So much. Well, hello, everyone. Thank you for joining me today.I hope you're having a great time at the conferences today and are learning a lot of things from different, presenters. My name is Anas Gerson, and I'm a senior manager at I lead our, AI enabled transformation programs in the AI and data practice. Today in this session, we will talk about the rapid evolving area in AI, which is basically agenda. Yeah. You must have heard a lot about this, and how it is poised to reshape the consumer packaged goods and retail sector, essentially. So, you know, as you all have been witnessing, we are at almost at a pivotal point in the evolution of artificial intelligence. Right?

Over the past few years, we have seen generative AI go mainstream, where we were, you know, basically generating text, images, even strategies, all within seconds. Right? But we are now stepping into the next chapter, which is agentic AI. Now you would think that why does it matter for CPG? Right? Because CPG is a very complex, you know, industries and or sector, if you can if you can call that. It has a lot of, you know, complex decisions that needs to be made throughout its value streams from, say, you know, what products to launch. Right? How to personalize marketing, how to run promotions, how to reduce waste or meet sustainability goals. It's very complex as well as it's very, you know, fast ecosystem as well. Traditionally, all of this requires, like, an army of planners, analysts, and marketers. But now imagine intelligent aided agents augmenting or automating these tasks at scale almost twenty four seven.

So today, what we're gonna do is we talk we'll talk through this. We'll talk through what agent API really is beyond the buzzwords, how do you carefully apply it in your value streams or business functions, some of the real use cases that you can act on today, and some of the guardrails that we need to safely scale it. So let's get started. Alright. So let's take a quick look on the evolution of artificial intelligence. It's also important for us to understand, what different term means, right, in the in the realm of artificial intelligence. It began in 1950, somewhere in September with rule based systems under the umbrella of artificial intelligence. So, basically, any machines that mimic basic human, you know, logic or, you know, kind of like automating some of the task was termed as artificial intelligence.

By late nineties, machine learning entered the scene, which was enabling algorithms to learn from data without explicitly, you know, being programmed for it. Then came the deep learning in 2010, which is powered by, you know, neural networks unlocking, you know, capabilities in language, vision, speech that we now use almost every day. And today, in 2025, we stand on threshold of something new, which is agenda key. But now you'll think of it or you'll ask that what where does generative AI fits in and, you know, what makes agent API different from generative AI? So generative AI really focuses on, con you know, content creation like drafting emails, generating images, or composing music, etcetera. Agentic AI takes it a step further, in fact, several steps further.

Agentic AI is very well, you know, self governed, meaning they can, you know, set sub goals, reason through steps, you know, use APIs or databases, and execute complex tasks autonomously with or without human input. This means AI is no longer just an assistant to help you find information, but it's becoming a collaborator that can manage marketing campaign campaigns, optimize inventory, or even codevelop new products with you. And this shift is really happening fast. According to Gartner, by 2028, just three years from now, a full one third of all generative AI interactions will use action models or and or autonomous agents to complete the tasks. Alright. So let's understand agent a little bit more in details, and it might because it sometimes, you know, the first might sound a little bit complex, but it's very simple.

It it almost mimics the way we think about, a certain problem or a task. It all begins with, you know, a natural language input. This could be a command or task or even a vague request from the user, like optimize this campaign or forecast next month's demand. After this input, is received, the AI agent then integrates this input and applies reasoning into it. It starts to ask where does like, what does it really mean? Like, what is the intent of this question? What constraints apply or rules apply to that? You know, this is where the agent agent's memory and context awareness comes into play, which has been trained with large corpus of data. Right? At the end of the day, everything, you know, embedded in the agent to KI is a little models, which is, again, the whole, corpus of data has been trained on it.

From there, the agent doesn't just answer. It starts to plan. It breaks the task down into, like, set of actions, tools, APIs to cause, any subtasks to delegate. Think of it like a project manager. How would a project manager start to delegate task to a team member? It starts to think like that. And then it, execution comes into play. The agent carries out the steps like accessing data, sending messages, triggering automations, and interacting with other systems or agents just like a team member would. Right? The end result is always, you know, a clear outcome, Whether it is a report or updated dashboard or, completed purchase or a launch campaign, the agent will really deliver the result without constant back and forth. And just like humans, these agents also starts to learn. Right? They are just based on outcomes, feedback, new data, often using reinforcement learning.

