How Much Can You Really Save with AI in HR?
In this episode of the HR Leaders Podcast, we sit down with Carlo Steenvoorden, EVP HR People Services, Analytics & HR AI at KPN, to unpack how a 100+ year old telecom company is moving from legacy HR systems to a fully conversational AI powered employee experience.
Carlo explains why KPN made a bold decision to declare that the future of HR interactions is conversational, with systems pushed to the back end and one intelligent interface in front. He shares how reducing human led HR queries from €15–20 per case to cents per prompt unlocked both massive efficiency gains and a better employee experience.
Most importantly, he breaks down the real transformation behind the technology, from rebuilding HR team capabilities, to adopting product thinking, to deciding where AI belongs and where humans must stay firmly in the loop.
🎓 In this episode, Carlo discusses:
Why in-house AI development accelerated transformation
How hyper personalized learning replaces one size fits all training
How HR query costs dropped from €15–20 to cents per interaction
Why 25–35% of the HR team had to be renewed to move fast enough
How KPN shifted from legacy HR screens to a single conversational interface
What if global compliance took 5 seconds?
It does now.
atlas copilot partnered with our friends at Deel, the all-in-one Global People Platform, to bring on‑demand employment matters and compliance intelligence into your daily flow. With Atlas' GenAI agentic platform and Deel’s vetted sources, get authoritative answers across 150+ countries.
Ask anything:
“Is a 3‑month probation period compliant in Mexico?”
“What termination notice is required in Germany for 18 months’ tenure?”
“How long are the maternity and paternity leave in the UAE?”
You get clarity, that used to take hours, in seconds.
This is not a feature. It’s a new operating model. The chaos of Slack threads, scattered PDFs and expensive calls collapses into one trusted interface. Your team moves faster, escalations drop, decisions align and you see regulatory change as it happens.
If you’re still stitching notes from outside counsel, you’re already behind. AI-Smart HR leaders are shifting to always‑on intelligence. Don’t let compliance feel like navigating a minefield.
00:13
How are you doing, my friend? I'm good, thank you. Nice to see you. Yeah, nice to see you too, great. You just got off stage? Yes, just got off stage. And you still got energy? Yes, I have. More energy, I think. Yes, indeed. So you were talking about from legacy HR to conversational AI. Yeah. Just a small topic.
00:31
Just small, just not transformational at all. No, Chad, so talk us through that journey. I think one and a half years ago, we made, I think, a bold move. We stated that the future of HR interactions is going to be conversational and that all systems behind it are going to be back-end systems. And that's cool. That's a bold move that's leveraging technology as early adapters. But it also means really transforming the employee experience, right? From...
00:58
going through screens, optimizing those employee experiences as good as possible to having just one user interface, Gen AI or even a Gen Tech loaded in front of people, which does everything for you. That's a cool transformation to speak about and to undergo at KPN with challenges and things. Walk us through the journey.
01:21
Yeah. Where did you start? We started one and a half years ago. We saw that a human-led transaction or query, and we have tens of thousands of those a year. We saw that this was costing us roughly 15 to 20 euros each. And that with new technologies, it would have been a good prompt with an answer, 15 cents. So imagine that cost combined with the employee experience, there was a major opportunity.
01:51
But it was a bold move. Technology was new. It goes really fast. So we decided if we want to have a chance of changing rapidly the employee experience to how they're going to see it as customers somewhere else with OpenAI, ChatGPT, et cetera, then we need to go there and all in. So we went there, started to build things in-house. Oh, you did it in-house? Yes. I was going to ask you that next question. Because a lot of the vendors that are here back then didn't even exist. Nope. And...
02:22
I think that's a choice. If you go, if you take a vendor or you wait for your vendor, you're going to have, I think, pretty good productivity gains within the system. But we wanted all HR transactions close to where people are. So all systems,
02:39
with one user face in front, and then you need to connect, build the layers, mesh, agents, everything. And then you need to build it in-house, otherwise there was no supplier doing that. Even now, right, to your point, because you have current AI assistants, but they'll only connect a few of those things.
