The Data Strategy That Separates Global Winners From The Rest

Jason Hemingway: Welcome to In other words, the podcast from Phrase, where we speak with business leaders shaping how organizations grow, adapt, and connect with customers around the world. I’m your host, Jason Hemingway, and I’m CMO at Phrase. Today, we’re joined by Willem Koenders, who is a seasoned expert in data strategy and AI-driven transformation. Willem currently leads data strategy at ZS Associates, where he helps enterprises scale operations globally, integrating AI and automation to drive growth. Over his illustrious career, he’s worked with more than 100 organizations worldwide, designing strategies that go beyond technology and embedding data as a core driver of business value, resilience and customer relevance. Also with us today is

Jason Hemingway: Simone Bohnenberger, our Chief Product Officer at Phrase, bringing her perspective on how data and AI intersect with scalable content and customer engagement. With that, I’ll say welcome to you both.

Jason Hemingway: Thanks, Jason. Thanks for having us, I suppose.

Jason Hemingway: Thanks, Jason.

Jason Hemingway: Yeah, nice to see you both. So let’s start off with you, Willem. You’ve had an impressive career, leading data strategy from loads of different organizations across Europe, Asia and the U.S and Latin America. But if we sort of take a step back and we think about what drives your passion in the space, how did you get started, and why is data such a critical lever for business growth, in your opinion?

Willem Koenders: Yeah, for sure. And so I think like a lot of folks that I’ve at least encountered, you know, both, you know, at the companies I’ve worked at and the client organizations I’ve worked at, like, it wasn’t a predestined thing. Like, I do know there’s not necessarily a data master’s or something like that, where you did, that’s what I want to do. It’s kind of like you kind of roll into it because data is just such a big problem often, right, at a lot of organizations. And for me, it was no different. So I started out actually as a strategy consultant, like corporate strategy, looking at kind of the highest level business objectives. And it just, data became the biggest theme, right? Very quickly. And I had the good fortune, and now a decade back, it kind of dates me a bit, but you know, when, kind of, the first time the chief data officer, that role became a thing, right? It became, you know, an office that was great in different organizations. I had the good fortune to work with a great series of them, like going through the very first, you know, biggest organization in the world, to kind of help define those offices. What do you do? How do you kind of break through that, you know, these challenges that they had? And that’s kind of how I rolled into it. But it was definitely not an involuntary kind of, you know, experience. It was just a fantastic experience. You know, I mean, I can talk about this for the remainder of the hour, but I could; it brings a lot of things together, right? Data sits within technology. It’s owned by the business. It’s a deep critical driver, as you kind of alluded to in your question, to a lot of, you know, opening up new markets and enhancing product design, serving customers better. Like, none of that really works if you don’t have the right data. So let me just leave it at that. That’s kind of, it’s kind of where the passion comes from.

Willem Koenders: And,

Jason Hemingway: yeah.

Jason Hemingway: And thanks, Willem. We’ll talk to Simone a little bit because she spends a huge amount of time, don’t you, Simone, on looking at how data and AI underpin that kind of experience and global experiences. And what’s the biggest shift, Simone, that you’ve seen in how companies are thinking about data and AI as part of their kind of growth strategies?

Simone Bohnenberger-Rich: Yeah, that’s a really good question. I think if we take a step back and think a little bit about AI in particular and how far it’s come, I think in the last 20, 30 years, a lot of AI was used in the back office. It started with RBA, rules-based automation, then some simpler neural AI models that were rolled out, supervised, and unsupervised learning. But in the majority of time, these AI models were really deployed to drive internal efficiencies, to automate existing internal processes. What we’ve seen now with LLMs is that they increasingly shift the use of AI from a cost center into a revenue-driving force. So our space is the space of multilingual content. Where we see that is that increasingly LLMs are being used to actually generate that multilingual content and adapt it. So the power of LLMs and the big difference between LLMs and traditional AI approaches is really the rich context they absorb and the nuance they can produce. And that nuance really unlocks what I call hyper-personalized content. So engaging audiences in a very different way. And we know that if you leverage and if you use personalized content, you’ll really affect business metrics like engagement rates double. You have better conversion rates. You win market share. So if we think about our own personal experience, my email inbox is also spammed with emails. And usually one initial sift through is, is this actually targeted or tailored to me, or does this look like a generic message? And if it’s tailored, I think LLMs allow you to tailor towards a microscopic level, meaning, you know, if I live in London and it’s sunny, then send me something about sunglasses and not rain jackets, but also use my Gen X language. Don’t refer to it as loafers or softies, but call them sandals, right? So I understand that. I think we see this more and more that LLMs are being deployed for revenue-generating activities. And I think that’s the most exciting shift, that comes with challenges and obstacles, but that’s, I think, the biggest shift I’ve seen in our customer base.

