Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Data Analytics Chat

Data Analytics Chat artwork

Nestlé’s Chief Data & AI Officer on Scaling AI

Oct 20, 202649 minAudio

In this episode of Data Analytics Chat, we welcome Jessica Matheron, Chief Data & AI Officer at Nestlé. Jessica brings a rare mix of experience to the role, from a French engineering and mathematics background and a Harvard MBA to a decade trading structured credit on Wall Street, followed by roles across fintech, ad tech and e-commerce before joining Nestlé. Jessica shares how she took a data team that was seen as a service centre, built on Excel and Power BI and reliant on external vendors, and rebuilt it into an internal department able to deliver GenAI at scale. The discussion examines what was really slowing projects down, why the bottleneck was upstream in the data rather than the data science, and how a single consumer data foundation on Segment and Snowflake made it possible to engage tens of millions of consumers across Europe. The data topic explores the three flavours of GenAI, from everyday copilots to coding copilots that more than doubled productivity and the AI agents now in production. Jessica also covers winning trust from the CEO down, why organisational change and technology can't be separated in a 270,000-person company, how to build a business case before measuring value, and why redesigning end-to-end workflows around AI is the biggest opportunity ahead. 01:20 Introduction to Data Analytics Chat 02:01 Jessica's Background and Career Journey 04:05 When Things Were at Their Messiest at Nestlé 05:59 How It Was Affecting the Business 07:09 The Turning Point for Change 10:51 Fixing the Data Before GenAI 12:48 Where GenAI First Created Real Value 16:41 The Moment GenAI Proved It Worked 17:55 What Had to Change in Data, Technology and Operating Model 21:23 Moving From a Promising Use Case to Scale 24:19 Winning Trust Across the Organisation 26:57 Technology vs Changing the Organisation 31:50 How AI Helps Engage Tens of Millions of Consumers 34:08 Measuring Whether GenAI Creates Value 36:50 Surprises of Operating at Scale 40:01 What She Got Wrong and Would Do Differently 42:11 Where Enterprises Should Focus First 45:24 The Biggest Opportunity Ahead Connect with our guest: Jessica Matheron: https://www.linkedin.com/in/jmatheron/ Thank you for listening!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Jessica Matheronif you plug AI onto bad data, you're gonna get a lot of messy results at scale. We've seen more than 2X productivity increase on the pure coding side of things. we were able to reduce the time from the... on the audience building from 10 days to 10 minutes, don't listen to the sycophants internally that are gonna say,"Yes, of course, everything's ready." four people with AI are superior to many people with AI, and that is vastly superior to many people without AI.

Jessica MatheronThe new AI workflows should not mirror the past ones. they really need to be redesigned with AI at the heart of them. seen the movie The Big Short. because actually Ryan Gosling in that movie plays someone who in real life is named, Greg Lippmann. And, Greg Lippmann was actually my boss at the hedge fund for several years

ben parkerHello, everybody. I'm your host, Ben Parker, and welcome back to this week's episode of Data Analytics Chat. Today, we're exploring a challenge many organizations are facing right now. How do you move gen AI from experimentation into something that creates value and scales across the business? We're going to unpack what that journey looks like, the situation. We've had to look across data, technology, operating model, and what it takes to scale successfully.

ben parkerJoining me today is Jessica Mathuram, Chief Data and AI Officer at Nestle. So welcome, Jessica.

Jessica MatheronHi Ben

ben parkerSo I guess just go, before we dive in, do you wanna give the listeners a quick introduction to who you are?

Jessica MatheronOf course. So I'm Jessica Matron, and I'm Chief Data and AI Officer at Nestle Zone Europe. And my background is a little bit of a mix of things. So I've been back and forth between the US and France my, my entire life. My mom's American, my father's French, actually. And I have a very traditional French background, so really high on mathematics and then engineering. But then I rounded it off with an MBA at Harvard, and then I went back to statistics actually a few years later to learn data science.

Jessica MatheronAnd so that's on the academic side, and on the business side, I have also quite a surprising, I think, path for someone at Nestle. So I was actually a trader in structured credit for about a decade, so between Wall Street, Boston, Paris. And after the statistics degree, I moved to tech. So fintech, pharma tech, ad tech, e-commerce, and finally landing in food in Paris at first, and then Zone Europe after that.

Jessica MatheronAnd so maybe I'm... If I I maybe I could share a little bit of a, an anecdote for those listeners that you have, which have seen the movie The Big Short. Because actually the so Ryan Gosling in that movie plays someone who in real life is named Greg Lippmann. And Greg Lippmann was actually my boss at the hedge fund for several years, so that's maybe something that your audience will enjoy.

ben parkerAmazing. And oh yeah, de- definitely, it's, it the more and more people I speak to, it's just fascinating everyone's come from different backgrounds, and I think that's what data and AI, you can come from any background now. It's, there's so much opportunity and growth.

