Introducing myself. So my name is Ketan Dola. I'm joining in from Frankfurt. I'm a manager in AI and data team at Deloitte.And, I have been working in this space over eight years now. I help organizations build data driven products, which can scale from prototype to enterprise wide solutions. And outside of work, I also serve as a women tech global ambassador at Women Tech Network. And it's the role, that is very close to my heart because I deeply believe in the power of representation and leadership, especially for women in tech. So for me, AI isn't just about, like, cool algorithms or, let's say, futuristic tools. It's about making an impact, and it's also about, like, how we use this technology to solve real problems, create opportunities, and build a better and more inclusive future.

And that's exactly, what we are going to talk about today. Let me check whether, Sandra is here. If not, then I would just, yeah, go in the into the presentation. So before we dive in, let me quickly walk you through what we'll be covering today. So first, we'll start with an overview of multi model gen AI. What it is, why it matters, and how it's enabling more intelligent and context of your systems. Then we will shift our focus to the real world use case, how enterprises are really using it and already seeing the value through it. And finally, we will explore the critical role of diverse voices, especially, for women in tech and why, it is important for building ethical and inclusive AI systems. So with that, let's get started. Cool.

So, as we all know, we are living, through a pretty incredible moment in tech, where generative AI has gone from being a buzzword to becoming a real force, reshaping how the businesses think, operate, and compete. Across the industries, boardrooms are buzzing with one question, like how do we leverage, this technology before our competitors do? And these numbers are hard to ignore. Around 87% of the jobs are expected to be impacted by GenAI. And that doesn't mean, that the jobs will be replaced, but it means that reemergent. And we are talking about, a potential 7% of boost in global GDP. And perhaps more strikingly, the gen air market is projected to grow to $1,300,000,000,000 by 2032. So these aren't just statistics, but, the signals signals that enterprises must not only pay attention to, Gen AI, but also actively prepare to harness it. So the real question isn't if Gen AI will affect your business. It's about, like, how quickly you can adapt and start creating value through it.

So how do we actually make, Gen AI work in practice? So here's the thing. We don't need to start from scratch or replace everything that we have built so far. But combining what we already know with what's now possible, which means bringing together the strength of traditional AI and creativity of Gen AI. AI. So traditional AI has been doing the heavy lifting for years, whether it's forecasting models, recognizing patterns, or driving data driven decisions. Gen AI, on the other hand, brings the new level of intelligence, the one that can generate, summarize, and communicate like, fluency, like a human being. Now when we bring these two together, we unlock a symbiotic AI relationship where Gen AI can generate the insights and traditional AI can validate forecast and optimize them.

So this isn't about better tools, but it's about, amplifying outcomes and accelerating transformation. Enterprises that master this hybrid approach will move faster, think smarter, and act with greater precision. But in order to truly scale, the value of, of this, we need to look into the the fuel behind this systems, which is data. And not just any data, multi model data. Today's AI models can go far beyond text. So they can, understand and generate code, images, video, audio, and even scientific and domain specific formats. And, that that means we are no longer limited to spread sheets, Sempna, and dashboards. We can build systems that interpret the legal documents, generate the marketing visuals, and even support r and d all in one ecosystem.

And this is where, we start to see the rise of agentic AI. The systems that can reason, act, and adapt across the different format, choosing the right approach, for the task at hand. And the message is very simple here. And more diverse your data is, the more intelligent your AI. And the more intelligent your AI, the more autonomous and impactful, your enterprise becomes. And with that foundation in place, I'll now hand over to Sandra who is now, here. Thanks, Sandra, for joining and who will walk us through the enterprises, which are already putting these capabilities into action. So, Sandra, if you would like to go in the quick introduction and then yeah.

Can you guys hear me? Yeah. So what actually happens, and we've seen that with our clients quite heavily, is, the agents are everywhere. Right? We're used to chatbots now. They're becoming more and more intelligent. We're used to our little apps helping us, and then Siri was the first one to help us, you know, find things on our iPhone. But also in the enterprises now, it becomes more and more data driven and combining tasks together. And agents are, at the forefront of everybody's minds because they can combine not only search functionality, but different datasets and then also put things into action. Having said that, it's a very, very fast progression that we're seeing.

I mean, I'm in the data and analytics space for quite some time now, and I have not seen a single topic moving so fast like AgentiQ AI. Last year when we started with JetGPT and the gaming models of of Google, it was everybody was very excited on how autonomously they can help us answer questions but also generate. But now it becomes in the workforce a little bit more tricky to actually evaluate where do I need, support and how can real agents, and in that case, then multiple agents coming together, help me in my day to day business. So if you go to the next slide, Kipna. We actually started, last year with one of our clients, our prominent bank, to help them not only with a chatbot solution, but with an intelligent chatbot. Right? And the the question was, is the tech secure enough to put an LLM into a real customer facing environment?

And so far, we've been seeing if you guardrail what the chatbot is doing and the LLM is trained on very carefully to make sure that it runs ethically, it runs on secure datasets, it runs in your environment, and it points to the right solutions where the answers are coming from, yes.

You can. And you can also do that in a very, very fast manner. It comes with limitations, of course, because as you all know, LLMs are evolving very fast, and you have to make sure, especially in an enterprise environment, that you guardrail what the answers will be. So we tested that with the service chatbot, that runs with Gemini in the back, the Google model very successfully. It points towards self serving, websites and services that, the company already runs on. And if there's no answer, it, automatically transfers to a real agent. So there's also a fallback. If the if the customer doesn't understand, he will be redirected. And the outcome is really impressive because there's only out of the test environment and then now it's been live, there's only a very, very few instances where the bot will not help the customer.