Jason Hemingway: Yeah. So we’re talking here about, you know, on one hand, it’s quite a mature way of thinking about the data and AI. But Willem, you’ve seen, you know, probably over the course of your career, lots of different stages of data maturity across regions and companies. What would you describe to the listener as the most critical data government missteps you see that companies make when trying to sort of scale internationally or within their business? And probably a second question is, how can they avoid those missteps?

Willem Koenders: Yeah, maybe responding to the one thing in your question first, like in a maturity curve that you can see. There’s different curves that you can see. You’re right. There are regions because they tend to correlate with, let’s say, regulatory authorities that just happen to put pressure, you know, on companies to do certain things. Europe is a great example of that, like where Latin America, for example, where I spent quite some time, was just, you know, following several years later, kind of following the right curve, you know, learning the GDPR kind of, lessons there. You also see them by sectors like the financial sector, healthcare. They had to really invest in data governance very early on because they were just being looked at with a lot of attention. I think potentially, in line with Simone, what you were saying, I think a lot of sectors are now kind of voluntary. Some of them were not really voluntary. You had to, right? The regulatory authorities kind of told you to do things, but you see, like, the companies are voluntarily, the sectors are voluntarily kind of joining this pack now because they really, genuinely, finally, potentially, realize that you need it, right? You need to understand certain things about your data if you really want to open up those opportunities that Simone sketched out, right? If you want to language models or other AI, you know, that’s kind of, I think where you can see the maturity growing. In terms of missteps, maybe two things come to mind. One is actually not attached to any, you know, international regional component, but nonetheless, I think it’s the biggest mix of missteps still, which is to postpone the business case too long, right? Data governance is something that just has a connotation still. People don’t like it. It’s a red tape. It’s a cost center, things like that. You have to be very aggressive very early on, like, why are you doing this? Like, how are you quantifying the value? Don’t be shy. You can believe in it. You can show the passion for it. Like, don’t push that. Don’t go implement things and do that in year two. Like, I’ve seen it happen in the, you know, when the next funding cycle comes around, they’ll just have to dismantle some things. The second part that is actually, I think, relates a little bit more to your actual question around scaling international is, so there are a lot of things that work, you know, from a universal best practices perspective, but you got to take into account cultural context, right? So like depending on where you are, depending on, you know, regulatory requirements, like you need to have a framework, an implementation framework that, in one hand, it requires minimum consistency still across your organization, but it also allows you to work with folks locally, right? So again, sometimes that’s, you know, just adapting to, you know, habits and practices, sometimes to, you know, how the organization is shaped. I think that’s a really critical one because again, like data governance is not something where you’d like to create a big central team to do all this work on behalf of everybody else. It doesn’t work. It didn’t ever work. But you need to work with folks locally, right? So.

Jason Hemingway: Yeah. No, I completely. So, it’s kind of that localized model, but having kind of a central framework that everybody kind of needs to catch up with and be on board with. So just to bring you a little bit in here, Simone, and you talked about global companies and how they’re using some AI and the importance of it. But do you think there’s an underestimation of how important it is to align data governance with their, you know, how they deliver the content or those experiences you talked about? Those personalized experience across different markets?