Jessica MatheronThere's a lot of back and forth, I think, with the skills. So the skills that you need in finance are actually quite similar to those that you need in data in tech today. So if you think of math and coding, it's at this essence of both these fields

ben parkerYeah, math's the found- it's the foundation, isn't it, really? I think it's sort of businesses need, like it's having that foundations is critical So then I guess take us back to the beginning. What did the situation look like when things were at their messiest at Nestlé?

Jessica MatheronOh my goodness, you're taking me back quite a bit. So let's see. There's, there was really three dimensions to to this question. And so I would start probably with people. So people are always at the heart of everything, and when I joined Nestle, the talent profile of the department was heavily skewed towards master data management and dashboarding. And I don't think there's anything wrong with that, it's just that it wasn't the distribution of talent that we needed to get to the next level.

Jessica MatheronWhich brings me to the stack, and on the stack, which is the second I mentioned here it was mostly Azure, but a lot of Excels and a lot of Power BIs, which honestly matched the people using them. And then the last piece was the ecosystem. And so I had never been in such a large organization before, and also I had spent most of my career in the US, and so that was a little bit of a... It was actually a quite a massive change because the, there were a lot of administrative hurdles to make changes in people, in stack, in project management.

Jessica MatheronSo that was the third dimension, really hard. And I think it was also internalized, so there was a perception of the data team as a service center. And the culture, it wasn't really a data culture not at all a tech culture. And there was this perception that it would be tricky to be anything else. There was a desire, but it wasn't perceived as being possible. So that was quite a challenge coming in.

ben parkerSo I guess was, how is it affecting customers or the business?

Jessica MatheronSo basically they had very long term projects with the business, but it was mostly deployed externally through vendors. And it was just sitting down with the business and explaining how to use tools rather than listening to what they really needed and building with them. So it had started to change, but obviously when you deploy with external vendors, you don't have as much control.

Jessica MatheronSo even if you might understand what the business wanted it wasn't it wasn't that easy. And so I was hired to change that really. And we started with with something that was very specific and that's also what I recommend honestly is not trying to change everything at once, but actually starting with something that is highly visible and important to the business and working your way back into the tech. And I can go into detail if you'd like on, on, on that first project,

ben parkerwe'll go into it. I'm just... What was the turning point that made the business realize that you need to change?

Jessica MatheronYeah. So that turning point happened before I joined. So it was really the head of the digital department in, in France who hired me to change that. And I knew it would be hard, but it was... That was really what I was up against. And as I said I looked at the project that was under strain. Their main project was marketing mix modeling deployment at scale, and it was really important to the business that it worked because they were trying to optimize their spend in marketing.

Jessica MatheronAnd otherwise it's, it's just opinions, but based off of very light facts or these facts were being interpreted by agencies, which obviously have their own agenda. So it was really critical to the business, in fact, across many categories pet care and coffee and food and water, just to name a few so that your audience will understand what we're dealing with. And in each one of those categories, we were... There are many brands. Obviously, Nestle is a huge conglomerate, and they needed they needed to have those insights.

Jessica MatheronThe thing was it... The process wasn't working really well, and I don't think it was necessarily the the platform, so the the dashboards that were being presented. It was more the process itself Of getting the insights, which was really painful, taking, nine months oftentimes and involving hundreds of stakeholders, and really a very tedious experience for everyone.

Jessica MatheronAnd so what I came in and did is I realized that it wasn't so much the, what we call data science, but here it's more econometrics with a lot... a very heavy dose of marketing experience. I realized that wasn't where the bottleneck was. It was really upstream with the data. So the data was coming in wrong from our partners the the panelists like Nielsen and Circana and also from the marketing standpoint, so from our media agencies.

Jessica MatheronAnd that process of verifying the data was taking months, and it was extremely tedious. And then there were mistakes made at multiple levels, including inside of the team. And so by the time it reached actually the the people doing the data science it was already not going in a very good direction. So I came in, I hired some data engineers to help out, and we restructured all of that process, trying to make it airtight everywhere we could.

Jessica MatheronNow, we didn't have control over the entire chain. Obviously, our suppliers, our data suppliers are, very instrumental here. But everywhere we could, we started making those those changes, and w- controls being much more algorithmic. And then something really important happened, is the head of the digital department in France got promoted to the zone board. So he became the head of digital for Europe plus Turkey plus Israel, and that really made a huge change because I wasn't...

