Welcome And Format

SPEAKER_01 0:10

Welcome to What's in the Box, the brand new podcast brought to you by Box Technologies, powering retail with purpose. Boxtech delivers innovative and market leading customer engagement solutions that turn business ideas into a performing reality. From design and integration to ongoing support and maintenance, we're with you every step of the way. Now, this podcast is a little different to what you might be used to. It's audio only and totally unscripted. Well, we say it is. We've had a little chat with our guests beforehand. And we're around about 20 minutes each episode because we're busy. Our guests certainly are, and we're sure you are too. So let's get right into it. My guest today is Director and Chief AI Officer at Heineken. He leads global data science and analytics efforts, driving the organization's digital transformation strategy. His team focuses on delivering incremental business value by optimizing value chains, enabling predictive insights, and leveraging machine learning to forecast outcomes. They collaborate closely across the enterprise to embed data-driven decision making into the company's core operations. With over 15 years of experience in data science, machine learning, and predictive modeling, he brings a strategic approach to solving complex business challenges. His leadership has empowered teams to implement impactful analytics strategies, leading to significant operational efficiencies and growth. Passionate about advancing AI solutions, he remains committed to creating sustainable organizational value while mentoring teams and fostering innovation. I'm delighted to welcome Surgi Gosh to What's in the Box. Welcome.

SPEAKER_00 2:02

Thank you very much, Andrew. Glad to be here.

Using AI To Summarize The Talk

SPEAKER_01 2:05

Now, Siurji, we first met a little while ago at an AI and retail conference in London. And uh I loved your session so much that I wanted to invite you on as a guest onto uh the uh what's in the box podcast. But I then and I and I I think we agreed we would we would kind of go over that uh that session and summarize that and then I thought ah I've got a bit of a problem here because I haven't got a transcript uh to it. So what did I do? I asked AI. So the prompt, and as you know far better than I do, it's all in the prompt now. And so I I asked AI, well, what uh give me a clear summary of Heineken's AI journey as described by Siri Gosh, chief AI officer at Heineken, based on his talks and public comments. And this is what it came back with. And I'll I'll read out the uh the five headline bullet points, and perhaps we can use that as a little bit of a structure for this conversation. So it came back with starting small and strategic from concept to scaled solutions. Then number three was building AI as a capability. Number four was looking ahead, digital twins and innovation. And then finally, number five, cultural and organizational adoption. Now, I hope you're not gonna tell me now, well, no, that's that's nonsense. It's nothing to do with that whatsoever. So I I hope that that's reasonably accurate.

SPEAKER_00 3:40

That is very that I think that summarizes quite well, actually.

SPEAKER_01 3:43

Well, that's good. My faith in AI. Uh and I must, whilst we're talking about it, I I decided quite recently I need to verse and educate myself far more in in AI. So I'm doing something at the moment this month, I'm devoting an hour a day to educating myself on AI. And today is day three, and I'll tell you what, it's fascinating. I'm learning, I'm learning all sorts of things, uh, which I won't go into because we're not here to talk about that. But yeah, all I say is for the audience, I think it is, I'm sure you'll back this up, it's well worth people educating themselves in all aspects of AI and what it can do, and more importantly, perhaps what it can't do.

SPEAKER_00 4:31

Absolutely. In today's world, having this under understanding, super important. Yeah, yeah.

Start Small Think Big Strategy

SPEAKER_01 4:37

Okay, so let's enough from me for the time being. So let's get into it. So number one, uh so starting small and strategic. Tell us about that.

SPEAKER_00 4:49

Yeah, I think the way we look at things, right, since I joined Heineken about six years ago. Our overall approach is think big. Think big in the sense that have a big vision, what you really want to do with AI across the value chain. But start very small. Because if you think big and start a very big project and it fails, the downside is enormous. So we want to be very conservative at the beginning, to start very small with small experiments, and then make sure whatever we are doing actually ties to some sort of a business outcome. And that's why the strategy part comes in. So what we do, we look at our value chain and we see which are the pockets where AI can be applicable. And you'll be surprised when we look at our value chain, even within Heineken, there are hundreds of areas we can actually automate and optimize using AI. But we cannot solve everything. That's just impossible. So we just pick and choose those projects that have maximum potential top line or bottom line impact. Start very small with small experiments. If it works, if the proof is there, if the value is there, then we start scaling it and making sure it gets embedded into the business processes as day-to-day ways of working. So it is not just stopping at an experimentation phase, it needs to embed into the business, and people need to start using those products to make decisions. That's where it comes to a full circle, and that's what we have been doing over the last almost six years now.