And that's ultimately what you want. Right? That's ultimately where you wanna go through. It serves as a helping hand to real people. And if you go to the next slide. And those of you who are in, in the past bank know what the chatbot is about. So if you're a customer, just click on it if you're already one of the few ones that can access it. But, it goes to show that if you do that, you have to have a ring fenced environment. You have to make sure that the bot doesn't hallucinate. You also have to make sure that it's with an authenticated user group because whoever has a mobile app on it downloaded is already pre identified to be a customer. And then, of course, and what I what I said, it's responsible design.

So even if somebody wants to, like, you know, out task the bot and say or goes badly with bad names or bad things to it, it doesn't answer. It's very polite, but it says, I I will not answer your question. So this is actually something that we have to keep in mind when we talk tech that whatever you feed the system with comes back. Right? So if you feed the system with really nice things, it will come back with a very polite answer. But if if you want to abuse it and it's part of the testing cycle as well, you have to make sure that it runs according to the code of contact of your enterprise. And that's something, that has been underestimated right from the start when the first LLMs came out.

It gets a lot better because all the companies that are providing LLMs make sure that they train the models, to along ethical standards. But it goes to show if you wanna use that in your own company, you have to even guardrail it much, much more than than just using it for your private use. If you go to the next slide. We're, obviously, from a consulting firm, and Deloitte is very, very keen on helping our clients putting tech in place that, follows certain standards. So we always make sure and the bot that I just talked about also does that. There's always a human in the loop. If the bot cannot answer, it goes to a real person. And that's something that you have to basically bake into your processes that you wanna automate.

At a certain point, you still need us humans verifying what every bot and even a multi bot system will do. Because at the end of the day, we are, as employees, responsible for the answers that the customers ultimately get. And it it works in consulting the same way as it works in a bank. You're the face to the customer and your bank's responsibility or your company's responsibility is on the line. So you have to make sure you're acting according to code. Manage hallucinations. I mean, the the LLMs are getting faster by the minute. There's very specialized LLMs now out there. And, yes, in former times, they couldn't calculate properly, but now they become a lot better. So you have to look into what model serves your purpose.

And not even the bigger ones, can do everything right now, and they've been trained quite a bit. But sometimes and that's also learning. You don't even need an LLM. You know? You don't need something. You can just use a normal AI model for doing the same and maybe also cheaper. Everybody talks about the consumption of energy and resources that we utilize going forward for the agentic area, and that's something that we have to keep in mind. It's like when you only have a nail, everything everything looks like a hammer or the other way around. I don't remember the saying, but, you shouldn't just look into the sake for for using Gen AI for Gen AI sakes. Sometimes there's a cheaper and a better version to doing it. And process automation has been on the planet now for quite a while.

So some of the solutions you're looking into can be solved differently. Transparency and explainability. That's something that's in the forefront of everybody's mind because we have to explain to, especially in a regulated environment, how we do certain things and how the system comes up with answers. And that also builds into the trustworthy AI principles where you say, look. We are not doing anything unethical. We're making sure that the LLMs are trained on on enterprise data properly, but we also are gonna make sure that going forward and and the EU is a little bit special when you look at the the the globe that we won according to EU AI act. And that's something that we as a company put in the forefront of everybody's minds when we're implementing tech like this, that we're following the principles and we are not violating any rules and regulations that are currently also applying to the EU. If you go to the next slide. And, here's the the last part. And, Ketna, you can also talk to this. Right?

We both are very passionate about increasing the share of women in tech and making sure that, AI professionals, especially also people that work with tech, are, being treated inclusively, that you make sure that the biases are out of the datasets that you train with, and that AI professionals, are being utilized properly in not only for research purposes, but also in the diversity perspective that comes out of those systems.

I I mean, you have to make sure and everybody knows that right now that, combined and cultural diverse teams work better. But you have to also make sure when you work in teams that actually the results of those systems work according to what what the company values are. And this this is just goes to show that the future of AI is definitely diverse. Right? You can, of course, manipulate certain things in certain ways, and we all know that on social media. The algorithms are trained on what you feed them with. But in an enterprise, you have to make sure that, whatever is the outcome is the desired outcome that goes according to the values of the company. And that's only achievable with mixed teams.

At least that's what I'm I'm definitely striving for in in my teams, and that's also what we should be emphasizing on going forward.

Yep. Thanks, Sandra, for saying that. And I would also like to, add, one example here that, if, we talk about last few years, I mean, it has been incredibly good in terms of, inclusivity, in AI. However, like, 10 back, when, I think, Apple, they they headed this application. I don't remember the name, but the health app, and they did not include, let's say they included all the features, like, how well you are sleeping, how good is your heartbeat, and everything, but they did not include, the feature for for women's menstrual, cycle tracking. And that itself says that, I mean, if we talk about the broader picture, in the past, there were not, the data was not really inclusive. And that's why the models were not inclusive.

But now, it is it is getting really better that women are shaping the future, for better, and we need to have a female, at each level. So starting from the junior level to the leadership, so that, the the model that we build now and in the future are not biased. So, yeah, this is really important. And with that, I think, yeah, we are through, the talk, and here is our, yeah, our contact details. Also, QR code for our LinkedIn profiles. Please reach out to us if you have any questions related to the session or even, like, outside of the session if you are interested to know about, more use cases. We can definitely connect outside of this. But, yeah, great to, have you, and thanks for joining our session.

Questions, Hal? I'm very welcome. Yeah. Thank you.