Jessica Matheronit wasn't like I was doing, one marginal change at a time, one hiring difference. It was really the ability to create a blueprint and hire an entire department from scratch at the central level. And that's where I led next and and it really had a big impact

ben parkerYes. You, as you mentioned you started to realize the data was not quite right. Was that when you had implemented gen AI or was it before?

Jessica MatheronNo, that was really before gen AI. So I was hired in 2023, so it was really, if you rem- remember the very beginning of gen AI, and the results of gen AI weren't that good at the time. So this was really more traditional more traditional machine learning or data science. But the data that powers gen AI now is the same as that, that powered these th- that project then. Today you would have a talk to your data, for example, on top of on top of those platforms. But the platform still needs to exist.

Jessica MatheronData is not structured if it's not modeled well, actually, and then your queries aren't going to work, the agent isn't gonna find the information it needs. So the basics are always the same, it just, you just change the surface of interaction with your clients or your internal clients.

ben parkerYeah. No. The only reason I ask is'cause there's a lot more businesses I think are finding out that, yeah, they need good data to get good results with Gen AI. So it's-- I think a lot more now is-- I think a lot more emphasis will be focused on like data foundations and in the near future'cause it just needs to be g- you just need good quality data now

Jessica MatheronYeah, I think it's always been that way, but it's just that there's a lot more visibility around those problems today. Because if you plug AI onto bad data, you're gonna get a lot of messy results at scale. And so the business is gonna come and see you far more than it used to. Because you, you used to have a lot more control with dashboards. You would see the errors, and you would just surface very few metrics. Whereas today, you can't control what the business is going to ask and to talk to your data interfaces, right?

Jessica MatheronIt's a lot easier to get it wrong than before. Data teams are focusing a lot more on the data foundations

ben parkerCool. So where did gen AI first start to create real value rather than just experimentation?

Jessica MatheronYeah. So I think it's useful probably for the audience that I disambiguate a little bit what we mean by GenAI into a lot of different sub-segments of GenAI. So the first thing that your audience would be familiar with is what I call everyday GenAI, and it's all these versions of Copilot that the audience has been familiar f- with, and also just at a personal level. And we started implementing that in 2024 at the company, and honestly, my team was not involved in that deployment because it's really a global IT deployment.

Jessica MatheronIt's just deploying software, and we were working with Microsoft Copilot. And so that deployment went well. Initially, the results were lackluster and increasingly, gradually through time, they improved. I think today it, it provides some value. It's not incredible, but it's g- I think it's a good to have. Then the second version of GenAI is coding copilots. And so your audience may be familiar with Claude Code and Codex and GitHub Copilot.

Jessica MatheronAnd that got implemented in 2025, and I took far more of a leading role there, in fact, trying to lead from the front. So I... even though I'm not supposed to code anymore, I, I got a license. I started using it. Initially, the, the results were interesting but not exceptional, but the progress that has been made with co- with coding copilots is nothing short of amazing. And so I really pushed the team, not just the data engineering team, which of course adopted quickly, but I actually pushed the entire team, so data governance and product information product o- owners and project managers to really adopt these tools as well.

Jessica MatheronAnd what we've invested in here, and this is really thanks to the data science AI team and the data engineering team, were system prompts to make it work, not just at an individual level but at a team level, and that's where you really see benefits. And we've seen over even just, I would say, less than a year, we've seen more than 2X productivity increase on the pure coding side of things. If you're coding alone, you will see far more increases in productivity, but we need to work as one team.

Jessica MatheronAnd so you're always gonna be a little bit slowed down by the team, but you will go further. So that's something to keep in mind. And then so the, the version of GenAI that we're working on in 2026 has been really AI agents, so that's another flavor and we focused these agents in a few areas where we saw value just, upfront. So for us it's been really on, on audience activation, so really with the consumer data and marketing. And we... I know we'll talk about that a little bit later, but so that, that was one of the very important things that we were focusing on this year.

Jessica MatheronAnd then also in a, maybe a little bit less of a sexy topic, but in data governance. So AI agents for the service of increasing our data quality. As we mentioned just earlier, the data foundations are super important. The quality of the data is super important. And if we can deploy AI agents that help us surface the issues and correct them, that's also been of huge value. And then I think as we're looking into 2027, where I want to take the department is really on agentic AI.

Jessica MatheronSo not just having a suite of AI agents with a lot of deterministic handoffs, but actually having those AI agents interact in more autonomous ways

ben parkerOkay. I guess just were you hesitant at first with GenAI, like the power it can do? And I guess what... And then when was, like, the first use case when you thought,"Yeah, wow, this is actually working. This is gonna change things"?

