Meet Sophie And Her Path To IT

SPEAKER_01 0:02

I'm Ben, your host today, and I'm very happy to be joined by Sophie van Nevel. Sophie has worked across several organizations in Belgium, was named Digital Coach of the Year in 2023, and gave a keynote on data governance that I still refer back to today. Sophie, welcome.

SPEAKER_00 0:18

Thank you, Ben.

SPEAKER_01 0:20

We're recording from the Data Roots office in Ghent with a view over a city you know quite well. You studied here, right?

SPEAKER_00 0:27

That's correct. I studied in this beautiful city.

SPEAKER_01 0:29

And what did you study again?

SPEAKER_00 0:31

I studied civil engineering with a degree in chemistry.

SPEAKER_01 0:35

So, how did you uh get to IT then? Because you started your career in ING, right?

SPEAKER_00 0:40

Yes, I started my career at ING. Um, and I started in a traineeship, and the first mission that I had was within IT. Um, and then it was a banking crisis, and I never uh left IT ever since, and I'm still working within the IT environment in the meantime.

SPEAKER_01 0:57

Could you elaborate on your career path a bit?

SPEAKER_00 1:00

Yes, so I started my career at ING, um, and there I started in several different roles, um, always with a huge link on transformation and change management. Uh, first more in IT strategy, then later also in data, data governance. And that was a bit the red wire through my career. I was always in data and things which were moving forward where there was a clear future in. Um, and that's how I landed in the whole data and AI space.

What TVH Does And Why Data Matters

SPEAKER_01 1:30

Today you're working at TVH. Could you explain what TVH does as a company?

SPEAKER_00 1:36

Yes. So TVH is in fact a Belgian-founded company, it was a family-founded company, Termote van Aalst, located in Warehem. Uh, in the meantime, it's an international company with uh entities all over the world. Um, we have around 5,000 employees worldwide, um, headquartered indeed in Warehem. And what is our main mission is that we keep your business going. So we deliver the right spare part at the right moment to your business. This is for material handling, industrial equipment, construction equipment, but also in agriculture.

SPEAKER_01 2:12

Interesting. And how many people are working in Belgium again?

SPEAKER_00 2:15

Around 2,500 people.

SPEAKER_01 2:17

Crazy, it's a very big company. Is it still family owned today?

SPEAKER_00 2:21

There's a part still owned by the family, but it's also part of the Dieteren Group, which has a part of the shares of TVH.

SPEAKER_01 2:29

Okay, interesting. And so you started working there about a year ago?

SPEAKER_00 2:33

Yes, exactly. First of September last year, I started at TVH.

SPEAKER_01 2:37

Did you have a mission right away, or why did you join TVH?

SPEAKER_00 2:42

I indeed uh joined with a mission uh to TVH, so I'm responsible for our data products uh teams and for AI. So concretely, that means that uh further building out our data strategy within TVH. TVH has as such a lot of data because we have a lot of different spare parts with a lot of different uh data linked to that, so it's really key and core to what we do as a company. And my main mission was in fact to further build out our products around data with equipment data, service data, and our product data, but also to further lead the strategy for uh AI within TVH.

SPEAKER_01 3:23

Just out of curiosity, because the the scale is huge, I think. How many products uh do you offer with uh TVH?

SPEAKER_00 3:30

Uh we have around one million uh different SKUs that we offer.

SPEAKER_01 3:34

Um what did you start with? Because you you joined uh the company, and then did you have

Building An AI Team Out Of Silos

SPEAKER_01 3:40

to formulate a strategy or did you do use case discovery, or how did that uh go?

SPEAKER_00 3:45

Yes, specifically on the on the AI part, there was already a high-level plan when I started, uh, which uh consisted of three different elements. And so there was a part for AI that we said, okay, we want to broadly engage everyone in the company and give also kind of change management and a training plan to get everybody acquainted with um day-to-day AI tools. That was one thing. A second thing, and there we really started, we did not have a team in place yet at that point in time. So I really started building a team around um our own AI development use cases because we had um really a strategy where we said, okay, we really want to invest in those cases that make the difference for TVH. And I was typically in the domains of our service, equipment, and product data, but also in e-commerce and search, where we are really building out a number of capabilities, and we wanted to develop a kind of internal muscle as well. We were before a lot dependent on external parties. That is the reason why I started now building out also a team, um, which is uh since January this year in place, and we're still further building out.

