So please welcome to the stage Ronnie. Perfect. Yeah, thanks for also saying that you know, we're not doing just slide where a lot of a lot of people know BCG as a strategy consulting company. We have also BCGX BCGX is kind of the builders piece where we have the applied AI advanced analytics and really try to help companies drive and accelerate their AI adoption for transformation. Talk a little bit about that today. But you know, there are two parts of today's discussion. One is a little bit about the generative AI how we position it also and how we talk about it with C-suite executives and I think apart from you know, all the technology that you know, it's a little bit how they perceive it and what their fears and what their what their opportunity seeking is around that and the other part is a very purpose-driven element of that that that I build a a solution around tackling some of the ESG topic specifically the S in there.
So diversity equity and inclusion using AI and we have a really new approach that that we believe helps a lot of companies around that. So I mean most of you know that but I think for the C-suite what I typically like to point out what the difference is of generative AI. You know, it's not just the chatbot we've heard it today. But that it's a pre-trained model. That sounds great, but it also has limitations, right? We know it's trained on internet scale data. With internet scale problems. Um it has the self-supervising training element in there which I think is very very special. So that gives rise to this power that it can create these images and create novel responses which of course has the implication of hallucinations as well that sometimes we need to control especially if it comes into fruition, you know, in within companies.
Um and I think you know, the point that needs to be realized by by most of these C-suites and they kind of know, it's going at a speed that is breakneck. Which also means that it's uncertain for them where it applies, who are the partners, who are the vendors, etc. in there and that makes them often very reluctant on this. So won't spend much time on this. I mean but it's highlighting a little bit that there's a lot of money that is pouring into uh into generative AI and there are you know, a lot of implications that we're expected. Of course, nobody has you know, the the the balls to see what comes in there but we've seen the traction and the speed that that that we we have all observed in the last couple of months, days and weeks.
We also see the first implications of effects in different industries in there and they're partially substantial. They're more to come but yet, you know, we see that there's a traction also on a commercial level that we haven't seen before. And I think it's a couple of examples that you know, are publicly perhaps able to be talked about. Is you know, perhaps a couple of examples here from Mattel that uses it really in a very interesting way to come up with novel toy designs. In a similar but in in in some aspect a very revolutionary approach in in Procter & Gamble, we look at the formulations of some of their products and we can see that generative AI also GANs can create new formula approaches in a much better way and to test them reducing the time to market for any product launches and that is significant effects for them.
And I think Nike, you know, probably I don't know if they learned something, you know, with some of the commercial controversial design partners, you know, trying to see if an AI commercial an AI partner for designing some of their sneakers, you know, might be less controversial. Not sure if that's going to happen exactly. One of the key slides that I wanted to show that I also, you know, talked about with the Fraunhofer Institute specifically Nico who's I think also here um is something that helps C-suite navigate, you know, how this is going to evolve how you can get started especially, you know, don't get started with the most complicated part necessarily but how is the journey towards it.
So I'll spend a little bit time. It's a very busy slide but we all know the left part, I guess. Um where you have the foundation model or models, you know, multimodal approaches are going to be substantially supercharging that. That that are based on that pre-training on a lot of different data sources. Now we know they can generate all kinds of stuff in there but they're different elements to it, right? We have the prompt and we have the results. Within that we can do a lot of stuff. Prompt engineering is one of the hottest job markets right now um and not without merit because in to understand both what you want to get and what the LLM or the foundation model understands is a crucial point.
There's also a lot of data that you submitted. Um Fabian just said it in in the keynote. You know, and for my clients, you know, the corporate clients. So they need to say, well, how safe is my data? Is it used for training? But also when I send it over, can I send it over to US? What can I send over? Is there personally identifiable information? German um EU act um about data protection. So you have these four wall concepts where they want to navigate and want to be felt secure within there. And there are different ways that you can empower this. I think the first one is sort of the out of the box where there's a lot of knowledge around this.
Where we see that really applying is when you have um when you have the unstructured data especially around general language. So a lot of customer service elements are being done there, the chatbots that understand the intent and can generate the kind of language that humans can communicate with. So there there's a a huge lever in it. But we see it often in a supporting function. So that's the circle on the top right there where you know, you have for HR for some of your non-core operations, you use these and that that's quite helpful. However, it also doesn't have that strategic impact necessarily that you might want to envision about generative AI. So that's where the second comes a little bit more in place and um you know, if you heard about LangChain the the the popularity of having an intelligent document processing where you control the hallucination by basically just ingesting smartly the relevant information from your corporate environment having it secured, you don't retrain the model necessarily but you do a first stage where
you identify the relative information the relevant information a little bit with more traditional I would say NLP or now it's traditional. You know, a half a year ago that was very novel way and then you send that tidbit of information that you better can control over to the to the foundation model and get a highly relevant result. And this you can drive quite a bit before you actually go to something like fine-tuning and I think fine-tuning the discussion that we often say it's it's typically expensive, it's fragile. You know, the first kind of effects that you can do with fine-tuning are great but you know, you have always a risk of overfitting and having very strange responses and not generalizable.