Jessica MatheronI, I think for me, the coding copilots were really instrumental in making me realize that something was changing massively. It went from nice auto-completes and cute answers to simple questions, to really being f- a game changer for someone who is not a developer every day, but actually trying to get something done, trying to get a proof of concept up and running.

Jessica MatheronFrom my perspective, this was really when I saw that change. But in terms of moving beyond the productivity simply of the tech teams and having business impact I think AI agents are... They're already important for those that get to production, obviously. And I see that only increasing over time

ben parkerBrilliant. And then you, so you mentioned obviously the technology had to change when you started i- implementing this. Was there... was it more... I guess what had to change? Was it more the data, the technology, or operating model? What do you think was the most important change?

Jessica MatheronYeah, so I think your audience is probably gonna laugh, but we were missing data. It was... the beginning to make a, the change possible was actually sourcing the data. So we went looking for it, and then of course structuring it into foundations. Stepping away from a mountain of chaos to a space that was small enough to clean up. And so we did that on the consumer data and I know we'll go into that a little bit later, but it was really lots of little data sets, Excels everywhere, in every category of every country in the European scope.

Jessica MatheronSo that was just getting all of that data into one place with one data model ensuring that the... we were using a new stack so we, we partnered with Segment Twilio, and then Snowflake on the data warehouse. And then, plugged in a bunch of, a, a bunch of sources and outputs. But just getting that place cleaned up was the first thing to do. And then we had to change a little bit of our ways of working around data projects or data products, if you will.

Jessica MatheronI think the whole process of having external vendors and it was very long back and forth, lack of control, lack of understanding that needed to change. And so that's where having worked in tech in the US really helped out in terms of, changing the way we operated. And and then we, we partnered with marketing. So we started discussing where their problems were, and in particular and we can dive deeper in that, on the compliance side.

Jessica MatheronAs in Europe, GDPR is hugely important from a regulation standpoint. And making sure we address that head on, from the start so that it wouldn't be so much of an issue because the stack wasn't aligned with those needs. I think that was really important

ben parkerYeah. But yeah, no, on your point on the data, it's been this... I speak to a lot of people about it. It's whether it's sourcing the data or actually getting access to it. Some companies, sometimes-- I spoke to leaders and they said it take, can take four to five weeks just to get access to their data sometimes, which,

Jessica MatheronAbsolutely. Absolutely. So that, that also comes from a lot of large companies have SAP, and so the access rights are controlled through that, o- older stack. And yes the- you get tickets, and the tickets go, three times around the world to other countries before, you finally get access to something. So yeah, it's those are the processes around the tools that... And the whole thing needs to change

ben parkerYeah. Yeah. That, that's just that could just be frustrating because you obviously you want, you wanna get going, don't you? You wanna get delivering and then obviously it's out of your reach, isn't it, really?

Jessica MatheronI can tell you coming from tech in the US and then being at a larger company in Europe and facing these technologies that I had never seen before that's It certainly, it changes your daily environment for sure

ben parkerSo then how did you move from like a promising use case into something that could actually scale across the organization? 'Cause look, scaling is the challenge, isn't it? Like businesses, again, there's so much... Again, it's the process. Like there's so much going on, isn't there? So how do you get around that?

Jessica MatheronYeah. So as I mentioned before I think it was really key to start on an area of data where things were structured properly. And wh- where scaling was possible by design. Okay? So that's why we chose the consumer data, because I had been at the very beginning of it, and so it was, it was started as a proper data project. And then the, the second thing was really the partnering. So partnering with the teams that would actually be internal clients.

Jessica MatheronSo I think this is... This will be familiar maybe to some of the people in your audience who are not building AI on top of products for consumers, but actually on top of products that are internal. I think you need to resist the desire to have this idea and just deploy what you have in mind. E- even if it's your clients are internal, it's really key to sit down with them to understand how they will use the product that you're building, and to really provide them with intermediate steps.

Jessica MatheronSo we went really quickly in an agile way. We built what we thought they wanted, but then we put it in their hands, and we got their feedback, and we quickly redesigned things according to that feedback. We didn't just, move ahead with what we had in mind internally, which was more a- agentic only. And in fact, some of the things that they needed were more simple, just algorithmic, deterministic.

Jessica MatheronAnd still, that was something that made it work for them, and therefore made it made it possible to scale in terms of the number of users. If the support teams, and I'm going to be specific here, if the sup- support teams are needing to stitch information across multiple tools, then the process will not scale even if the tech does, right? So we needed to see what they were using from other platforms and make sure that information was inside of what we were building, the web app we were building for them.