SPEAKER_01 4:57

Interesting, and so there were no in-house build tools, uh AI systems before, right? So it was all embedded AI or everything was was bought?

SPEAKER_00 5:08

No, there was embedded AI, but there was also already in some areas of the organization, there were already indeed certain teams or certain people working indeed on AI. But it was more siloed and not really brought together in one strategy.

SPEAKER_01 5:22

So the goal was a bit to centralize and then have a company-wide strategy?

SPEAKER_00 5:26

Yes, yes, because of the importance of the topic. Um, we really thought that centralizing it in a data and AI organization uh brought value to really strategically um invest further in AI and also to bring together different people from different angles of the organization to start sharing best practices, to set up necessary practices on MLOPs, for example, the platform behind, and that that was all centralized in one organization.

SPEAKER_01 5:55

Makes sense. And so you'd start with discovering use cases then, I imagine. What did the way of working look like, or how did you tackle that process?

SPEAKER_00 6:04

Yes, there was already, in fact, a list of different use cases. Um, but these were all things that were tried out, that was in more in a POC modus, and by further building out a team, we started taking up a number of specific use cases where we saw there was a high value case. And that is in fact how we started. We started with a few use cases with high visibility, and that we directly started uh delivering upon, plus also demoing to different uh business stakeholders, and that started to get traction because people see what you actually start delivering. Yeah, um, and that is the reason why then people said, okay, we want to have more. And in the meantime, we started further building out a team, and that is the reason why we got more and more different use cases uh that we started upon.

SPEAKER_01 6:53

Along the way, you noticed that this was the right way of working, or did you need to change course a bit?

SPEAKER_00 7:02

Yeah, well, I think what was important first uh was in fact to get started so that we could really actually deliver upon

Early Use Cases Plus Platform Foundations

SPEAKER_00 7:10

a number of those uh those use cases. Um, but in the meantime, what we did in parallel was building up indeed our AI platform uh to really start scaling, to start setting up a number of practices behind, uh, and that we did in parallel. So really making sure that we could towards our businesses show the actual value, and in the meantime, building those foundations behind because these these were not yet there, uh, and we're still further building out those foundations. So um, what was of course important is that we had a number of those different use cases, but it is all more use case-based. And that is the reason why it was somewhere in in March, I think, when we presented uh the different use cases also to our board and our global management team that he said this is very nice, and we really see the value of AI. But what is in fact now the actual value case of AI? And that is in fact the moment where we started also shifting a bit our strategy from more bottom-up use case that we had, also to really looking at where can AI strategically have an impact for our

Shifting From Use Cases To Strategy

SPEAKER_00 8:20

organization.

SPEAKER_01 8:21

Yeah, that's what you see in in most businesses today. I think before, let's say, the Gen AI hype, I think it was important to, for each domain, find the right use cases, get a value feasibility case, prioritize, and then execute them. But today it's not only about those siloed cases. I think AI has an impact on the organization itself and on the business model, even sometimes. So is that then where you're focused on right now?

SPEAKER_00 8:49

Yes, exactly. Yeah. So we really started in off in March, where we then really had a look, and we did that also by using AI uh with our own team to really look at where AI is bringing extra value for us as a company, and that's from different angles, really looking at a customer perspective, and how is our customer going to change with AI, and what is the impact then, for example, on our e-commerce uh platforms. Um, and next to that, also looking at impacts um of AI on our workforce, on the different capabilities that we had. And by doing that, we could really see also where is for us the biggest opportunity, and that means that you can then start really prioritizing the biggest impacts for AI on the company. And I see by doing that that this has really changed the storyline because we started first, yeah, getting everybody enthusiastic, use cases, bringing in. But the more you start also looking at it more strategically, the more it becomes kind of serious business. And then you really start transforming your organization towards embedding more AI in your processes, but also rethinking part of your business with AI.