Who Heineken’s Customers Are

SPEAKER_01 6:14

Just for before we move on, a little bit of context for the audience. Heineken, your customers, just uh tell us a little bit about who they are and to just sort of position this for people. Because we we know the end customer is the likes of all of us who hopefully uh consume Heineken and perhaps preferably Heineken Zero, I'm not sure. Tell us about your customers, the business customer.

SPEAKER_00 6:42

Yeah, so I think walking back first from the consumer, so we are also very, very consumer-obsessed in the sense at the end of the day, if consumer doesn't consume our products, there is no business indeed. So that's the first pillar top priority. But then what we refer to as customers are the retailers, the wholesalers, the outlets, the mom and pop stores, the bars and restaurants. These are actually our customers. But then you go a little bit deeper level, where I sit from for me, any stakeholder around me, within the business functions, let's say commerce, supply chain, procurement, finance, HR, doesn't matter. They are all my customers. So in this sense, I am building products for the business, not for my department. So our the the whole customer obsession is something that at least I try to embed as much as possible within my team members so that we are really stakeholder obsessed, that what they need, we would like to build it so that it works best for them.

SPEAKER_01 7:38

And I in in the introduction there, I was saying about my a little bit about my sort of personal journey with AI and so forth. Do you find that other people across the Heineken organization are really interested in what you and your team are doing with AI and how it can benefit the business and your customers?

Building Trust Guardrails And Demand

SPEAKER_00 8:00

I I think very good question. I think it's it's uh it's also a question of maturity. When we first joined, when we first started this team, it was six years ago, it was a green field. There was no AI at that point at Heineken, at least nothing structurally. I think over time, as you start building products and move away from experimentation to real products that people can actually use and they love those products, and it almost becomes the business cannot do their day-to-day job anymore without some of these AI insights. That's when you know you realize that it has made an impact. And as that happens, also word of mouth travels within the company. So one operating company pitches on behalf of the AI team that we have a big product that we are using, why are they not using as well? So then the word of mouth it travels, and today we are in a very good place. In fact, we have a good problem, which is we have quite a bit of organic demand coming from the business functions. And I'm trying to keep up with the demand. But that is a problem I will take any day, as opposed to having to sell and pitch concepts over and over again. I think the company is now quite comfortable working with data-driven insights and actually making business decisions, especially very big ones, with some of the AI insights. And of course, we want to make sure that we have guardrails in place so that there is always, if there is a big error in some of the models, we catch it beforehand so that you know nothing significantly harm is done. But with all the guardrails in place, I think there is quite a bit of organic demand in in the company as of today.

SPEAKER_01 9:34

And I guess that that's quite a lesson for for people listening to this, is it to make sure that you have your guardrails in place.

SPEAKER_00 9:42

Absolutely. Otherwise, it gets into uncontrolled territory, and then if the predictions start not making any sense, not only will we have reputational damage, we will have business impact, and also the trust and the credibility is also gone at that point. So we need to be very, very on top of things when looking at model monitoring and robustness of the model and how good it is. These things we have to absolutely take care of. Yeah. There's no shortcut to that one.

SPEAKER_01 10:09

Yeah.

SPEAKER_00 10:10

Yeah.

From Concept To Scaled Solutions

SPEAKER_01 10:10

Okay, right. So number one was starting small and strategic. Number two, from concept to scaled solutions. Tell us about that.

Local Nuance And Co‑Creation

SPEAKER_00 10:21

Yeah, I sort of touched upon a little bit in my previous answer. I think when we look at a business problem, we try to find similarity of the same problem existing within multiple operating companies. Heineken has quite a few operating companies across the globe. So if we build a product, the first thing that comes to our mind is can we scale this? Because the real value is realized when you're able to scale it. Just one bespoke product in one area of a business, yes, it does give you incremental value, but not at scale. So we want to make sure from the get-go, the design of the model, how the model is being hosted on a platform, how the data foundation is as standardized as possible, and how we build it on top of that with that design in mind. So that eliminates later on the need to overly customize over and over again new deployments for the same model. This is how we tackle it. So having that scale that I have to scale the model if I build it from the get-go helps us steer in the right direction from the initial phases of model building. So that's what has worked so far. But it's also, keep in mind, this is not very easy to do as well, because Heineken is a very big org with our presence is in all four continents, actually, four regions rather. And there are nuances within each region. The business operates slightly different than somewhere else. The tax laws, the pricing laws, the card rates, the change, so it could be a little bit different. Those things we have to keep in mind. So, what sort of bare minimum customization I need to do in a new deployment, but the foundational engine should be more or less standardized so I can scale quicker. So that's how we approach this. And this has helped us a little bit to in places scale quicker. Otherwise, you do it one by one by one, which is really tedious and time-taking.