Simone Bohnenberger-Rich: Yeah, that’s the first thought I had when Willem said, “Don’t push out the business case for data governance.” And when you’ve mentioned earlier the rise of the chief data officer, I think that’s actually the prerequisite to be successful with deploying an AI and moving it onto the revenue-generating side of the business. So, as an example, I’ve described this nicely hyper-personalized world that you can enable with LLMs. The reality, though, is that LLMs are trained on very generic data sets, namely the Internet. So they’re not tailored towards your customer base. They’re not tailored towards your tone of voice. They’re not tailored to anything. And to unlock that potential to reflect your tone of voice as an organization, to ingest your customer insights and understand how to target an audience and what the intent is, you need different types of data sets. And in this entire debate around LLMs, that’s something that’s often overlooked. They tend to work out of the box. You put a prompt in there, and you think it works. But the key to unlock this is really in the data. And that means data has to be curated better; there’s a data set, assets have to be clean, they can’t have duplication and noise because if you ingest that into an LLM, you just get very bad outputs, the same as AI has always operated. They also need to perform differently. They have to be dynamic. If you want to move as an organization into a world where Gen AI delivers content that allows you to target audiences, maybe even in real time, these data assets have to be dynamic, and they have to be updated and curated and optimized. And you cannot start with that right now. You should have started two years ago. And that’s really critical to unlock that; the other pieces, the type of data. So I think some organizations place big bets on knowledge graphs, which are very, very rich in context. They have relational, they have relationships, they are too big to put into an LLM, right? LLMs have fixed context windows. So there are different technical issues that come with that. So you really need to think through the entire, I would say, AI journey end-to-end, starting with the use case, starting with the data, and then you can talk about the LLM and the AI model. So I love that about Willem’s response. It’s really the data governance. It’s really the secret sauce.

Jason Hemingway: So if the data governance is the secret sauce, Willem, how have you, in your experience, got leaders to shift that mindset of governance being kind of that, you know, compliance burden to more of [00:12:00] that foundation of global expansion and AI innovation?

Willem Koenders: I think, for me, two things really come to mind. One is more of a way of telling the story, like crafting your narrative and explaining what these data foundations are and how they matter. And so what has really worked, at least for me personally, with a lot of client organizations is to kind of frame it up as an offensive capabilities and offensive objectives and a defensive set of capabilities, objectives. And you can actually put them on a page, like as a Venn diagram, where you say like, there are certain things from a defensive side, we’d love to make sure, like, we’d love to make sure that we know where our sensitive data is, you know, it’s classified, it’s protected, we don’t give access, if we shouldn’t do it, we are able to evidence compliance and privacy, things like that. That’s kind of on the defensive side. Then you have the offensive side is a little bit different, usually, depending on the sector, but it could be about, you know, serving customers better, better product design, you know, commercial use cases, who knows, but then you have this diagram that intersects in the middle, where there’s actually both of these sets of objectives require some minimum data foundations around, what data do we have? Where is it? How do we define it? What’s appropriate usage versus what’s not? And if you explain it that way, it tends to be pretty clear to people to understand that it really helps you to protect it, right? Because, again, if you understand it, you classify it, you can confidently lock it down, but you also can confidently democratize it. You can actually say, “That’s appropriate. Here’s where it goes.” There’s a whole different angle to this, you know, if you project data mesh, data products onto this, which kind of helps supercharge this. And then that’s kind of a narrative at a higher level. So that works the first time. After that, you’ve got to get specific, right? You actually have to get specific. So you have to actually talk about, I don’t know, a cross-sell algorithm that didn’t work, you know, or like account managers who are going into different accounts and don’t know that the other account manager was there last day, like, you have to get into specifics and kind of make that real. And so that’s at least how I’ve seen it.

Jason Hemingway: Yeah. I like that idea of defense and offensive strategy. That’s quite interesting. And Simone, kind of in that balancing that offense, defense, or rather a consistent data strategy and the need to kind of respect, you know, maybe regional differences or, you know, defense and offense in Willem’s case. What do you think comes into play, and how have you seen people approach that kind of problem?

Simone Bohnenberger-Rich: Yeah, I think that’s, I would say it’s a universal problem in our sector, organizations who operate globally and in different markets. I think Willem mentioned that there are different data requirements and privacy laws. So in Europe, we have GDPR and other things. There are questions around where your data centers are. On a different level, perhaps the way that our more sophisticated customers are doing that in the market, they are taking a very strategic approach to each of the different markets. So, for each of the different markets, they really define what they want to achieve, but also what the risk profile is of each market. Not each locale is of equal importance to you as a business. So if you’re an American business and you’re going abroad, maybe you target bigger markets first, like, I don’t know, Japan, China or Germany. They may have very different requirements each in terms of what type of personalization is required, what data customers want to see. So in Germany, German users are very interested in what personal data you digest, and you need to showcase that at any step along the chain, whereas in the U.S, you’re more interested that the information that you receive actually resonates. So take a very strategic approach, risk approach to each of your market geographies, and then wrap your data strategy and AI strategy around that through orchestration. I think that’s sort of the tiered layer cake is what we see.