SPEAKER_01 10:02

And and what does the process then look like? So, first you try to identify the domains where you think AI can add a lot of value. Do you then engage with the business stakeholders at first, or do you first come up

Mapping Opportunities With Department Leaders

SPEAKER_01 10:15

with a strategic plan? Could you could you help us with that a bit?

SPEAKER_00 10:18

Yeah, yeah. So, what we did, we made in fact a kind of opportunity map. So we looked at our full landscape that we have within TVH or core uh capabilities that we have, and then every time we did an analysis on the capability as such. So, meaning, for example, if I take IT as a capability, you could say yeah, there is impact uh on the way we do software delivery today. So that is a capability that we then further investigated and really looked at okay, where is the impact exactly? The same on the type of jobs that we had. So, really um, we made a kind of AI model based on industry practices to really look at yeah, what is the impact on certain roles and which part of a role is going to change with AI. And with that analysis, we started getting in detail engagement with different um departments that we selected, and then we started really with the leadership teams of those departments, really start engaging around this is the opportunity that we see. We still need to put that in their context, and that is where we then engage with them through a number of workshops, um, and that they then really say, Okay, yeah, this is how we take that into account in our strategy. And that is finally what comes out is in fact a kind of strategy from that department. But us as an as an AI team really deliver then, yeah, here you see an opportunity, and this is what you can seize as an opportunity.

SPEAKER_01 11:46

Remarkable and impressive because I think many organizations today are struggling a bit with what position do we want to take. We know that there's agentic, there will be agenc organizations, but there is like a full spectrum of possibilities in how you want to do that. So it's uh it's cool to hear that at TVH that's quite clear already. I can imagine that it will evolve over time.

SPEAKER_00 12:08

It will still further evolve.

SPEAKER_01 12:10

You also mentioned um the importance of the the workforce and change management. Could you show some or share some initiatives that highlight that?

SPEAKER_00 12:19

Yeah, yeah.

Training The Workforce For AI Tools

SPEAKER_00 12:20

Because I think based also on the exercise that we have done on the impact on our workforce, we really see that in fact most of the jobs are impacted. It's not necessarily impacted in the sense of uh there are our job losses, but more in the way every job will look like with AI, will be different in the future, and that will be in every type of type of job. So that's the reason why we really have also a program, we call that Forwards with AI, um, where we take our full workforce. So everybody of our workforce is fully involved in there, um, along in a journey where we give them training, coaching, guidance on uh, for example, all the tools that we have in our workplace. We do that with we have a Google stack, yeah, so it's with Gemini. Um, but we really train them and really take them along into all those AI tools in their day-to-day work.

SPEAKER_01 13:14

So, with a strong focus on both transparency and collaboration, because from the start you want to show them and have the message we're in this together, and we really want to guide you and help you the best way possible in yeah modernizing uh their jobs, let's say. Yes. Much has been said already about SDLC software development lifecycle. Is that also a use case you're looking at within TVH?

SPEAKER_00 13:38

Yes, yes, that is definitely also one of the key priorities in the IT department to really look at how can we embed AI within the full SDLC. And it's not only within development, because you can say, okay, you use AI within development, and there you increase productivity and you go faster, but it's in fact in every step of the SDLC, already starting from your requirements, from your analysis, then all the way to testing. So I think we're still at the starting point, but it will dramatically change how we work today, also within a typical IT department.

SPEAKER_01 14:14

Because indeed there are ongoing discussions on where we should have agentic and where not. For example, saying we will do both development and testing agentic, or only development agentic and then testing is done by human. How do you make those decisions? Is it by experimenting or is it something you also studied and then made a choice, or is it still ongoing?

SPEAKER_00 14:37

I think today it's still ongoing. Yeah, so we we will now start more in development, but I think we will see quite soon that it will impact also other functions. And that is the reason why currently we're really looking at an end-to-end product to really see all the different steps within SDLC and then see how that will now change our way of working.