Jessica MatheronAnd when we did, it was a huge success. So we thought our internal client was marketing, and we realized, actually, we had more clients than we thought. It was also the support teams that needed to adopt the product. So we, we went r- back to the drawing board, and we made sure that it would scale for people and not just the tech

ben parkerI guess how do you go about winning trust? 'Cause that's a big challenge for firms. Did you start small and sort of stack wins, or how do you go about it, Jessica?

Jessica MatheronSo it really depends on which trust, right? So the, the trust from the CEO is about, taking charge of the projects that matter to him, and making sure we have a steady stream of small successes around those topics. And, it's never gonna go as fast as, as he would like. But making sure he's informed of that progress, and also informed of the things that are holding us back. And he has been incredibly supportive and pushing the organization to get us what we needed, both from a resource standpoint, but also just from, agreements, an internal agreement standpoint.

Jessica MatheronSo that's, how you get trust from leadership. And then on the other side of the spectrum, I think we got a lot of trust from junior people in the organization, people in s- in shared services, by simply being good listeners. They hadn't had the experience of their feedback being heard, and so we made an effort to show them that we were going to proceed differently and really taking that feedback into account. And it's been incredible because now they are the strongest supporters of what we're doing, and we definitely need them to use the platform every day and to continue to be a driving force behind it.

Jessica MatheronSo a little bit of a different different perspective depending on who the stakeholder is, but trust is important everywhere

ben parkerYep. No, definitely. And it's obviously it's a hurdle for many. And I, now I think now it's just becoming more and more important. The softer skills are so critical now, like the communication aspect is just so important'cause it's so much going on now with businesses. They're getting flipped upside down, aren't they, really? Like with AI, you've got to pretty much

Jessica MatheronSo I think, yes I think communication has always been important, but I think here it's something specific. It's communication by people who understand the tech. Because I think communication for communication's sake sake is not necessarily sufficient. I think where we've been lacking is we need to hear more from people who have tech talent, but of course, that tech talent needs to communicate in a way that is understandable by the business

ben parkerSo what was harder, the technology itself or getting the organization to change around it?

Jessica MatheronSo organi-- I wouldn't oppose the two, honestly. Organization is, in a large corporation is extremely tricky. And this is I've come to understand that it's a result of the size itself. And I have this nerdy comment, but if you think about the number of emails, the number of meetings that you have in large corporations, it's actually proportional to the interactions between people, which scales as the square of the number of people that are involved.

Jessica MatheronAnd you can really see how that hits home really hard when you have a, a corporation that is as large as Ne- as Nestle, so 270,000 people. However, circling back to your question, that that complexity comes around organization, comes in several flavors, and I will discuss three of them. The first one would be the politics, and I'd rather say that, from the start. It... You need to get your politics right, and that's really the, the role of a chief data and AI officer, is making sure that there is alignment across stakeholders at a very senior level before you start.

Jessica MatheronWe s- we spoke about the support of the CEO. That's key, but it's not sufficient. You need to also, for example, if you're working on consumer data, get the support of the CMO and IT, et cetera. And certainly on some projects, like with the marketing mix models, that's where we were lacking, I think. But on the consumer data, it's been mostly quite aligned, so that's really important. However, there's also issues of even if there's alignment on the vision, there's there can sometimes be problems of conflicting perimeters, right?

Jessica MatheronIt's easy to not spend enough time defining the scope of each team, and then at the boundaries you sometimes have gaps or overlap. And so I think- again, it's really important to have clarity there from the very beginning s- if you want to be successful. Then moving into maybe tech org, which would be, like, the second flavor of this. In large organizations there's a habit of working with external vendors, and that can be, quite effective if you're trying to scale quickly and you just can't afford to have an internal team.

Jessica MatheronBut at some point, in particular with AI, you will need to have people internally who have, the right tech talent, the right tech skills to power you forward. And this is a new challenge that is both tech skill and org, right? And it's about how to get large organizations that are used to that way of working to start collaborating with internal teams effectively. And there's a corollary to that, which is that you need to hire different people, and that's a change in culture, right?

Jessica MatheronThe code is cool kind of culture. And that means also creating new paths for HR so that tech people can have opportunities to lead inside of the organization. And again, circling back to what I was saying on, tech people need to be able to communicate because otherwise people who don't understand tech will communicate for them, and that is not the most effective way to move forward But then I guess the last the last one of these is tech implementation.

