Join the Insider! Subscribe today to receive our weekly insights
David (00:00)
Welcome, you’re listening to the ITOT Insider podcast. I’m your host David. Subscribe to our blog or podcast to learn more about our work or go to itot.academy to discover our training. And if you follow our work, you know that it’s all about data and AI this year. ⁓ So instead of talking about technology, why don’t we talk about applying data, applying AI in our day-to-day job? And to do so, ⁓ yeah, we have invited Nathalie Rigouts.
She ⁓ was the head of data analytics at Borealis until a couple of weeks ago, and now she’s the new head of business applications and data and AI at Umicore. Nathalie, thank you for joining us. As always, why don’t you kick this episode off by introducing yourself.
Nathalie (00:42)
Hi, good evening.
Yes, happy to do so. as you said, I recently made quite a change after more than 20 years in the polymer industry. I moved to Umicore, which focuses on recycling. ⁓ Similar challenges, ⁓ similar contexts, but still getting used to the new position. When I left Borealis, ⁓ I was heading the data and analytics team. ⁓ We were…
building BI reports, were building AI solutions, we were responsible for the governance and the architecture. ⁓ And that all sounds maybe quite technical. And in fact, I am not a technical person at all. I have a finance background. ⁓ I worked in ⁓ consulting in finance for a couple of years and then joined my previous employer as a controller, in fact. But also as a controller, you heavily rely on accurate data. And already in that role,
David (01:30)
Yeah.
Nathalie (01:44)
Every month again, I was struggling a lot with getting all the correct actuals. And then of course, you have to make your forecast. There’s no predictive model for financial forecasting yet. think, unfortunately, that is still human work. ⁓ yes. Yeah, massive files. ⁓ And actually, that was sort of then how I got ⁓ into IT. I was then the business project lead for implementing
David (01:57)
And we get lots of excels I assume.
Nathalie (02:12)
a financial planning tool because at a certain point in time we were ready to discard the big Excel files and do it a little bit more professional. ⁓ And I also established the BI capability in Borealis many years ago already. And yeah, over time, ⁓ then it was more logical for the team and myself to join the colleagues in IT. And in IT, I was then responsible for BI.
I actually also made a very interesting sidestep to IT innovation. Now will remember, David, the times that everybody was talking about blockchain. Remember that? And, yes, and virtual reality, augmented reality, all very cool technologies. And then the clue of the matter was, of course, is there any value in this technology for our company being an industrial producer? ⁓ And I must say that my experience with that was that
David (02:53)
you did that as well?
Nathalie (03:12)
Very often the value was a little bit overrated, a little bit the same as what we see today maybe with Gen.ai. Let’s see how that story turns out.
David (03:21)
interesting one. And I definitely want to touch that, that the topic as well later in the later in our conversation, you come from finance, ⁓ it like, where is the, where is the step into manufacturing? How did how did that happen? How did you discover the pipes and reactors and
Nathalie (03:29)
Yes.
Well, already ⁓ when I joined as a controller, we were ⁓ educated about what are all the numbers about. ⁓ And we got to know the competitive landscape. We got to know the customers, which product do we ship from, which plant to which customer at which one of their sites, the whole logistics network. So again, lots of data ⁓ that needed to be gathered from many systems. ⁓
David (03:51)
Yeah. ⁓
Nathalie (04:11)
So already when I was leading the finance teams, ⁓ we were very, very close to the business. And then ⁓ in IT, it remains, I think, really, really important that you have a deep understanding of the business processes, because it’s about finding those golden nuggets. It’s not about implementing Microsoft Copilot. You’re not going to gain any sustainable advantage there. But if you can have a deep understanding of the processes in your company, be it industrial manufacturing.
David (04:40)
Mm-hmm.
Nathalie (04:40)
or anything
and you also have an understanding of what are the challenges and could a data-driven solution help? That’s when you start to create value.