02:58
Whereas if you want to bring in benefits, payroll, learning... Learning is a big one. Yes, it's a big one. You hyper-personalize it through this. Yeah, I haven't seen many companies do a great job. Connected behind the screens, the learning with your role, with other things you're going through and let that agent layer take care of it. Obviously, we're not there yet. But by going through that future of HR interactions is conversational.
03:25
We started that journey at least one and a half years ago. I mean, I wish we had an hour. We can do it later. I'm assuming you're going to tell me the first and most important thing is to get the data right. Because you can have the best system possible, but if you have garbage in, you have garbage coming out, right? Together with getting the skills to do it. Skills taxonomy. No, not also, but that's for learning. Skills in your team. Skills in the team, in the department, right? Because...
03:51
Imagine we are traditional, let's say we are a forward-looking company, but more than 100 years old. And imagine the HR teams brought us here, but the change of skills in those teams needed
04:05
I think a lot of companies outsource their tech qualities to IT, not in the teams. And now suddenly this technology can let you do it yourself. So you need tech-savviness back. You need a learning mindset back. You need that experimentation back. It's so different, at least in our case, was so different compared to what there was. So with the data, right, but also
04:28
the good volume of people with the right skills to make this work instead of leveraging the cloud to one or two experts in IT. So that transformation came together. Yeah. What are some of the skills and capabilities now that you have in the team and that people listening to you feel that every HR team should have? Yeah, I think it's threefold. I think it comes with a mindset. It starts with the mindset being open, curious, experimentation.
04:55
Secondly, it's tech-savviness. You need to know what you're doing, how it is,
05:00
in the back, how this also can connect to this new world of agents or people handling cases should not be handling cases only, which is a lot of value add, huh? But they also should think what part of these cases can we leverage through the new technologies themselves? And where do I bring the value and the new technique? And nobody's going to do that for you. You need to have the skills yourself. And the third,
05:27
I think, yeah, as I said, it's experimentation, it's learning, it's in those cases, the technical savviness and that product thinking of HR. Yeah. What was the balance of building skills versus buying? I think the transformation journey has been that roughly 25 to 35% of my department has been renewed in the last year. Okay. And that was needed to get this transformation within 24, 30 months going.
05:57
otherwise it would take too long. What were some of the challenges that you faced on the journey? I think it goes so fast that even our efforts to be such an early adapter, and I think we're quite further ahead than some other companies on this, it's the customer expectations.
06:17
from the outside world exceeds what you can build in-house or what you can buy on HR space that fast. So that drives apart. So somewhere you need to go, biggest challenge, somewhere you need to go from experimentation and learning mode to full operation mode.
06:35
And when is the moment when the world is moving faster than you are? That's such a headache. Yeah. That's a headache. But it's ongoing, right? It's ongoing. There's no such thing as a... There's no destination. No. And it's bringing the efficiencies already. Yeah. But you want to... The speed is unparalleled outside and you need to bring it inside. Yeah. But it's such a cool journey to go through and how you see that employee experience has been changed or is changing in the last 18 months. Yeah.
07:04
That's super cool to see. What's the biggest use case?
07:07
The biggest use case, I think, is twofold. I think one is really on the operations side. So I think 60% of your queries are easy transactions or missing knowledge, not able to find knowledge, don't want to find knowledge. And that can be done easily. And that brings time back to employees, managers, HR itself. That's a big one because there's a lot of that 15 euro versus 50 cents cost. And then you space up your energy within HR to drive the bigger
07:36
business strategy. And the other one is, as we just touched upon, is the learning. The learning piece where you can actually don't have personas anymore. You don't have 20 or 15 personas of career moves. You actually have individuals. You have individuals suddenly because those agents connecting your wishes, your skills, your learning,
07:57
they are never getting tired. Those agents work on this agent, their context is so different. These two, one is operational efficiency, and the other one is driving hyper-personalization, and both are very good to do with this new technology.
08:14
But practically, how are you connecting it to learning? Is it recommending a link to a course? Because this is a big challenge I'm seeing where learning and our AI assistants stick in silos. Yes. So this is still, for us, it's a journey we will build in the next...