Jessica MatheronAnd that is generally a problem of product. So bad specifications, right? So it's both org, but also tech. If you don't have the right specs then it's very unlikely that the product you build will be successful. One of these things was getting GDPR by design, as I mentioned, and this was both tech and org in that we needed to get a lot of people in the room together, right?

Jessica MatheronSo it's both stewards for data contracts, but it's also the data protection officer, w- who's part of the legal department, building that close relationship so that there would be trust again. And and also so that he would not prevent us from moving forward, right? So it was really getting that, that relationship to work. Of course, there are more more techy kind of topics like handling real time, which is always a, a, an extra topic in the implementation.

Jessica MatheronSo we had a little bit of a difficult moment on that front with the consumer data, but I think it's getting resolved

ben parkerOkay, cool. And then obviously you mentioned that, like you now engage sort of tens of millions in, of consumers. So what role has AI played in helping you reach and serve that scale?

Jessica MatheronSo again the first thing of that project, and I really wanna make that clear, was the foundations. So before AI played a role, we needed to get the foundations right. And what I mean here is the stack that would power the right speed, the right compliance, and also the right unification against decentralized stacks in every market. So that was key. And to do that, we needed to have an internal team to build versus being, disintermediated across lots of vendors.

Jessica MatheronSo that was the first step, honestly. And then of course, yes, we deployed AI, which really changed the way that we can scale in terms of providing value, so with advanced models. So if you think of an example, s- very simple example would be before we had that stack, we would have every team in every category cross-country in Europe creating a simple model to know if a consumer had churned from our CRM.

Jessica MatheronAnd those models were very basic. And now because we have one data model everywhere, we were able to do much more advanced machine learning and deploy the same model across the entire scope. So that was, one of the ways that we were able to engage tens of millions of consumers efficiently. And then more importantly, as we move to AI agents, is the way that we're reinventing the marketing ways of working at scale.

Jessica MatheronSo making them secure from a regulation standpoint, and also making them quick. So there was no way that every team in every category cross-country in Europe was going to have enough tech skill to be able to reinvent the way they were working. But because we put all of that together, we were able to to deploy the right stack for them

ben parkerAnd other, now, there's so many... each business unit has got their use cases. So obviously choosing the right use case is important. How are you going about measuring whether gen gen AI was creating value?

Jessica MatheronSo before we can measure whether it's creating value, you actually need to build a business case. I think this is... it's really key before you start any project to understand what you're trying to get out of it and actually go through the process of building that business case. But of course, acknowledging you will need to make a lot of assumptions. So for example, on the s- specific topic of the consumer data we split the value between what we expected to happen in the CRM, from media efficiency, the tech cost, and also the time gained.

Jessica MatheronAnd the... the, the numbers that we achieved in each section are really dependent on a lot of assumptions. But that's what's interesting, right? Having those assumptions be very clear and so that before you can actually see the value of the project itself, you can actually check whether the assumptions are holding up in each specific bucket of value, right? And those buckets are not at all of the same size. So for us, in this specific project, the, the uplift in the CRM is much larger than all of the three other pieces of the pie combined.

Jessica MatheronAnd so we know that's where the opportunity lies, but also the risk, and so we'll be, controlling much more tightly there. The second phase, of course, is then tracking the benefits as they arise. I think in some companies this is easier than in others. In our case, because we actually don't sell food to consumers directly, but we go through retailers most of the time it's tricky to link everything back to actual sales. Of course, we make assumptions but you're not gonna get to a very clear dollar amount in the same way that you would if you had a e-business, for example.

Jessica MatheronThere's more certainty around costs. So if you're, removing some legacy platform that was expensive to manage that's a lot more certain. And then the time saved, of course, in, in, in wages is a lot more clear as well. So circling back to just, what we can measure we were able to reduce the time from the... on the audience building from 10 days to 10 minutes, and for us, that was a key success

ben parkerIt must be a good feeling seeing the success as well.

Jessica MatheronYeah, absolutely

ben parkerSo I g- I guess, what surprised you the most once you started operating at scale?

Jessica MatheronSo I'm gonna focus in here on the word surprise, because there, there were many challenges, but those were not unexpected. What surprised me was the difficulty or the, the, the lagginess really of the handoff between our agent that we were building on LangGraph and leveraging Cortex because of course w- our stack was on Snowflake. And we realized that agent to Cortex handoff was extremely laggy to the point where it wasn't okay from a, a user experience standpoint.

Jessica MatheronAnd so we had to make ch- you know, serious changes to our agent structure. We hadn't expected that to be as, as laggy. So that was, one of the bad surprises and reworse we had to go through. And then the second thing, and I briefly alluded it, to it previously, is the fact that marketing teams needed more than the agent that we had in mind. In fact, they needed sometimes some quite simple things on the side of that agent, things that were very deterministic, in fact.

