-hand corner, Marco. And she lives in Pamplona. So see, she's studying there. And we wanted, as family, we wanted to send a little something. And my wife said to me, you know, let's send her a voucher, a breakfast voucher, one of those wonderful pastries they have in the pastry shops in Pamplona. And she said, you know, you take care of that. And so I was in front of the computer and I thought, what if I ask AI to help me with this? I was prompt engineering, if you like. I was putting all of that into AI. And then I sat back and what should happen. And it was indeed, as you say, almost hybrids. It was a hybrid experience because I saw it going through all kinds of searches, selecting things, sometimes asking me for feedback. Is this the kind of thing you like? Yes, no. And it progressed. And so I saw it going through all the motions, and this is crucial. I saw it going through all the motions that I as a human being would have gone through. It was stripping over the same wires that I would have tripped over as a human being, such as, oh, I see a pastry shop. It looks nice, but there is no voucher capabilities. Or, yes, there is a voucher, but oh, I'm trying to fill it out and something's broken, it doesn't work. So it was going through all those motions that I would have had to go through. And it was tripping over the same problems that I would have tripped over. It was trying to solve them, failed, came back to me. And then at the end, maybe you know, 40 seconds in, 50 seconds in, where it would have taken me about 10 minutes, it presented me with three choices. What I then did, hybrid, I looked up those things, just clicked on it to see what kind of picture shop is it exactly. Selected one and said, okay, go ahead. It then filled out the fountain sheet for me, sometimes asking me extra questions, and then presented me with this is it. Now I'm gonna give you the reins, dear sir. I'm gonna give you the reins and you can pay. Unfortunately, it didn't do that on my behalf, right? Could have been nice to have sent Margot a present with the greetings of Sam Altman, but it didn't do that. But it was a hybrid experience and it was fantastic for me. But for me, it was also an experience where I noticed as a CX person is that this machine, this black box, if you like, prompted correctly, did a wonderful job, but had the same or encountered the same hindrances that I, as a human being, would have encountered. And therefore, I think that the big breakthrough is gonna be from insights to action, namely to be able to indicate to us as CXS, to indicate where are the hurdles, if you like, that we have let in place, where is the voucher not working for a customer? Because if it works for a human being, it's gonna work for this thing as well. And that's something to remember. In another sense, even though a lot of things are changing, I mean, two years ago I would never have done it this way. I would have looked for that voucher myself online. That's changed. But the underlying process hasn't changed at all and needs to be if the base rish wants to get my business, they have to get their customer experience top-notch. Otherwise, I'm gonna buy, but this thing isn't gonna buy either. So I thought that was quite an interesting experience from a safe point of view.

SPEAKER_01 40:24

Yeah, absolutely. Yes, it's the perfect example again of the same thing. You have the human now, you, bro, who had to make decisions at crucial moments, at critical moments, when there actually needed to be a purchase, when there needed to be a payment, but otherwise the computer could do much of the things for you, the i system. And the good thing in this example is as well, those websites they have been optimized for human experience to make it as easy as possible for you as a human. Now that you actually see the bot clicking through, it's very similar. There's with the current state of technology, not yet a need when something is perfectly designed online for a human to optimize that for a bot. But that could change in the future.

SPEAKER_00 41:03

That could that could change. I'm looking at the clock, and I think we've kind of run out of time. If you do have questions, feel free to reach out to us. I see some questions coming in. If you have more, do keep them coming. Thank you. I was going to put something on the screen before anybody leaves. Do not miss this appointment with us if you want to see the products and especially the AI aspects of our problem for CX people in the webinar of the 10th of December at one o'clock. That's where we are going to explain that. So do feel free to join us for that. But there's a couple of questions actually coming in as we speak. And one is a person asking the question what should I focus on as a CXer? What should I do in terms of AI? Should I study something? Should I get to know one specific tool? What would be your advice?

SPEAKER_01 41:57

I don't think you actually need to study something. I'd rather say get your hands dirty. Atomic habits, try one thing, one use case that might interest me you in one platform. Try it out for a week, see if it works or not. Then the next week try something else. And by working with the machines, with the algorithms, you'll find out what works, what doesn't. Share your experiences with co-workers and yeah, atomic habits.

SPEAKER_00 42:22

What is atomic habits exactly?

SPEAKER_01 42:24

I mean, just uh is it something everybody knows and the I don't know? I don't know. It's a concept saying, like, don't try to completely change everything at once, but try to form a habit by doing small things at a time.

SPEAKER_00 42:36

That would be your advice for our audience. Like in AI, just little little things, little steps. Yeah, right. Little steps. Yeah. My advice on top of that would be so atomic habits. I have to look that up. Atomic habits. It's a good book. Good. I love books. The other thing is don't give up. I think we all have our part to play in this. We cannot expect AI to solve all of our problems. I think some of the CX problems are so fundamental to the success of businesses that we have an enormously important role to play, but we mustn't give up. We can teach AI a lot, we can expect more from it, we should expect more from it. And together we will raise the bar. One more question. What is the biggest trend that you expect in CX and AI in 2026? One thing that you would put forward.

SPEAKER_01 43:24

I'd say that's closely related to statement four that we discussed. It's going from insights, from purely generating insights to also taking actions with the human in the loop. So yeah, less less dashboards, but more actions online. But yeah, be very prudent. Don't let the bot decide everything. Make sure that you have a human in the loop. There's multiple ways of doing that. It's also actually one of the important parts of the AI Act in the European Union. So oftentimes it's even by regulation.

SPEAKER_00 43:58

Well, thank you. Thank you very much, Jonas. Thank you, ladies and gentlemen, for joining us. Again, if you have questions, you see our email addresses are on the page. But thank you so much. Once again, December 10th.