So it has to be, you know, handled very delicately. And then there are of course, you know, the superstar kind of cases where you create your own foundation model. Sounds really really interesting but it's you know, anything right now up to 50 million dollars to create these kind of foundation models. Sometimes, you know, substantially higher. They can be relevant but we see them often as a way to change your business model to adapt to a new one, you know, like Bloomberg GPT is doing where they go from providing analytical service to actually providing the foundation for others to do that. So I think it's a very exciting possibility and what we try to do is both help CEOs navigate through that space where in your case, which kind of use cases, where does generative AI apply?
What can it help rather than just cost saving, you know, can it help you not eliminate necessarily customer service but grab the knowledge that is implicit in your company, use it ideally in a way that supercharges your core processes. So you know, like we did with AI, have a sales forecasting so that you can sell your existing product better. If you have unique data that is a competitive advantage, you can leverage this actually to make your core process or your core product into a competitive advantage and I think that's where we want to ideally steer our clients there but there is a path throughout these different models on that. And I think was a little bit mentioned by Fabian but we also see generative AI is really a step change but it's not replacing the traditional AI.
It is actually complementary to it and it supercharges it. So So, extends some of the capability not just, you know, the human interface, but some of them is semantical understanding that we can have in order to get a better handle of the unstructured data. Remember that, you know, up to 80% of the data in companies is unstructured. And it's untapped, typically. It contains a lot of very critical and very personal information. Personal, I mean, to the company, which has not been unlocked yet much. So, I think the combination of this is the smart thing, and that's what you saw a little bit, the foundation model on the left side on previous side, and you know, some of the smart modeling, some of it being more traditional on the right side on it.
So, we see a very good combination of them. But, you know, it's it's right now through the hype trying to navigate that it's not replacing that. So, what we tell our C-suite is, "Okay, you need to get started. You know, there's a gap that that that that finds it." A lot of my programs that I do is where we do an ideation before we where we see not, "Hey, is it customer service or something of that kind that we do?" But we go a little bit deeper and understand what is the impact for you on that. And identify these first real use cases, not starting to experiment with a technology, but rather see where it can apply to you really in a meaningful way and how how the path is.
And then, they need to get started as quickly as possible because there's a lot of implication on it. It's the way that you ideally change your products or you you supercharge your products, you supercharge your R&D, and the the whole organizational structure, not just what we heard at the key notes about what we typically say shadow AI, you know, where where people go off and and do some rogue stuff, but the whole way that we leverage the knowledge, that we actually apply it in order to supercharge either the distribution or the margin, but also to change the way that the products are designed. And I think that we see in the operating model that will change quite a bit.
And you heard from the previous talk, of course, a lot of the responsible AI guidelines that are both, you know, on organization technical and a process level. So, the way that I typically approach it uh with clients is that we look along the value chain. We understand where the value actually gets created within the company or within the products that they develop at their clients. And then, you know, here's a a first kind of selection of a couple of, you know, industry relevant use cases that could be applied to it. You know, we have much longer list, and often the client also has its own ideas. But what helps us really is to actually identify through the top where is the biggest unlock capability in terms of, you know, the the success in the business impact for the client, not just having a long list and see, you know, what's what's easiest.
And what we do then is we try to map these use cases along typically two dimensions. Anybody who knows BCG, you know, we love two by twos. Um and here I changed a little bit to a yeah, now almost three-dimensional. But on the one side, we look at the impact, and the impact is on on two levels. One, what you can achieve now, you know, when we start programs, we cannot wait 2 years with companies to see that. We want to have something within 3 months that shows historically that is achievable, but also within 6 months that you see a real P&L impact. And, you know, 6 months is not a lot of time. At the same time, it needs to have kind of a road map.
And you see here a little bit, you know, of of an example befitting of that that you can say, "Okay, we start here, and then we go into this direction." And the the second dimension is um the feasibility. Feasibility or speed of uh speed of implementation or complexity. For me, they're all correlated, but it's kind of like how long will it take you to get this to fruition. And with this, you know, like I said, we get typically the frontrunner use cases. They're the ones that you you will have impact. That's the ones that you can see quite quickly and then go from there. So, I'll switch to the other topic um that I wanted to talk about where, you know, this is one of the examples.