Jessica MatheronAnd I hadn't thought that those things would be necessary but when we sat down with them and looked at the flow of their work, we realized that those things would be a bottleneck. And so those were the biggest surprises. In terms of things that I, that were challenging, but not a surprise to me, was the complexity of evaluating and monitoring agents in production. And quite honestly we're still in the process of getting the right tools and processes around that.

Jessica MatheronBut it was expected. I think the the market speaks of this a lot and and so I think the... collectively, the industry is trying to figure it out. And then one other tricky thing, but which wasn't really so much of a surprise is that tension between generalization and personalization, right? So the whole idea behind our, putting the stack together was being able to provide value at scale. But to do that, you need to generalize, and we can't do something very different and tailored to every team in every category in every country.

Jessica MatheronSo we need to generalize. At the same time, AI enables you theoretically to provide more personalization. And so we're still in the process of finding that balance be- between the two, but still providing value for each unit, but reaping the benefits from scale

ben parkerIt's good. No, but I like it. It's a s- good success story, and it sh- shows you that the technology works. And like you said, I think the... I know obviously it's, we're still early days with generative AI, aren't we really? Like it's still a baby. And it's gonna just get more and more powerful and like I said, ev- everyone's on their data journey. Everyone's learning. So it's, only things are gonna improve and it's, you've gotta learn, haven't you? You've gotta learn to go through the journey

Jessica MatheronAbsolutely

ben parkerSo I guess then from that aspect, what do you, what did you get wrong along the way and what would you do differently if you were starting again?

Jessica Matheronwrong every day. It's small things which is good. And I think it's best to have small mistakes every day or every sprint, so for us it's every two weeks rather than realize six months later that we really went in the wrong direction. So for me, and I really have that vibe inside of the department that it's totally fine to get things wrong, to put something out before it's ready, get that feedback, iterate quickly. Totally fine with that. I think if you step, if you take a step back on, bigger things that I might have gotten wrong, I think the jury's still out on whether I should have built the consumer stack alone rather than partner with the global organization.

Jessica MatheronI certainly thought about it quite quite a lot before I made that decision. I think I've made the long-term bet that integration would benefit everyone. But it's true that my department, most people come from tech, and so we're used to operating at a faster speed. And working with the larger organization comes with a huge administrative burden and loss of control. But I think in the long term this will be beneficial.

Jessica MatheronWe're, we're building that trust and the stability to operate this for, many years to come. So time will tell. I'm still hopeful. And also even if you had perfect hindsight, if you were starting again and you knew exactly what would happen, it's quite tricky to change the dynamic internally. So i- it's, I don't think that insight here is the most important thing.

Jessica MatheronThere's such a critical dimension of navigating the organization and the people and building that trust. I'm not sure that I would have done things differently even with perfect hindsight

ben parkerOkay, brilliant. Thanks for sharing. So then for another enterprise that's trying to scale GenAI today, what would you tell them to focus on first?

Jessica MatheronI, I think we've covered this already, but honestly, the, the audit of the data stack is really key. I don't think you wanna start over-enthusiastic about that. Actually go in, check out what is the data that you will need for your project, and make sure that it's ready for AI. And don't listen to the sycophants internally that are gonna say,"Yes, of course, everything's ready." Once you start pulling that data in you're gonna be in for a, a surprise. So that's really the first thing.

Jessica MatheronAnd in fact, you should probably prioritize your projects, not just by the impact they will have, but by the data readiness behind it. Because if you can get a medium-sized impact in a few weeks or months, you should start there, rather than have that gigantic impact but that's gonna take five years. It's really key in terms of speed to audit that data. But then, of course, you need to start with something that matter, that matters to your senior stakeholders, in particular the CEO.

Jessica MatheronAnd so here I think it's important to iterate quickly on some of the small, small things that he might have in mind. They don't necessarily understand what is going to be the most complex. So if there are multiple things that the board is interested in and some are easier than others, then definitely starting there, getting that speed, getting that momentum is key. And then I would say don't un- underestimate the importance of the tech talent.

Jessica MatheronSo- in larger organizations where that tech talent might have been outsourced for decades, you're gonna need to have a few key people inside. So make sure that before you get started with, consulting firms, you actually have a few people internally who will be able to audit the quality of what you receive. And you're gonna wanna pick people that have implemented rather than simply advised. I, I was at a conference I think it was last week, and the CPO of a tech platform said few people...