David (04:52)
So I think one of the struggles many companies, from small to big, facing today, probably also in the last years, but now especially with the AI hype, is…
We’re all focusing on the best words. We all want to be data, we used to want to be data driven, now we all want to be AI first or Gen.AI first or co-pilot first or autonomous first or whatever. All hypes, all best words. So why don’t we, as a first topic, I would like to get your vision, your idea on…
Wherever you are at, I would say at whatever point you are as a company, like how do you start building data capabilities? How do you start building a data strategy? What would be a good starting point for the people who are listening?
Nathalie (05:49)
Mm-hmm.
Well, I’m a very pragmatic person and I’ve learned that that pays off ⁓ being pragmatic rather than coming with fancy concepts, often coming from overpaid consultancy companies. I have deep respect for the companies that we work with. Don’t get me wrong. But starting pragmatic, what does that mean? Well, like I said earlier,
David (06:01)
Mm-hmm.
Nathalie (06:14)
try to really understand the business. I recently started four weeks ago ⁓ in a new company and that’s the main question that I ask everybody that I meet. What is the particular business that you are working in about? What are the main challenges and are we sitting on a bunch of data and could the data-driven solution help? So even if you are low on resources and limited on infrastructure, there are always these elements that you can find where
Often people have been waiting for somebody to ask that question, to come along and help them, even with limited resources. So find the enthusiast, find the potential issue that you might be able to help solve, and there you go. ⁓ I always say at Borealis back in 2016, we started with one data scientist and a laptop. And that was really it. Python notebooks on a laptop, and we started. But you know what? Success sells itself.
David (06:49)
Yeah.
Nathalie (07:12)
when you start to show value even on small scale, people start to get interested. ⁓ What we often forget, we are typically quite modest being Belgian, we often forget to sell that success, to market it, to talk about it, to communicate it. And as it so happens, I’m pretty good at the talking part. So I would go everywhere and talk about small things that we did. And then…
David (07:29)
Yeah.
You
Nathalie (07:38)
you get some additional support and you can hire the second data scientist and you can ⁓ invest a little bit in bigger and more infrastructure that can handle more data. So it’s really, if you haven’t started yet or if you’re somewhere midway your journey, I would say really take it step by step. Try to find some really good use cases. Don’t forget to market them internally to communicate about them.
And in that way, try to gain budget support from your management to take the next steps, whatever you think that might be.
David (08:18)
but still there is this constant trade-off between ⁓ investing in foundational things, whatever that foundational thing might be, ⁓ versus investing in the customer-facing things or the user-facing things.
Nathalie (08:38)
I would not call that a trade-off, David, particularly. For me, they are all equally important parts of the puzzle. It’s actually also ⁓ a picture that I like to use in communication, that we need all the pieces of the puzzle. You need the infrastructure foundation. You need the data to start with, and you need a way to bring data from different sources together, and you need some sort of capability.
to then develop a machine learning model or a large language model or whatever it is. You also need people to do that, either internal or external. And at the same time, think about the people outside of your IT teams creating awareness for the responsible use of such solutions. That becomes more and more important in light of the recent legislation. That’s maybe also an interesting topic to touch on later because that’s definitely a new element in the game or a new piece of the puzzle.
⁓ And at the same time, especially when you start with Gen.AI type of solutions for which you have a big audience, don’t forget governance. Make sure that you determine the rules of the games. Are you going to allow chat GPT or not, et cetera, et cetera. ⁓ So try to determine the rules of the game. Try to also have a, as I said, success cells itself. So be prepared for lots of questions coming to you.
Can you do this? Can you help with that? So also be prepared with a process on how do you do that intake and that prioritization. It ideally is not first come first serve. Try to be value driven as much as possible. So try to think a little bit about processes and governance. It’s always the people, the processes, and the tools. think whatever you do in IT, even with AI, it’s still about people in your team, outside of your team in the company. It’s still about the tooling and it’s still about the processes and the governance.
And the main component of all of it around which all of those other pieces of the puzzle center, that’s the value case. How do you determine those golden nuggets? And my opinion is that you have to invest equally in all of them as you grow your maturity. Because to have a wonderful team and not a good idea of which use cases to work on or vice versa doesn’t really help. And to have a lot of work and no governance.
David (10:55)
Mm-hmm.