It actually uh started within something that we did for ourselves. So, we looked at our um own DI numbers. We're a very meritocracy-driven culture. We believe, you know, the best advance, um the best get promoted, the best we we retain. Um yet we did see that there was a difference in perception between, especially, our our female employees versus the men with regards to some of the metrics that we had. Um and we wanted to understand why. So, in a larger context, you know, which is very quickly, the promises of DEI, you know, we've heard all about. There are lots of studies out there that seem to prove its significant amount of EBITDA. You know, we see that if you have more diverse team, you're going to have more uh more, you know, business impact, more revenue, higher productivity, etc.
Yet at the same time, we see that these promises in those companies who apply it don't seem to really take traction. And what we see what we sent out and we we we work with a lot of companies, the main reasons we see is that they don't understand where the gaps are. That included us. Their inability to find the root cause. A lot of these D&I initiatives are are done through best thinking, you know, um somebody who has a hypothesis, I think we need to do that and that. But they don't measure it. And there's no way of defining the success of this kind of hypothesis. So, this is where we see that you do a lot of stuff, doesn't seem to have a lot of fruition on it.
So, what we designed is a concept where on one side we use what pretty much all of these DEI kind of tools use, but in a way that also goes historically. So, you go much much further into the past, not just a snapshot. Think about surveys that have a current snapshot, and then if you do a survey a year later, you don't know if they're the same people, you don't know what correlation to the initiatives you have. And here, we go and we see what kind of path did our employees take along, you know, the the last 10 years. And then, we unlocked, and that's where we use the generative AI or the large large large language models, some of the more underutilized or sometimes never utilized data uh into, you know, from from a qualitative inter- uh inputs like um feedbacks that we have, like promotion emails, etc., into a quantitative model.
And this is how we can actually measure, you know, within a cube from different dimensions and see these kind of gaps. We can answer and ask these kind of questions. So, you know, to be honest, I didn't do much around, you know, the natural uh interfaces of that. A lot of this is hypothesis-driven kind of questioning out of the box, but then you have for the root cause analysis these kind of questions. If you do them in a natural language interface or you do them in a more traditional, you know, I I think that's not um that's not the big difference on it. But, you know, just to give a a small flavor of how we used the generative models.
On the one side, you know, let's say you have feedbacks uh and and, you know, in our company, we have a lot of feedbacks that we have structured after after projects. And there, we have a first set of hypotheses that come through DEI experts. You know, we know that mentionings of personal train straits uh traits, strength, and weaknesses, etc., those are relevant points for somebody to feel, you know, how how they're being treated or not. And we feed them into a generative AI model, and the generative AI model allows us to generate better taxonomies. So, I know that, for example, if I talk about intelligence, there subcate- gories, you know, clever, um intelligent, or sharp-minded, those are all expressions that nuance a little bit that that kind of personal trait.
And I can create with the ability of embeddings and large language models a way to quantify how much I'm talking about that. It's not just presence or nonpresence like we did before, so zero and one, but a very granular way to represent how much of it is there. And this we put in into the large model. So, an example um very quickly is we we analyze, you know, we found that there is a a difference with high performers, you know, between women and men, that these high performers that have actually really good career path, the voluntary leave of women was almost double as high as that of men. And this is expensive. We're losing people that, you know, are are valued members, and and why is that happening?
What we found is that these women often times get what we call unconditional feedback, meaning that we say, "Oh, your work is wonderful, fantastic, good work." Whereas men typically get things like, "Oh, when you're in front of the CEO, your concise way of presenting, you know, helps seal the deal." So, it's situation, action, and outcome, where I can measure saying, "Okay, I'm doing good on that or I'm doing bad on that and this is how I can change it." And this is something, you know, we observe in through research in general, but here's the first time we can actually quantify it and we can measure it. Now, interesting, a lot of the people who give that feedback don't actually want it, but it's like a societal bias.
But the question is, do we want to overcome that? So, what we created is an actual AI agent that helps monitor your feedback as you write it. And they typically they don't control it, but they say, "Hey, you know, there's a lot of mention of personal traits in there. There's a lot of unconditional feedback, you know, are you sure about that?" And just that feedback improves and we can actually measure the improvement through that kind of initiative and really see, you know, does it work or not? And, you know, we have other elements where I looked at recruiting um and I actually identified by looking at the past that there was a region that went from the lower tier of, you know, the ability to recruit women and and actually, you know, send out an offer and that the women accepted the offer from from bottom tier to top tier in very short amount of time.
And then I asked, "What what what what happened there?" And they did a a set of initiatives. What I was fascinated is I could actually measure the impact of that initiative. What is the net increase of women retained in there? A lot of clash that comes often against things like that is like, "Yeah, well, you know, they're just putting a quota for women. They're letting everybody in, right?" Which is sometimes unspoken, you know, a kind of critique. But we were able to show on that little graph there on the side is yes, the funnel, you know, has opened and there is of course, you know, if we go to a lot of non-core universities where we typically recruit, you have more women perhaps that actually don't fit.