Jason Hemingway: Yeah, interesting. So if we shift gears a little bit and focus a bit more down into that kind of AI area, Willem. When you think about AI and how it’s being kind of embedded into sort of global operations across all the different industries that you’ve worked in, and the one you work in now. How do companies ensure that that kind of AI and automation and all the things you can do complement what they’re doing, or rather what the human is maybe doing, rather than kind of disrupting it when you’re thinking about different markets and different dynamics and different regions, perhaps?

Willem Koenders: I think there’s a couple of common foundational enablers that you would hopefully logically think about, you know, they’re just managing the change journey here. So, like, actually doing the homework around, what are your business processes today? Understanding them, respecting also the expertise that’s in there today, like there’s a very interesting anecdote where Elon Musk, back in, and he talked about it, like where he was like, has these supply lines, there was AI and automation everywhere. And eventually, he actually was going end-to-end, he starts stripping it out, because it was too complicated, right? And so, like, it was easier sometimes to keep relying and actually more effective. And so I think there’s a part here around doing your homework on really understanding the business processes, and like working with the individuals in there to make sure you respect, you know, what they do, what good looks like, what’s realistic, what’s feasible. There are these common foundational enablers that I would think about: training, making sure that you don’t just pick the tool that, you know, has a certain connectivity with the technology landscape, but also the right kind of interface with the folks that are supposed to be using them. And also here, I mean, it’s very hard to kind of stay abstract without kind of going into like the specific challenges that a company seems to have. So in one example that I’m personally very excited about, I will [00:18:00] admit, both in our own company, as well as some clients, is kind of a CRM platform, for example. It’s like one of these platforms that they’ve been out there forever, and the data is always, well, let me not use an expletive here, but like, it’s not never great, right? Like, so how can you, with the language models, is that fantastic opportunity, for example, now where people are like, literally in a teams chat talking about this, you can actually invoke it and say, like, “Hey, create me this entry, update the status”, you know, that is a fantastic example where adoption has always been low, like, it’s always been bad. And then I think that’s like, I’m using it, that’s like, fantastic. And I think that finding those real examples that have meaningful change, and actually make life better for a lot of people using it, and kind of, in this case, that gives people more time to actually work with their clients, instead of worrying about some of this admin, is just a great example, and you’ll find more examples like that, and figure out what the two or three are that, in your company, is what I’d say.

Jason Hemingway: Yeah, and then maybe celebrate those wins more widely so more people kind of adopt that, right? Yeah. So, Simone, where do you, in your experience, we talk to a lot of global companies, and there’s a constant debate of balancing human and machine, where do you see the balance being right or wrong as people try and deliver those experiences consistently to global customers?

Simone Bohnenberger-Rich: That’s a good question. I think it really depends on the use case. So I’ve mentioned earlier that one of the great shifts that we are seeing is that AI is much more used for revenue-generating activities, and that means AI interfaces a lot more directly with customers, be that through a chatbot or automatically generated content. And that, of course, bears risks, for as long as AI models are built on probabilistic math or statistical regression, they will make mistakes. And these mistakes carry a potential high risk. I think here in London, Lloyds of London, our reinsurance market just launched a product in May to ensure companies against the legal and reputational and costs they incur if their AI chatbots makes mistakes, right? So it tells you this is a genuine risk, and it costs money. So the way to handle that is quite different. In a spectrum of extremes, you can argue that one way of doing that is to review everything that AI has produced by hand, which is an old-aged, a very outdated approach that we still sometimes find in our industry that AI translates. And then at the end is still a linguist or someone who ticks the box and says, “That looks right or wrong.” And that doesn’t scale. That does not move you into this world of content that gets generated in real time, is highly targeted because there are not enough humans to do that, but also it’s a waste of time. So at the other end of the spectrum, you could just run risk-free and say, “We just take the AI output as it is.” And I’ve mentioned earlier, that’s not a good recipe. So the balance lies in the middle, but really in terms of how much risk and reward you want to accept for your [00:21:00] different markets. So if it’s a high-value market where you are under big competitive pressure, that’s a market where you want to accept lower risk. And then the question is, how do you involve humans? A good approach is not to run everything through AI and then involve a human at the end, but rather, use humans at the beginning that define the strategy for each market, the risk profile you want to take, and then embed risk management tools at each step of this chain that produces that content, meaning, is this content fit for purpose, has solid evaluation frameworks, customization thresholds? So, all the good things that a good platform makes available that you automate within the market. So the balance really depends on what risk and reward you’re interested in, but it generally starts with a human expert at the beginning and not at the very end.