SPEAKER_01 14:59

Okay, and and you just mentioned a big program. I think that's indeed a smart way to uh be transparent about what you will be doing with AI. Are there any other initiatives uh in change management that you could elaborate on?

SPEAKER_00 15:12

Specifically an AI, you mean?

SPEAKER_01 15:14

Yes, I it's because uh we we had a conversation some months ago, and I remember that you said something about an AI lab, and I think it was very interesting, but I don't remember the details.

SPEAKER_00 15:24

So I okay, yeah.

Labs, Tool Choices, And Shadow AI Controls

SPEAKER_00 15:25

So what we what we do as well as part of our forward with AI program is that we have transformation labs. Uh and transformation labs that means, in fact, that we put together um a specific team. They have a specific need today. For example, they say, uh, this is uh within my business a problem. Uh then we put them together in a two-hour workshop with the necessary tools that we have available within TVH, can be with Gemini, with Notebook LM, uh, can be even a part five coding, and we put them together around a concrete example. And the fun thing is that typically the people who come up with the ID they have already experimented things, uh, but also it has an effect on their colleagues who are together in that workshop. And it starts really, people start really practicing more and more, the new ideas come out, and that really feeds also further our backlog and feeds also the cultural change within our organization.

SPEAKER_01 16:24

Yeah, it gets the conversation going.

SPEAKER_00 16:26

Yes.

SPEAKER_01 16:27

You you just mentioned Gemini and also Notebook LM. That's also a very interesting topic today, the tooling strategy of an organization. Um, because uh everyone wanted to have access to some kind of generic tool, like for example, ChatGPT, Clot, and so on. Um, but now there's also more specific modules like cowork, code. So, what you saw is that organizations shifted towards ChatGPT because it was very important, there was a lot of chat AI. Um, but then Clot came up with cowork, and then it was important for organizations to think about should we shift to co-work, yes or no? Some of them did or no on the CLOD stack, but then now you have also co-pilot cowork, there's also chat GPT work, so it's it's an evolving landscape. So, both on tooling and technology, how do you maybe two questions? How do you follow up on the tooling landscape and the the evolution in AI? And then second, how do you battle shadow AI and make sure that you have a good tooling strategy?

SPEAKER_00 17:32

Yeah, well, it's it's a difficult uh topic, yeah, because uh definitely today a lot of there is a lot of evolution in tooling, uh, and sometimes it can be that now today with cloud, for example, we see okay, it's front running, but in six months it can be different. Um, so uh our overall strategy is that for the large organization we provide the Google Stack because that is what we invested in in the past already, and that is in fact the standard tooling where we really say, Okay, we put uh Gemini first as a strategy. But of course, we see that in certain areas, like for example, in IT and in coding, and even also in finance, uh, with um the interlink with Excel, for example, uh, that we see a lot of people also asking for cloud. Uh and um we have a responsible AI committee in place within TVH, and within that committee, we really then decide okay, if we see a specific value case, for example, we had a case with uh with finance recently. Somebody in finance experimented in his personal life with cloud, and then said, Okay, but I see a really good use case within TVH, I would like to experiment on that. Um, and of course, what we don't want is to have the free version um used with data from TVH. So that is the reason why we really had a look. Is Gemini capable of doing your use case? And if not, then indeed we can go for enterprise licenses for cloud. And that is the reason why we currently have a kind of test uh group in place with licenses uh for cloud, only six months, uh, because we really believe that Google will also catch up, or there will be other models coming in, but really looking more from the value case that we have, and then see, okay, at this point in time cloud can solve it, let's do that. But really on specific cases and really uh with a kind of tailored approach.

SPEAKER_01 19:34

And by that, I think you uh replied to to both uh questions because the the second question was about battling shadow AI, but I think indeed by offering the enterprise version, typically it's more safe than if they would use it in the free version, and certainly if they share company data. On that, do you have certain policies within the company, or do you also um share best practices and guidelines and so on?