Nathalie (11:01)
is a risk for a bit of chaos. So, yeah.
David (11:03)
Yeah,
yeah, I get that. But this calls for, or I would say screens for some examples. Like, are there named or unnamed, anonymous or whatever your choice, but are there some typical, from a process manufacturing, whatever point of view, examples where you see like, okay, these are typical use cases we see quite often, or I see quite often.
and those are maybe easy to quantify, maybe easy to show value, maybe other use cases where you go like I also see these but it’s hard to… I don’t know.
Nathalie (11:44)
Yeah, definitely. think in a manufacturing context, we all want our plants to run smoothly. That’s good for the cost and for the safety aspect. ⁓ So if there are ⁓ pieces of equipment which fail often, I think the first thing to look at is anomaly detection and ⁓ predictive maintenance as a next step. In my experience, that is not an easy nut to crack. ⁓ Quite often, the critical equipment
David (11:53)
Mm-hmm.
No.
Nathalie (12:12)
you have already invested in a lot, so you don’t have a lot of failures. That means not a lot of data to feed a predictive model. ⁓ So yeah, that is easier said than done. ⁓ But I think if you have good support and maybe have a go with some synthetic data, there are ways to make progress there. But in my experience, that was a long journey. ⁓
David (12:16)
Hmm.
Nathalie (12:40)
I believe another important aspect when we are producing intermediate goods or finished products is the quality of our products. Typical use cases there is computer vision to monitor product quality. Means of course that you need to have some online cameras installed. ⁓ But that is definitely a use case which I would say is pretty mainstream. And if you have a sufficient amount of pictures,
David (12:54)
⁓ yeah.
Nathalie (13:09)
and they are annotated, meaning indicating this is good versus bad quality, then it is not that difficult to get such a solution running. So that’s, think, also a very good one to look at. And then typically, another one, when you want to ship your products to your customers, you may have a bit of a logistics spaghetti to solve what are the optimal routes. And that is also one where some AI optimization models, that’s not.
necessarily predictive models, but optimization models can also really help. So I would say those are the ones to have a look at if you haven’t done so yet. ⁓ Some are easier than others. ⁓ And that’s also ⁓ because I said you need pictures that are annotated or you need whatever other data points that are annotated. That reminds me of an experience we had
David (14:02)
Mm-hmm.
Nathalie (14:08)
quite early on when we started in the Borealis context, we were informed that there was a lot of good data available annotated. So we were very happy because that was a bit of the exception. Turns out the annotations were done by our colleagues in Finland in Finnish. It’s a beautiful language, but nobody outside of Finland speaks or understands Finnish. So there we were with our beautifully annotated data and there was no Deepl yet at that moment. So yeah.
Voila, there you go. ⁓ And that’s where then data governance, I mentioned all pieces of the puzzle, that’s when data governance comes in, because we had to go and tell those guys kindly ⁓ make your comments in English from now on. why? Well, because otherwise, us down the road in IT, we can’t really do anything with your data. So yeah.
David (14:44)
Yeah.
Yeah.
And this is also again, data quality related to not only the language, but probably also I would say in general, I did some work on maintenance orders, but that’s many, many years ago, but we try to extract meaningful information from maintenance, yeah, logs, technician logs basically, but…
Nathalie (15:06)
Correct.
David (15:27)
The variety was so wide. The way people annotate is so diverse and some write a whole booklet and others they just write okay or change or something like that. And then, yeah, you come to this point where you go like, how much time should I actually spending cleaning this?
Nathalie (15:48)
Yes, yes. And that is, think, where typically when you start, you will spend a lot of time collecting, first of all, and then cleaning the data before you can start to see if there’s any predictive power in there. But I think that’s sort of the struggle you have to go through. We experienced that struggle as well in Borealis. Quite soon when we had our second data scientist, we realized if you want to scale this operation, we need a data engineer. And that was then a new role which came up at that time.
David (16:15)
Yeah.
Nathalie (16:18)
By now everybody knows about the importance of such a role. But yeah, you learn along the way what you need and what your team and what your company needs.