Jason Hemingway: Yeah. So, Willem, do you have anything to add on that, particularly about the kind of moves that large enterprises can make to kind of embed AI effectively, given that complexity of operation?

Willem Koenders: Yeah, there’s a multitude of things that kind of can come to mind if you think about all the things that are required to do, to make it work, from the AI technology to talent. I’m not necessarily the AI expert, so I’ll leave those to Simone and others. And then I’ll kind of go to my kind of, you know, that’s the area of interest mostly, which remains data, right? And so for me, there’s at least one or two different things that will come to mind for me. And one is to have a deliberate data strategy associated with it. Again, like, AI does not work without the data. It doesn’t matter if it’s external, internal. So, like, have and be deliberate about that. You don’t have to; you don’t need potentially an army of data stewards or things like that, but you do have to be deliberate about it. So I think that’s maybe one thing. And then the second thing is, especially when, as you mentioned, kind of the complexity of global operations or like the global processes that different companies can have, what really does tend to help is to kind of then take that data strategy that exists at a higher level and translate it to some kind of customizable, localizable concept like data ownership. Like, if I have a process, it could be onboarding, it could be a commercial process, it could be supply chain driven. So what does it mean for a process owner to be a data owner at the same time? And like help them understand what that means, right? You know, one of the most easy ways to say it is if you create it or change it, you own it, right? And so, like, if you do that like that, you now have the accountability. That’s not because I now say that because I helped you write a data strategy; there was always like that, you just maybe weren’t aware. And so that really does help in kind of helping them to potentially, you know, understand the impact of data, how to care for it as a true reusable asset across teams.

Jason Hemingway: Yeah. I think that’s great, isn’t it? It’s that idea that you’ve got to care for it, look after it, and keep constantly keeping an eye on it because, you know, it’s got to be organic in a sense, hasn’t it? So let’s back up a little bit and just a little bit of a fun segment that we have halfway through. And we all ask all our guests, Willem, you know, for an inbox confession. And the question is, what’s the one thing you could, if you could wish for anything that you could automate in your workday, what would it be?

Jason Hemingway: Well, I feel

Willem Koenders: embarrassed because I feel like this is automated. It can be automated, and I probably should have done it somehow, but I’m spending absolutely way too much time on my actual inbox. It’s absurd. There’s like 50, 60, 70 new ones every day, and, like, each night I’m spending time I could have spent, you know, in a creative way, a constructive way on different things, and I’m actually going through emails still, trying to sort out and mark them, it’s bizarre. So that is an actual inbox confession, I suppose. And I’d need to, I’d want to do something about that, let’s say this week.

Jason Hemingway: Yeah, good one. Simone, have you got anything else? I don’t think we’ve asked you that question before.

Simone Bohnenberger-Rich: Yeah, for me, it would be my Slack messages, similar to the inbox, because there are just too many channels and too many messages. And the nice thing is, it’s not just direct messages. You get tagged as well and on sorts of places. But my calendar, I would love some form of automation that optimizes my calendar and automatically finds slots when I try to set up meetings between seven stakeholders. And it’s just playing the calendar lottery.

Jason Hemingway: Yeah, you’d be surprised how many people actually say calendar. It’s quite a standard answer, I think, for lots of people. Let’s get back into the data side of things again, Willem, that’s why we’re here. And data transformation, you know, it’s core to kind of modern enterprise. And there’s a customer trust angle with data, isn’t there? So from your perspective, what ethical or trust considerations should, you know, data-driven leaders have in their mind when they’re thinking about marrying the data with that kind of customer first approach?