SPEAKER_01 12:09

That's quite interesting, because in in um retail, and you know, we're we're we're in that uh domain if if you like, yeah, there are so many stories of retailers either from the US trying to make it in, say, in the UK, vice versa, or it and those are just examples, it could be anywhere. And is that you think down to that there are local nuances and it's very easy to overlook them and to make assumptions, wrong assumptions about your consumer, the market that you're entering?

SPEAKER_00 12:42

Absolutely. And it's actually a little bit of a make or break. If I try to embed a generic model without some sort of customization or without understanding the nuances of the local business, what are the exceptions I need to make into the model? It will just not work. It won't work up to the expectations. So, what we do to tackle this, the business and the operating company is involved from the get-go as part of a squad when we are building these models. So they give constant feedback on the features and upgrades that we need to include in the model that is going to cater to the local demand. Otherwise, you build something which is my version of truth, is not applicable on the ground. So yeah, we are very careful with this. We need to make sure that we we accommodate those local nuances and needs in the models. Yeah. Yeah.

AI As A Company Capability

SPEAKER_01 13:31

Okay, so that was from concept to scaled solutions. Number three was building AI as a capability.

SPEAKER_00 13:42

Yeah, I think it's a broad topic, and there are multiple ways of looking at it. One way could be we have a lot of models in the company across different business functions and people are using it. You could say that my job is done, AI is now embedded in the company, and people are using it. I would say that's one step. I think the step that we want to reach towards is most of the biggest business decisions are being driven by some sort of data-driven insights. And it doesn't have to be a fancy AI model all the time. A good analytical model in the background that really makes sure that it's getting the right trends and patterns from the data to make some predictions and businesses using it. That's the ambition we want to go to. So that actually means AI is so ingrained in our company, almost like it becomes involuntary use of AI or unconscious use of AI. So, to give an example, when we look at our phones, there are a lot of apps that are running AI, we don't even realize, we don't even care, we are not putting any effort. So, can we really embed AI in such a way that it becomes business as usual day to day in the operations that we do? So Agent T AI is a good frontier that we are going towards, which will actually help doing that. But the idea is that AI should be just bread and butter for a company like Heineken, which is actually very much operational and very much not sure the exact word is not, it's brick and mortar business, actually, a lot of it, right? So that's the holy grail that I'm going towards. So unconscious use of AI by our own employees across all business functions. I think that's the change we need to make next.

Hype Cycles Speed And Adoption

SPEAKER_01 15:23

That that's really fascinating, and I I totally agree. I wasn't at someone as retail, but uh it's relevant. I wasn't to NRF in New York last month. Uh David and Mysisia. What I was told was, as you might expect, that it was almost wall-to-wall AI. And I wondered, picking up on what you said there, uh do you think that we'll even be having this type of conversation in a few years' time? Because AI AI will just be in the same way that we don't think when we, as you say, we use our smartphone and we make a phone call, we have Facebook, FaceTime, or whatever it happens to be. We don't think about the technology behind that. We don't discuss it as a separate kind of entity, if you like.

SPEAKER_00 16:12

That's correct. And and actually within Heineken, it's already happening as we speak, as we transform the org in in making people believe the value in AI and the fact that they're using it with a lot of trust because of the guardrails we have in place. That transformation is happening. But I think eventually it will get to a stage where you have unconscious use of AI. You just it becomes natural to you. The question is who gets there first. I think it's the speed of getting there, knowing that that's the end state anyway. How quickly can we transform to leverage the benefits from that state? That might give us that edge. Because this technology is going to evolve even further. I mean, at this point, there is no stopping it. It's just as quickly how quickly we can get there and on board. And I'm I think there is also moving goalposts because the tech will also change five years later, like almost guaranteed. But can we quickly get up to speed as quickly as possible? I think that is what I'm targeting.

SPEAKER_01 17:07

It it's kind of for me, personal view on it, it's both very exciting but also a little bit scary. Uh, because it's it almost feels like it's a a journey into the unknown. Do you think does anybody really know what? You said five years, there's a horizon there. You know, what what is the world going to be like? Because because of the pace of change, it is so, so rapid.

SPEAKER_00 17:33

I I think, yeah, this might be a little bit of a speculative answer because nobody really knows the future. And and of course, such a dynamic technology that's changing every day. But I think a lot of the the adoption will go to a complete different level. I think today there is a there is also a lot of hype baked in, by the way, to be very real and transparent. And that hype probably will go away at some point because people will, and that happens with any kind of technology cycle. AI is no exception. And once that hype goes away, you will get into a more steady state phase and more slow growth in AI and adoption. I think that's what we're going to see in five years. More steady state, day-to-day AI baked in in our daily lives. Already a lot of things we are using today is actually driven by AI. We don't even realize. Our business transformation will also happen more or less at par with that sort of uh embedded technology. I think that's where we're heading. But what AI will do, is it going to be a self-conscious individual? Not going to speculate on that because there is a lot of theoretical misconceptions in that whole idea. So maybe that's for a separate day, not for two. Yeah, I think that's very wise.