David (16:29)
you have if you look back now what would you have done differently 10 years ago?
Nathalie (16:34)
That is a very good question. ⁓ I don’t really ⁓ know if it would have helped if 10 years ago I would have talked about the importance of data governance and compliance because I don’t think anybody was ready for it. So I think with hindsight, ⁓ we probably would have ⁓ done some things differently, but.
David (16:46)
Still at this point.
Nathalie (17:00)
As I said, it has really been a learning journey. And I think that’s also fine. So be kind to yourself, right? Be kind to yourself. You’re gonna make mistakes along this journey. You’re gonna invest in the wrong use cases, or you’re gonna, I can share another example of where it went wrong for us. And that’s then something that we changed immediately afterwards. We had a very nice model running back in 2017, 18, which helped to support hedging decisions. So it was a model that could predict whether
David (17:03)
Mm-hmm.
Nathalie (17:30)
commodity would go up or down and then of course if prices are going to go up you’re not going to you’re going to hedge because otherwise you need to buy ⁓ more expensive and if the price is predicted to go down you’re not going to hedge because you can probably buy cheap. And it actually worked very well and we had a very enthusiastic person ⁓ in the business and then that colleague decided to leave the company. The gentleman who replaced him
He didn’t believe in data-driven solutions at all. He said, I know this much better than your strange magical model. And he stopped using it. So our total investment in that model was down the drain. And it was a pity because it had really proven value. So after that experience, we learned that it is very important, first of all, not to have the one enthusiast. I know I said earlier on, start with that and you have to start somewhere, but try.
David (18:22)
Yeah.
Nathalie (18:24)
as soon as possible to find more enthusiasts, especially the boss of the enthusiast should also be in on it. And secondly, also try, and then I come a bit to change management. I don’t think I mentioned that already, but it’s also important. If possible, try to make sure that when you build a data-driven solution, whether it be a new dashboard or an AI solution, try to make sure that
David (18:29)
Mmm.
Nathalie (18:48)
that there’s no way around it anymore, that the process is changed in such a way that people cannot revert to the old way of working. And that’s a little bit of change management, but that will avoid that good solutions are then potentially discarded simply because people are not believers in data-driven solutions. So you learn those things the hard way. ⁓
David (18:55)
Mm-hmm.
Yeah, that’s,
That’s always a big risk. So when we were discussing the examples a couple of minutes ago, you mentioned predictive and vision, et cetera, et cetera, ⁓ and said for the one, the business case is easier than for the other, then I already thought like, let’s plug in Gen.ai into this discussion as well.
Chen.ai, obviously already years in the making but boomed since Chetch.pt was released ⁓ to the public, has been on the agenda in every company, I guess, in some way or form. Is in literally every pitch deck of technology vendors today.
Apparently you can’t sell a product anymore without a Gen.AI component. So from your perspective, where do you see potential value? Is there already potential value today? Will it be that or is it more for the years to come? What have you seen until now?
Nathalie (20:29)
Yeah, I think it’s a question that many people ask themselves, me as well. I was also on the weekend reading some articles in the newspaper about ⁓ so much investment going into AI, but are we seeing in fact any value? ⁓ Tried in my previous role to make the business case for Microsoft Copilot because 30 euro per user is not a small amount if you talk about a big company. And how much time can you actually save?
David (20:42)
Yeah.
Nathalie (20:58)
for a person during a day and are you then able to reduce your workforce? Typically, that’s not what happens. Typically, we say we want workplace efficiency. People can spend time on more value add activities. But what does that bring to the bottom line? Question mark. I do not have the answer to that question. I believe in the workplace efficiency. ⁓ Gen.ai is not, ⁓ or companies should not invest in it because of
the huge value it’s gonna bring. I honestly believe that companies have to invest in it because it will determine their attractiveness as an employer. Today, when we hire people or when we try to hire good people, they do not ask us, ⁓ can I work with Excel ⁓ or do you have a ticketing system that I can raise a ticket if I have an IT problem or all that.
David (21:39)
Yeah.
⁓ yeah, I can’t imagine so
people going like I really hope you’re using this type of ticketing system.