Jason Hemingway: Two things

Willem Koenders: for me. One, transparency of source, and I mean the right kind of transparency. So there’s, we were so used to it now. You signed up, you know, you order something. You have to be a member. You have to sign up. You get a thousand letters. Nobody reads them. I don’t. I probably should, but like you just click and like let’s keep going, right? And I feel like for some cases, maybe that’s okay, but I think if you’re trying to establish a meaningful relationship with a customer, like invest in making that very simple. Like, how do you actually explain what do I do with the data that I need from you? And be transparent about it. I think that’s one, it’s not actually even necessarily just me saying that. That’s actually a regulatory requirement in a lot of parts of the world. But in any case, like there are different ways to do it. And like use that as an actual opportunity to engage the customer rather than trying to quickly move in through, you know, one particular screen. That’s one. And second, I would try to explain the value of it, right? Like so, where, like be specific and like so, where and how is this going to help you? And so like I personally, I guess that there’s different personas out there, right? Where the balance between, and this correlates regionally. I think Simone, you were referring to Germany a little bit earlier. Like, where maybe the balance between how much you actually value inherently your privacy versus, you know, you get back for it. Make sure you explain it. Like, how does it work? And so for me personally, I still have that as well. I think I was natively raised right next to Germany in the Netherlands. I don’t know if it’s because of that, but I have that too. But at the same time, I don’t know about you guys, but if I’m engaged by a chatbot, an old one, I mean, I get frustrated beyond measure in a couple of seconds, right? And so, like, it is very, very clear to me, like I’m actually wanting it. It’s like I even get these big airlines, for example, it’s like, how can I not talk to a chatbot? I don’t even want to deal with a person. Sorry, it sounds awful. But like, just you know, help me this way. And so just these two things, you know, being transparent about what you do with the data and trying to figure out, explain, like how the value can really get to people. And those are kind of the two things I would focus on.

Jason Hemingway: Yeah. I think there’s a lot in that kind of, is the data being helpful to the customer? Are you actually using it to be helpful rather than maybe something like target them inappropriately or something like that, I guess? And Simone, you know, building that trust at scale. And how do you see kind of AI and the ethics of AI influencing that kind of brand loyalty and long-term customer relationships, but without stepping too far, you know? I mean, we’ve all had, you know, that feelings where something’s a bit too creepy, I guess, in a sense.

Simone Bohnenberger-Rich: Yeah, too creepy is good. I was laughing when Willem mentioned the airline chatbot because I sometimes try to hack it and go, like, give me a human, refer me to an agent, because I get so frustrated with them. Just get me some different type of sophisticated help. So I think brand loyalty and engagement, and data and data privacy are the different sides of the same coin. I think the way to drive brand engagement and loyalty is by addressing individuals and customers in a more personalized fashion. To do that, you need customer data. The big, I mentioned it earlier, the big differentiator in the game of LLMs is really to feed your in-house proprietary data as an organization into the LLM. And that’s what you know about your customers, that’s what you know what has worked in the past in terms of your demographics, that’s your proprietary data set that you’ve accumulated over years and years and years. Now, the challenge here is to make clear to customers what data you’re using, but also cleaning that data and being transparent across the end-to-end. And that’s not just transparency towards your customers, but also for us as software providers, transparency towards our customers when they use our platform. So, for example, we have a cleaning, a data cleaning and optimization tool to make data sets fit for purpose for the use of AI. And we make it very clear what we’re filtering out, how the data asset looked before and after, what happens to it, right? An audit trail. We have custom evaluation frameworks where customers can set their own risk threshold for AI evaluation frameworks to say, “Look, my risk appetite is higher or lower.” And we explain that we allow them to set it, right? So it’s control, but also transparency, pretty much along what we call the ML or AI pipeline end-to-end. And that’s critical.

Jason Hemingway: Yeah, no, I think that transparency and control, it’s kind of more on that trust, isn’t it? It’s that, how can you trust something you don’t really understand? It’s really, really difficult. So it’s opening it up. So let’s talk a little bit more about leadership in this area. And Willem, you’ve worked at large organizations, some smaller organizations, how’s your leadership style or what you focus on evolved, especially when thinking about data globally and managing those local or culturally diverse teams that you’ve come across?