SPEAKER_00 20:01

Yes, that is part of that responsible AI committee where we indeed we share, we had a kind of e-learning for all our uh employees as well around the responsible use uh of AI. Um, but also we share uh we have a whole guideline on confluence where we share as well like the best practices, what is allowed, what is not allowed. Um, and we're currently also working on a kind of AI catalog where we want to show in the organization these are the tools which are available, these are approved tools, uh, and for specific other tools that they want to use, they need to go through a process of asking uh specific uh requests.

Agents, Guardrails, Costs, And The Road Ahead

SPEAKER_01 20:38

So that's for these tools, then what about creating agents? Because you sometimes have uh people in the workforce that are a bit more advanced who are using this at home as well, and they want to build a somewhat more uh advanced agent. How will you enable those people in in doing that? Um, I would like to make maybe a parallel with uh business intelligence where you also in the beginning had a shift towards centralizing to have a bit more control, to share best practices, to build the most important uh reporting, but then there was also a decentralizing uh wave, let's say, where there was self-service BI and you wanted to make sure that business domains with business knowledge and um expertise could also build reporting on their own. I think we will see a bit similar patterns with agents. Is that something you're already looking into, or are you really very much focused on centralizing, or what does that look like?

SPEAKER_00 21:39

Yeah, that's a very good question because I think today um businesses will anyway experiment and they will, anyway, uh you see it already that some businesses are already making their own agents, they're already vibe coding things. Um, and that is the reason why we want to have a kind of open setup with the businesses to clearly understand. Understand what they are doing. Because if you you really centralize everything and you don't allow a number of things, I'm sure that they will find a way you will even get more shadow AI. So that is the reason why with uh the initiative around the catalog and really logging as well. What kind of prompts do we have, what APIs do we have, what MCP servers uh do we have within the company. We try to also centralize that, but also make it available to other people in the organization. Um, and on the other hand, uh definitely also engaging with our different business uh stakeholders. We have, for example, now a track around vibe coding with uh a kind of limited set of people to really start looking at okay, how can they vibe code within certain guardrails and within certain yeah, I need guardrails to make sure that afterwards, if you say, okay, there is this is a non-risky uh process, you can of course keep vibe coding, but as soon as you need to have something enterprise great, you need to have IT further involved, and we really want to keep having that link, yeah, not business developing something and then saying, Yeah, now it becomes tricky and it becomes more complex. I hand it over to IT, like we had in the past with yeah, indeed with BI, but also with other kinds of tools that they made in Excel. Um, and that is indeed what we now really bring together, yeah, because that I think is the only way by bringing it together that you can really avoid all those shadow AI uh tools.

SPEAKER_01 23:35

I couldn't agree more. I think a big difference with BI is that the accessibility and the ease of use is much, much higher. Yes. And thus it's already happening, it doesn't make sense to be ignorant. It's better to from the start already collaborate, be transparent, um, be open to collaboration and uh licenses and also agentic development. So very cool uh to hear that. One downside, however, could be uh the the token usage, the the costing. How do you handle that? Because that's also a big topic today. There are um more uh expensive models, there are cheaper models, also open source, closed source. But uh let's let's discuss that later, maybe. But how do you make sure that everyone can still develop build agents? So you want to enforce them in doing that, but at the same time, you also need to make sure that costs aren't adding up too much, of course. Uh, how do you find that balance?

SPEAKER_00 24:35

Yes. Um, I think uh there was recently also, and I think somewhere in July in the newspapers also that a lot of companies struggle with token usage and they saw their costs at e-cost going up. I think we don't have that at this point in time, yeah, because um everything that we have in our Gemini stack uh is part of a kind of fixed amount. Uh so it's not that we pay for extra token usage, the same what we have now with Cloud. But of course, it has a downside as well, because uh some developers say, ah, after two hours uh I'm out of tokens and uh don't have uh the limits anymore. But they'll be quite strict on uh at this point in time, and only if there is a kind of specific value case, and then we can indeed further open token usage, but we're not opening it uh for a large audience because otherwise you cannot control your costs anymore.