Nathalie (21:58)
So
people are taking a lot of stuff for granted, right? And I believe that the generation which is now in college and which is going to start looking for a job soon, they will take for granted because they use that. My kids who are in high school use this. They have grown up, so to say, with chat GPT or copilot or whatever it is. So they will take that for granted. So I think that sort of workplace efficiency tool
it just will become a scenic one-on, not for the value, but simply because otherwise, yeah, you will not be able to hire talent. ⁓ There are of course cases where you can ⁓ save significant time of people. There are the typical cases where people deal with lots of documents like in legal or like in research where they have to revise research papers or they have to revise patent applications, et cetera, et cetera.
I believe there you can save significant amount of time. Again, what will that bring to the bottom line? Not sure, but it will for sure ⁓ make the job more attractive. So I think it’s definitely worth investing in using the large language models that are available and tuning them to your needs and training them on your own documentation.
think it becomes really interesting when you try to do like pharma is doing and using ⁓ large language models trained on your specific IP to try and come up with new formulations for drugs, for instance. And if you can speed up that process, then you talk about a lot of euros or dollars, of course. That is maybe not relevant in every industry. I do not know. But that is, I think, one of the golden nuggets of gene AI.
where you take it away from the workplace efficiency type of applications and you really deploy it on your own intellectual property, on your own documents, on your own knowledge, and then it can really become a digital co-worker.
David (24:08)
And have you seen it in a real manufacturing context already, ⁓ used by operators, technicians or not yet?
Nathalie (24:17)
I have not yet seen a company where they are in that stage. think pharma is the most advanced in manufacturing. I’m not aware, but that doesn’t mean that it’s not there. ⁓ I do know of some investigations ongoing in my current company. ⁓ And I believe it is definitely ⁓ worthwhile investigating. Absolutely.
David (24:27)
Mm-hmm.
But yeah.
But I think
you’re right. that’s probably the point where we are at right now. The point where we need to just dive in to it, figure out where we can add some value, be cautious about what data we’re using, be cautious about how the output is being put to use. Just a while ago, I had a very interesting discussion with an AI company.
Nathalie (25:04)
Absolutely.
David (25:11)
where if we look back to typical machine learning models, and especially the ones which are really running in production, so the ones who are making recommendations and are kind of closed loop into the process. Basically, what we’ve always done is while we are closing the loop, we are closing the loop within very specific guardrails.
So the data flows into Scala DCS system, whatever, but then the DCS system is really going to validate that output, going like, is this deviation too big from my previous set point? Is this within certain limits? Do I need an operator to acknowledge the input? Yes or no? Et cetera, et cetera, et cetera. What we are experimenting now with Gen.AI is there are no guardrails.
Like we just, ask for an output and today the output is A and tomorrow the output is B. I guess to make better use of it, we need to understand better, like, okay, how can we make sure that these models, they give us a more or less predictable output.
Nathalie (26:19)
Yeah, and I think on the one hand, it’s about awareness creation with the users so that they know, we used to say, take two with the output, think about it, don’t take it at face value, double check it. Ideally, you check the references that are mentioned. If there’s no references mentioned, it’s for sure fishy. If you start to ⁓ train the large language models on your own documentation and information,
you will indeed need to invest quite a bit of time in the beginning to have your subject matter experts review the outputs and retrain the model. And you will need to go through a bunch of iterations, but that is a flywheel that does accelerate quite rapidly. So that’s definitely worth the investment. yeah, I believe that that’s where we can generate some value if you train these large language models on your own.
David (27:01)
Yeah. Yeah.
Nathalie (27:13)
contracts on your own, research documents on your own, whatever it is that is important for your company. ⁓ But that requires ⁓ time and investment of your subject matter experts for sure to make sure that people who are then not a subject matter expert can also trust the output. Absolutely.
David (27:32)
And well, AI in general, especially now, Gen.AI is also linked into the legislation. You already mentioned it. It’s something new. ⁓ think differences in regions, but especially in Europe, we see attempts to create some loss, some guardrails, some boundaries. ⁓
Should we treat in your opinion should we treat legislation as a problem or as an opportunity?