Willem Koenders: Yeah, so I spent, maybe for context, right, I spent in my career a couple of years in Europe. I went six years, I think, United States, went to Latin America for two years, came back to the States, spent some time in Northern Africa, then came back to the States again, where I am today. And so I have been able to see, I think, a cross-section of kind of how these, how kind of maybe cultural factors play a role, you know, regionally, also across sectors, like a very fortunate position to have, I’ve kind of had the luxury to kind of see and experience these types of interactions. I don’t know that I’ve had a very deliberate kind of style change, but I definitely was like cognizant when getting into a new region to try to figure out, like how are things differently here, like what are the practices, the habits, even the words used. And I think for me, the strongest example of that was coming to Latin America, where I came without English, without Spanish. So that was something that I need to remedy right away. Two or three months, but that was one thing. Language actually is super critical itself, right? In a lot of different ways. So that’s maybe just one thing, just language itself. And then, actually, that can have very funny conceptual anecdotes because if you think about data governance, you know, you think about frameworks, ownership, accountability, et cetera. And so there’s a very critical difference in governance between somebody who’s accountable and somebody who’s responsible. But in Spanish, there’s no different word for the two. It’s very interesting. So how do you even explain? So that is very odd, maybe a little, you know, anecdote, but like you kind of run into some of these things that you can eventually deal with. But more, I think importantly there was just like trust, right? Relationship basis. Like folks there don’t buy. Their trust is built up in years, like the United States is way more transactional, for example. And so it took like time to, it was only COVID like that that helped us actually scale the practice we were building at the time, because everybody had to be online. And so, you know, that was actually one of the opening doors. I think in terms of leadership style, like you incorporate these things, like you choose your words carefully. In the U.S, you got to get to the point within two, three minutes, or yeah, that’s it, if you even get to it at all. Like, there, don’t even talk about it. Don’t ask about, you know, some of the, as a general, almost general rule, don’t go for the sensitive bits right away. It takes time for people to open up. And so I think that those are, if you think about communication and kicking off big programs, I feel like have been some of the tactical ways that you can tailor how you go about it.

Jason Hemingway: Yeah. And Simone, just to ask a similar question to you, how have you sort of seen, you know, businesses or leaders navigate that complexity and scale and still keep agile as they kind of do it rather, you know, trust takes time to build, but you still need to remain agile. And you still need to remain aligned to the core brand and the brand promise. So how do you sort of see businesses, you know, aligned to that?

Simone Bohnenberger-Rich: Yeah, I think for me, the question is not so much the people management side, although that’s important, it’s more the tech side and how you stay agile in the world of AI that’s fast moving. I think one of the most important rules in this space and what some of our customers are doing very successfully is to constantly innovate and focus on innovation. And that means find a playing ground where you can test and iterate, for example, how content performs, experiment with content. But also don’t treat AI as a feature, treat it as a use case. It’s not a feature in isolation. And Willem drove the point home really nicely. It’s about data governance and data is the foundation for AI. And so take a holistic perspective on the problem you want to solve, have a hypothesis, how best to solve it and then use data and AI together and iterate quickly. So I think that’s important, but also the understanding that it’s not a magic bullet. I think we’re now ending up a little bit more in the valley of reality in the last one and a half years. There was this hype about LLMs. But if you look at LLMs right now, we haven’t seen any major improvements in the last six months or so. At least my inbox didn’t blow up because someone said, “Whoa, mind-boggling something happened that I’ve never seen before.” So I think we’re coming a little bit down from the hype and the realization that it can’t do everything. So be clear that it’s a probabilistic tool you’re playing with, and that means it will make mistakes and be prepared for these mistakes and have a bit of a more of a risk and reward attitude towards it.

Jason Hemingway: So, that’s interesting. We’re talking about focus on innovation and things like that. Willem, in the sphere of kind of data and AI, I’m coupling the two concepts together, perhaps they’re honestly. But if you think about that, do you see any or if you look in your crystal ball, can you see anything in the future that’s there that’s going to reshape how things are happening and perhaps how our leaders might prepare for that future?

Willem Koenders: So I’m going to, again, kind of maybe, you know, interpret that in my kind of data context mostly. So we made the point earlier about, you know, you want AI, but all the kind of, these advanced different ways of doing things, you know, more customized, more automated. But like I wanted to turn it around as well. And so like there’s not just data that requires AI, like there’s a very interesting kind of niche area out there where AI is being applied to data actually right so you kind of have different AI capabilities automations around observing your landscape you’re creating your definitions doing a ton a ton a ton of work that you would have to do you know manually before better today automate it in a different way it’s like, I mean, I remember right spending a month on definitions for big data sets and they’re there now in like literally a flip of a second and like I would have to admit that they’re actually better, like I wouldn’t be able to do it any better myself like you know having done it many times and so, there’s just this one example of that and so, that is actually one, so if data is truly critical to your success like you harnessing how you do exactly this is going to make a huge difference in the future like if you can define, because it’s not, you know, about having nice technical definitions in a catalog, it’s really about does that drive, for example, a true customer 360 understanding. I understand where they are, what they want, how they engage with me, who’d engage with them, or what product, you know, depends again on the company. It’s like that depends on you being able to artificially, you know, automatically manage this stuff at scale, right? And so, like I feel that, for me, the companies who are able to harness that are going to be successful.