Jessica MatheronSo f- four people with AI are superior to many people with AI, and that is vastly superior to many people without AI. But he really put it in that order, and I thought this was really interesting, and it's true. Because when you have just a few people that are powered by AI, so with, very good AI-assisted coding, like cloud code they can make... They can align really quickly. So it's much better than having a large organization that also has AI.

Jessica MatheronSo you wanted just a few people on a very clear project powered by the right AI stack, and that's where you're gonna get the momentum

ben parkerYep, I'm hearing that a lot. They've been more lean, it's just a bit more powerful.

Jessica MatheronYou said the exact word. It's really lean management

ben parkerI c- what do you think is the biggest opportunity from here then?

Jessica MatheronSo I think the really big challenge in large organizations, and even in smaller ones, honestly, now that I think about it is redesigning the end-to-end workflow. So you have ways of operating, and they're mostly, driven by humans. And some of these tasks are high value add and complex, but a lot of them are not. Even at the senior level, I... When I see, the... what I spend my days on, there's multiple things in my personal workflow that can be automated.

Jessica MatheronAnd, I go by priority, but a lot of it could be automated. And if you think at the level of a company, those successions of tasks between people create, constitute an end-to-end workflow. And those workflows should not... The new AI workflows should not mirror the past ones. They really need to be redesigned with AI at the heart of them. And so that's a, that's a, a large amount of work that needs to be done, and it's going to be both with tech, but also with the business people who have a breadth of context but lack the technical skills to do it themselves.

Jessica MatheronSo I think I really believe you're gonna have small pods of people coming together and just redesigning the companies from the bottom up. And that's gonna bring reduction in cost, a lot of more speed, a lot more security, a lot more compliance. So I'm very hopeful on that front. And then I also think that we're gonna find, once we do that or in the process of doing that, we will find new ways of doing business. And so I'm thinking of better integration with our suppliers and our clients.

Jessica MatheronWhen I see the data that we get from, panelists, from media agencies, it's obvious to me that we can do much better in terms of integration, but also downstream of us. And so if I think of the retailers and, Nestle is investigating doing such things, we could actually im- optimize the supply chain by simply having that integration. And I think there will be new ways of doing the business of the past l- with those, those suppliers and clients.

Jessica MatheronAnd then maybe the last thing would be completely new businesses. And I think it's still very early days to imagine, businesses that aren't yet into existence, that you're not simply going to optimize, whether internally or externally, but actually completely new f- new ways. We've launched a very early stage initiative called Nutritech, which is- basically to use the new technology to improve the habits around eating in, in Europe.

Jessica MatheronAnd it's still, those, that business is still in the process of defining itself. It's something that we haven't done in the past, and wouldn't have been possible without AI. And we're discovering, how we would go and operate that

ben parkerBrilliant. Jessica, it's been a pleasure having you on the pod. You've, yeah, shared some great insights, and it's, yeah, it's been a pleasure speaking with you

Jessica MatheronThanks, Ben. It's been a pleasure on my end as well

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What It Takes to Become a CIO Today

    Nov 10, 202638 minAudio

    In this episode of Data Analytics Chat, we welcome Saadia Khilji, Global Chief Information Officer at Coverys, a medical professional liability insurer that has served the industry for more than 40 years. Based in...

    Credit the people on this episode

  2. Are Companies Wasting Money on Data & AI?

    Nov 3, 202648 minAudio

    In this episode of Data Analytics Chat, we welcome Lior Barak, Author & Founder of Impact Operations. After 15+ years in data organisations, Lior now helps data teams and their leadership treat data and AI as a capital...

    Credit the people on this episode

  3. Your AI Is Live. But Is It Actually Working?

    Oct 27, 202639 minAudio

    In this episode of Data Analytics Chat, we welcome Ryan Stevens, Senior Director of Applied Science at Ramp, the financial technology company helping businesses save time and money across cards, bill pay, expense...

    Credit the people on this episode

  4. AI Is Moving Fast. Is Your Data Ready?

    Oct 13, 202646 minAudio

    In this episode of Data Analytics Chat, we welcome Erica Howard, Head of Consumer Data Ownership at TD. Erica has spent 18 years at the bank, starting in the call centre before working on TD's early data governance...

    Credit the people on this episode

  5. They Tried 3 Times. Then AI Changed the Enterprise Data Build

    Oct 6, 202646 minAudio

    In this episode of Data Analytics Chat, we welcome Mary Alfheim, VP of Data and Analytics, and Joseph Kroon, Data Architect, both from Headspace. Mary previously led analytics teams at Amazon Prime Video and HBO, while...

    Credit the people on this episode