Nathalie (28:06)
Well, I’m a positive person by nature. ⁓ That’s one, but I honestly believe it is an opportunity. That’s at least how I chose to treat it. Let’s put it like that. Of course it entails work. There’s no doubt about that. You need to organize some projects or bring together some people ⁓ to do the necessary things. But I do believe it is a very strong lever towards
David (28:08)
you
Nathalie (28:33)
your upper management, the people who are going to give you budgets to do what you need to do. It is really ⁓ good, I think, because it also ⁓ allows you to work on all these different pieces of the puzzle. It asks that you train and make aware of the people in your company about the potential risks and make sure that they use the outcomes of models in a responsible way. It asks you to think about ethical considerations so you can
David (28:36)
Yeah.
Nathalie (29:02)
bring together some governance body if you don’t have that yet. It asks you to think about ⁓ careful prioritization, documentation, monitoring of models. So all those sort of technical safeguards or guardrails, if you don’t have them yet, it’s not a bad thing to invest in. So I would see this as an opportunity, of course, it is going to take time and effort, but I do believe ⁓ it is,
good to make that investment. They ask for ⁓ things that make sense in my humble opinion.
David (29:40)
I would like to end our discussion with ⁓ one final question. ⁓ Your take, or I’d say your advice, that’s a better word, your advice to people who want to start a data career. Like if you have somebody…
somebody in a finance job, example, who is today, who goes like, I also want the thing Nathalie did. So what would be your advice? Because what I often hear from people, example, younger people in my team, or they go like, David, it’s like so much. Like, where do we, where do I even start? ⁓ What would be your answer?
Nathalie (30:07)
Okay.
Well, it depends, I guess, a little bit on what you want to do. I’m not a technical person at all. I’m more of the advocate and the coordinator of everything and the girl who gets the budget so the experts can do their work. So you need those type of people. So if you’re good at that, if you’re good at the storytelling and the marketing and I see myself as a translator between IT and business. So if you’re good at that.
David (30:50)
course.
Mm-hmm.
Nathalie (30:58)
So lawyers who can explain things very well and make cases before court, maybe they can make a career switch to the sort of thing I’m doing. If you want to go for a more technical career and become a technical expert in a certain field, then I think you’re up for a lifelong journey of learning because the technology evolves so fast. ⁓ But a good basis of statistics never hurt anybody for that career.
David (31:19)
Hmm.
Nathalie (31:27)
And then, it’s going to be really lifelong learning and to stay up to date as much as possible. ⁓ And don’t forget, if you want to do something fancy with AI, please don’t forget about the data. The data is the foundation of AI. So don’t forget about data governance. That’s a very noble career if you want to become a data governance specialist. And if you want to convince business that it’s important to take ownership of the data so that the data
quality becomes better so that data is classified and can be securely used, etc. Data governance is so crucial for everything that we want to do. So I think a career in data can be very versatile and you can pivot from the one to the other if you want. I love it. I love the topic ⁓ because I really honestly believe that there’s so much value that we can create with data-driven solutions. And I always say keep it as simple as possible. We build beautiful
solutions with just simple scripting, scripting business rules and the business was happy and it helped them and nobody needed a fancy machine learning model. So keep it simple, be pragmatic ⁓ and then ⁓ it’s a wonderful area to work in.
David (32:32)
Yeah.
And I remember the healthy dose of statistics. ⁓ Which is absolutely true. It’s absolutely true. So for the ones who are thinking about why do I still need my maths? Well, there you have it. ⁓ All right. Thank you. That’s a wrap for another episode of the ITOT insider podcast. ⁓ Nathalie, thank you so much for sharing ⁓ your insights. And also of course, to our listeners for tuning in.
Nathalie (32:45)
Yes, yes.
Yes. Yes.
My pleasure.
David (33:11)
If you enjoyed the conversation, don’t forget to subscribe at itotinsider.com or go to itot.academy to discover our trainings. See you next time for more insights in bridging the gap between IT and OT. Until then, take care. Bye-bye.