At the same time we look at what is the the the the kind of criteria, you know, for entry or not and we could show we use the same criterias and the same kind of judgment for these women then for men as well. Meaning that we apply the same standards and therefore the net gain of, you know, women that we have is actually a true net gain and stops that kind of discussion. So, in in in short, you know, what we are able to do with GenAI now is once surface the problems, find the gaps, see where is the stuff that that hampers you, but also have that measured against KPIs, against business KPIs. Some functions it's easier if it's, let's say, sales.
We have done it front office with a large coffee chain in the United States where we see an actual increase of revenue happening and a decrease of cost because suddenly they don't need to hire, you know, that many people again and again and again and each employee can work longer. Um and then the the root cause analysis I think is the most interesting part, the diagnosis of it. And there the more unstructured data we can foster through and have these correlative effects and narrow them down, the more exciting it is. And this is something that that wasn't there before. But then the initiatives, right? We have a lot of companies that have, "Okay, let's do an unconscious bias training."
Doesn't help or at least if you have nothing else, it it gets you started. But it needs to be customized for you. And this goes until a very very granular level and can say, you know, this is the part you need to pinpoint on and this is what we would expect as an effect. And this allows you also to measure the effect, but also see in a first controlled environment, you know, okay, here it works. Can we apply it from, say, Europe to the Middle East? Is it the same ground conditions, the same baseline in there? And so, this is something where you can do this deploy and and scale of it. So, coming back a little bit to the generative AI topic.
I think, you know, what we see there's a lot of excitement, you know, at the CEO level. Okay, that we need to do something on that. What is it? So, I think the three main messages that we try to give is on one side, you know, these barriers for the new entrants will get lower. I used to work before for Airbus and the CEO, the then CEOs told me, you know, the barriers are falling. We want to be aware of them because once we identify a competitor, a new competitor, it's typically too late. So, this is the big danger. It it democratizes access to AI, makes it much much easier out of the box, get you started.
Might not have immediately that impact, but you have whole new ways of encroaching incumbents that you didn't have before. And I think if you don't move, your competitors will. They will use this in the first stage to get more efficient. More efficient means they're going to out-edge you, you know, on some of the market share, which you know, will hurt you in the long term, no matter how they do it. Of course, you can say this is a race to the bottom of efficiency what we'll be doing, but now it can go faster. And I think the the other problem is, you know, you have a fast-moving environment. If you say, "Well, let's see first, you know, where it where it coagulates, you know, who are the relevant players?"
Um we we're discussing, you know, from the large language models there's going to be an explosion, then there's going to be an implosion. But when are you going to start working on it? But what we know is these gaps that create right now, they will grow bigger and if you don't start now, these gaps will likely not be able to be overcome over time. So, I don't know if we have um you know, one quick time for one two questions, but you know, happy to discuss otherwise outside of the out of of the stage. Thank you very much. We do actually have some time for some questions and I'll kick proceedings off with you, Ronnie.
Um you talk about talking to C-suite. Um what are the biggest challenges that C-suite specifically face when thinking about generative AI? Is education for the C-suite a big part of what you have to do or have we moved beyond that point now and thinking more about application? I think it's it's a really good question. So, the education is on different level. There there is a culture, especially in Europe, especially in Germany. I I spent a lot of time in the US that is about experimentation and de-risking. So, okay, this is dangerous stuff, you know, we hear a lot of warnings. We heard it in the talk before, rightly so, about bias, etc. Let's start in some safe space somewhere, which often happens to be a non-strategic element.
The thing is, even if that works and often times it doesn't because it's just a a technology proof point, what happens next? How does that imply, you know, going forward? So, the the biggest time we spend is actually, you know, it's good that you start thinking about it and start running, but don't just run around like, you know, a chicken with a head chopped off. Try to see where does it apply for you? Where does it become a advantage for efficiency? Where ideally does it become an advantage for a competitive advantage and can propel you, you know, into a new era? Or for some companies and that is, you know, my desire to find these kind of unicorn cases is where, you know, somebody like Bloomberg, you know, goes and says, "You know what?
The knowledge that we have amassed and data over the last 40 years, you know, and I think some of the data even longer, we can leverage to create a new core business, which might in fact eat their current core business." It's a bold move. Um so, I think to guide the CEOs to that direction is probably where we spend most of the time right now. Fantastic. Thank you, Ronnie. Well, once again, thank you for your time. Thank you. And we'll move on to the next speaker. Thank you, Ronnie. Thank you very much. So, from one of the big guns in the industry to the next titan of the ecosystem, I am