08:32
six to eight months, so we have the plans, not yet the full execution. What we probably will do is we have one vendor, which will help us through skills taxonomy, through connecting our skills taxonomy with people's individual wishes on their careers, even inside or outside the company, and then connecting it to how you mitigate the gap you may have. But then you still need that interaction layer, people going into this, and if you connect that one also to the user interface,
09:02
the questions about these things will bring these answers together. We need to test it, we need to deploy, but I think a year from now, let's have the conversation. Yeah. I think my concern is in the future, and I'm seeing it now, is you now have agents with all of this rich context from the individual, but then we send them to a one-size-fits-all course. Yes.
09:23
It's such a shame. So you have this incredible knowledge on the individual, but we're still serving them the same course. But the whole learning in this time, the whole learning proposition might change. I think so. You don't go to classroom or two-day trainings anymore. You do
09:41
snippets and bytes of 20 minutes, which is contextual and relevant for you. So also the learning departments within all companies need to reimagine with AI how this is going to bring it to other people, to your employees in a different way, drastically. Where do you, how do you find the balance of when you have, when or where you have the human in the loop? Yeah, yeah. I think every process,
10:09
needs a human in the loop on improving, on seeing the accuracy rate, et cetera. It's just not on each individual query prompt anymore. So you have to establish a learning loop and set your guardrails where they're looking at. But that means that it's scalable and human net processes are not scalable at all.
10:34
Yeah, no, I love that. I think it's like also from the context of if someone's asking a question around
10:42
Mental health. Yeah. You don't want. No. The agent. No. But you have. You have certain processes. Where you can do. Human in loop. And you have certain processes. Where you want. An individual. Behind. Yes. The prompt. So. Complex individual cases. Okay. Health. As you said. These kind of things. Maybe not ready yet. Yeah. But a lot of things. About your expenses. About your benefits. About your personal data. These kind of things. There. It's not.
11:12
It's okay if one out of 100 questions is wrong and you have the learning loop to continuously improve. You need to make that contextual balance. But there's a lot of processes which can benefit from this. Yeah, of course. I'm just saying there's certain things that it's great for, like the transactional stuff, right? But I think we need to be...
11:32
But it's also being thought for where HR is really driving value. And now maybe we did a lot of things on the energy side where we didn't really drive the value. I know it's still a journey, but what do you know now that you wish you knew when you started? That's a good one. I think
11:57
If I knew how fast it would go, I would deploy even earlier. No, in beta and learning mode. So your employees... Get more feedback. Get more feedback. And then it's not to drive the exponential factor outside, mitigating already how good it is inside to be earlier in a journey in learning mode. We had, at this point in time, more data to have even a better product instead of having less data and improving ourselves. I think...
12:25
That would even be better. Yeah, I love that. How are you measuring the success of this? So first of all, adoption rate of how many employees are using this frequently. But also the ticket load, as we said, the query load, is it really reducing? So far, it has been reduced 30% in a year, more or less.
12:48
It's not yet where we want to be. That's still a huge unlock. It's still huge. And now we're measuring what are the biggest volumes and how do we tackle these one by one. And if you then improve the product and see that it's going down, you do something great with that human in the loop learning loop. This is how we measure it more or less at the moment now. And it's giving you the feedback of where you need to invest.
13:13
more right it's the time and energy and also contextual feedback because it's giving you the feedback not only on a bucket but this domain versus the other main what are the what is employee listening almost what is what is happening what are the questions in one domain compared to the other yeah and then you can learn and tailor your answers even better that's really interesting
13:33
How does this show up? Is it embedded into Slack and Teams? Until now it was in our ERP system and in two weeks time it's going to be in Microsoft Teams. So it's going to be just another colleague. It's going to increase the adoption. But then you get that chicken and egg story like is it good enough compared to the outside world because then it's mass adoption.
13:58
Yeah, but to your point earlier, you'll get the feedback and then you'll learn. It's a balancing act. What advice should you give to anyone who's starting this journey? Be open, experiment, have a learning mindset on this and do it iterative. Don't do a full Big Bang because you will learn. And make sure HR teams itself have the right skills. Don't leverage everything out externally.
14:23
Yeah. Listen, man, always fun. Appreciate you. Enjoy the rest of the day and I'll catch you soon. All right. Thanks a lot.
More from the HR Leaders Podcast
Carlo Steenvoorden, EVP HR People Services, Analytics & HR AI at KPN.