Digital Twins For Consumer Insight

SPEAKER_01 18:39

Okay, uh, number four. So looking ahead now, digital twins. We hear a lot about that. Digital twins and innovation. And I'll just uh quote uh for the first time here what AI told me that you have expressed excitement about future AI capabilities such as digital twins that could predict both short-term and long-term business trends. So come on, AI has said that's what you're saying.

SPEAKER_00 19:07

Yeah, I think that digital twins is a very broad concept. I mean, I think you can create a digital twin of a manufacturing facility. You can create digital twins of customers, you can create digital twins of in-office transactional processes. You can do a lot of things with digital twins. But what we were trying to do is get more towards where is the impact, where does it matter? And that's why consumer digital twins. As a retail FM3G company, we today we send out a lot of surveys to our consumers, say, what do you like, what do you not like? Imagine you can do that at a massive scale with very limited cost. When you have digital twins of actual humans, they are replicas of humans that tend to behave exactly like humans. And then you get firsthand feedback from this large population out there that you can ask any sort of question. Hey, what's the new product I'm going to launch? I'm launching a new advertisement campaign. Do you like the look and feel? And then you can fine-tune your campaign, your product launch. And I'm just giving two examples, but you can imagine there's a lot more use cases from this. I think that's a little bit of a holy grail. And this technology is not new, we didn't invent it really, but it's it's more about how can we quickly apply it so that we get the edge. And it really gets us a sense of what really consumer demand means. Today we look at consumer demand through proxy signals, Nielsen data, et cetera. But what if you were able to ask 100,000 people out there, what do you like, what do you not like? That's first hand information. And I think that's that's very precious if you can get hold of that sort of data set. So that's what we're aiming for.

SPEAKER_01 20:43

Yeah, that that that is really, really interesting to see. I mean, I guess that's also getting getting into the realms of consumer psychology. You know, what's going on in our heads that drives our decision making.

Culture Anxiety And Gamification

SPEAKER_00 20:57

So that's exactly correct. Instead of getting proxy signals, though, for consumer psychology, I'm getting direct signals from the consumer. And then you try to fit a model on consumer preferences. I think that's yeah, that's probably the next step in in retail FMCG industries. Wow.

SPEAKER_01 21:13

Okay, time is marching on. Right. Number five, fine, and possibly, it's kind of leaking on from what we've just been talking about there, possibly the most important, cultural and organizational adoption. Because I remember at that conference where you and I met, one of the things I think uh it was in a panel that I was moderating. And I think the question that uh I was posing was what's the what's kind of the one of the biggest barriers to AI adoption? And the answer was people, us. So whether that is uh what this is about, I'm not sure. But anyway, cultural and organizational adoption.

SPEAKER_00 21:53

I I think it's it's about there is anxiety, of course. Let's understand that that's the ground reality that people in general, we humans, we have this tendency, what I call inertia of rest, in the sense happy where I am. I do not want a disruption in my life. That is the average human being. All of us are the same. But when you embed a new technology, immediately there could be an aversion towards it because it changes, it disrupts my daily way to day-to-day working. So the way to tackle some of this is actually demystify some of these models and make humans or our employees actually interact with these models in a very user-friendly, easy way. So now what we are doing is these big black box models, we are creating a Gen AI layer on top of it. And then you can ask questions to those models in a very easy, prompt-based English language. How do I optimize my budget? How do I optimize my promotion? And the model will give you excellent feedback based on the back-end black box that is running. And moment people see that, they start relating with the business outcomes. The black box is no more black box anymore because you have kind of opened it up a little bit. And gamification also helps with more adoption because people love to see products that they can interact with. I think that has worked worked quite well so far. So but there is a long way to go. But I think this is we can be very creative with this sort of gamification of AI products that takes away that anxiety away from end users. Yeah, no, absolutely.

Closing And Where To Learn More

SPEAKER_01 23:23

Well, we're just about out of time. Um we could go on talking for hours on this. It's a huge topic, and I know that we've only just skimmed the very top of the surface of this. But Surji, thank you so much. Really appreciate that. And that's all we've got time for from what's in the box right now. New episodes will be dropping every two weeks, so please stay tuned. And if you want to find out more about Box Tate, please follow the link in the description. Thank you so much for listening.

SPEAKER_00 23:51

Thank you very much.

SPEAKER_01 23:52

It's a pleasure.