Jason Hemingway: Yeah, interesting. And Simone, to sort of give you the same question, is that what can you see? You said that there’s been little innovation over the last few months, but what could you see in the future if you also took out your crystal ball and had a look?

Simone Bohnenberger-Rich: So I think we will see more innovation. I was a bit facetious there earlier in the space of LLM, but I think it comes from a different place. I think the key [00:38:00] quantum leaps we’ve seen just came from the fact that LLMs have more computational power and can churn through more data that’s easily available on the web. And we’ve done that now. The internet has been digested. So the next leap forward doesn’t come anymore from big data sets; it will come from small, highly curated specialist data. LLMs will become smaller. They will become more specialists. So I think what we will see is a much bigger specialization and proliferation of smaller, large language models or smaller language models that are lighter in production, that are more controllable, they’re more domain-specific and not as generic initially. I think that’s sort of the trend I see right now. And then in our industry in particular, a bigger appetite towards risk-taking. So our industry is very, really hooked up on this concept of quality. A translation has to be high-quality; it has to be the best quality possible that resonates with the end user. And that is okay. But sometimes the quality isn’t worth the return, right? So rather than thinking about quality, think about the risk you’re willing to take and the return you get for it. And then that will unleash a little bit more willingness to automate in the first place. Because if you’re hung up on quality, you will always insist that two linguists debate whether it’s the best piece of translation or not.

Jason Hemingway: Yeah, is it that adage of don’t let perfection be the enemy of good, or something along those lines? Anyway, what a great way to finish the set of questions. And I’ve just got three quickfire questions, mainly at you, Willem, because Simone gets these questions all the time on a daily basis from me. But anyway, if you were going to describe global growth in one word, what would it be?

Jason Hemingway: I would go with digital.

Jason Hemingway: Okay, excellent. And then, if you had a book or podcast you would recommend to the listener?

Jason Hemingway: I would go with

Willem Koenders: a book called

Willem Koenders: Thinking Fast and Slow by Daniel Kahneman. It’s not a new one. I just rediscovered it. And it’s just, I think that’s a longer answer, but like, it just for me, just made me realize that we’re so focused on. I am too, right? The data itself, its supply, where is it? How do I classify the quality? We spend almost no time at all on how we make decisions. And so I feel that’s an area that we need to do a lot more.

Jason Hemingway: Yeah, excellent book. I can recommend it as well. One of the first texts I was reading, you know, a few years ago, when I started looking at customer experience, because there’s elements of that in there as well. And then lastly, just other than Kahneman himself, is there anyone you think we should speak to next?

Jason Hemingway: That was my answer. So I guess I have to come back to you then.

Jason Hemingway: No, that’s fine. That’s fine. I thought I preempted it, but excellent, excellent. Well, look, a fantastic discussion. And thanks for your insight, Willem. And thank you, Simone, for joining us today. And what an enlightening discussion. I’d just like to say thanks and speak to you both again soon.

Jason Hemingway: Thank you.

Jason Hemingway: Willem, Simone, thank you both for highlighting how data and AI, when built into the foundation of a business, can really drive meaningful growth and scale and actually strengthen those customer experiences and customer relevance. And with that, that’s it for this episode of In other words, the podcast from Phrase. I’m Jason Hemingway. I was joined by Simone Bonenberger. And a big thank you to Willem Koenders for sharing how leaders can design data strategies that are built to scale, ready for AI, and are grounded in that idea of business impact. So if you enjoyed today’s episode, be sure to subscribe to In other words on Spotify, Apple Podcasts, or indeed your favorite podcast platform. You can also find more conversations on leadership, growth and what it takes to really scale globally at Phrase.com. Thanks for listening and see you next time.