SPEAKER_01 25:26

So because of the pricing models of Gemini and Claude, it's per seat, and you don't have the risk of having too much tokens. So there was no token maxing at TVH.

SPEAKER_00 25:36

No, no, and I was happy about that.

SPEAKER_01 25:38

I can imagine. Looking forward, if we first maybe focus on the short term, what are the next use cases you see?

SPEAKER_00 25:47

I think for the definitely still for this year and also beginning of next year, I think we really want to focus still um our program around further engaging everyone in the company so that forward with AI program continues with uh now also like vibe coding, who knows what is next that we uh that we take into account. Um, and on the other hand, uh, we will definitely further focus our own AI development activities more in those core areas for TVH, so everything related to product data, equipment data, growing new services, and then as well in in e-commerce and search. Um, and definitely also, I think definitely there will be some disruptions in the whole e-commerce uh area, uh, because you will have more agentic e-commerce that is really also an area where we really need to further invest upon.

SPEAKER_01 26:38

Could you share a bit more about that? Because I always like to make it very tangible and concrete for our listeners. Could you share a bit more about the agentic e-commerce?

SPEAKER_00 26:47

Yes, I think the the way that customers uh will interact, you see that already today, and that they will interact with typical platforms or typical, yeah, like you you search something on ball.com. People today take a picture, uh, they ask to Gemini of OpenAI. Um, where can I find this? And then, of course, you need to appear there and and they need to find you as a company, and then have also definitely the right data that they can say, okay, we order with you. So that whole experience, and people will not necessarily go anymore to your website, but will find through other channels uh your company, and then they need to find you. And that is typically going to change the way customers will interact with us, but also much more, you will have agents who do necessarily buying and selling, and so there also data is really key. If you give the right data or you have the right data specifications, of course, you will be higher ranked than others. So that is the whole change that will happen in the whole uh e-commerce and and search uh capability that will affect a lot of us uh who do in that uh space.

SPEAKER_01 27:59

So, really focused on the customer journey and then what technological capabilities need to be enabled to make sure that they can find us. We already discussed that TVH owns more than I think one million parts. So I can imagine that it's important to have good metadata and descriptions of the products and so on. Is that also something that you are enhancing with AI?

SPEAKER_00 28:21

Yes, yes, and that is really in our core business because that is making sure that we have the right product specifications for both products for equipment that we have to be able to provide the right service, to be able to um yeah, to find it also on our e-com uh uh channels. And that is really something that we're heavily investing in, and also in that area of yeah, our master data, there we're really um also looking at the necessary AI tools and having already a number of AI use cases in place.

SPEAKER_01 28:54

Exciting times for TVH. And then if we look a bit at the longer term, if we go back to the start of the conversation where you said we have a whole new strategy in place, a transformation program, what will there be the next step, let's say in in the coming one to two, three years?

SPEAKER_00 29:10

Yeah, I think in AI, yeah, looking three years ahead is always a bit very long term. I think what is most important is that we have a kind of adaptive strategy so that we know uh to always look at where is the biggest impact, that we can evaluate that regularly. And then I think for me, the most important is that we keep further developing a number of our use cases in strategic areas, but also uh that we really look at uh today already things that will change us in the future as we start building the foundations now to be able to scale when is the right moment.

SPEAKER_01 29:46

I think we need to invite you back here in uh one or two years because uh it sounds very exciting. Um, I think we can round up. Uh, thank you very much for joining today, Sophie. I think uh you were very transparent about the way of working, the strategy, some use cases, and I think it's very um inspiring for listeners working in organizations to see how a focus on transparency, collaboration, and so on can really um accelerate the development uh of AI. So uh thank you very much for sharing that.

SPEAKER_00 30:18

Um I was happy uh to do this as well because I think it's a nice way to share what we're doing uh and also to hear further feedback from other uh people.

SPEAKER_01 30:27

Yeah, again, transparency. Thank you very much. Bye bye.

SPEAKER_00 30:31

Thank you.