And over time, it will start to become, like, sharper, faster, and more aligned with your goals. But it is also important to note that human intervention or it will still happen at any point of the time, whether to validate the validated or, you know, approve or redirect the request. Your human in the loop is also very important. Now it's it really depends upon, you know, the manager or whoever is controlling the agents. They can design it to be collaborative. They can design it to be totally autonomous. You know, however you want it to be. So now you can understand, like, when we say agenda k I, it's not just talking about some smart chat bots like we are used to now with generative AI. You're talking about intelligent systems that can interpret, reason, plan, act, and learn all in the service of real business, outcomes. Alright.

Let's talk about the evolution of agentic orchestration design, which is really evolving rapidly, not just because what it can do, but it, but in the way how multiple agents can work together. So let's explore this journey. It all starts with a single agent, which is in the simplest form of agent API. Here, one agent handles a linear task like summarizing a document or pulling inventory data. It's efficient but often very limited in scope. Think of it as the early phase of AI task automation. It is fast, focused, but without collaboration or any kind of adaptability. Next comes the two two agent chats, chat models. Now here the agent can talk to each other, enhancing decision making through dialogue. For example, one agent can, generate a product idea and the second l can evaluate its visibility. The interaction layer can make the system much more dynamic and flexible.

Then comes the third, stage, which is agent group chat, where multiple agents come together to solve a problem collaboratively. Now in this setup, you may have agents specializing in, you know, pricing, logistics, or marketing, and they're all communicating in a shared workspace or maybe reporting up in hierarchical structure. Now this kind of system is much more powerful, but it's also much more complex and often require, very smart coordination between them and a lot of shared context, which is often enabled by, your knowledge graphs and, a lot of unstructured corpus of data. And finally, we reach to the, multi agent orchestration, which is the most advanced and scalable model. Here, many specialized agents operate in parallel or in sequence, when they are orchestrated to complete multi phase task, sometimes pretty much end to end. And each agent have, or may have its own workflow, but together, they all deliver a unified outcome.

So think of it as, a CPG company deploying agents for demand forecasting and supply chain optimization in the supply chain function, content generation in, say, marketing or customer support in commercial, but all are working in the harmony towards, you know, productivity led revenue growth.

Right? So now if you zoom out a little bit from this, flow, you would, see that this evolution is not really just technical. It is also very strategic. You know, by understanding and designing for these steps, organization can deploy the right agentic architecture for the right level of complexity and skill responsibly over time. Alright. Let's move to the next slide. Now I've talked all good things about agenda k I. And while it is amazing, I also wanna caution everybody, a little bit in the sense that not every use case is agenda k I, very at least not for right now, or not right away. It it it takes a lot of resources and money to build these systems. So we have to be mindful of, like, where are we throwing these systems on. And to, you know, ensure value and avoid over engineering, you know, we need to be thoughtful in selecting the right right starting point.

So now your question would be, where do we start from? Right? Here are the four class criterias that we have built, which is basically around making a process a strong candidate for, you know, agent API. And it's a very, you know, high level criterias, on it. First is the volume. Agent API is based, is best applied essentially where there are high volumes of interactions. For example, CPG company processing thousands of orders daily, is, you know, better fit for, this agenda KI system than a low frequency, high complexity, process like, you know, aircraft manufacturing. And you would say why? This is because the repetition amplifies the ROI of automation, and agent tends to learn faster when they are thrown a lot of examples. Right? At the end of the day, it's reinforcement learning. So as much as data you can feed to it, the better it becomes. Second is the interaction across systems.

Now great candidates are processes that require pulling data from multiple systems. It requires updating, syncing, systems like CRPs. Sorry. CRMs, ERPs, and, you know, inventory platforms simultaneously. So if a human is spending time moving data or reconciling data between systems, that's a sign that it could be, a good candidate for agent and agent can step in and and do the tasks. Third is the human touch points. These are the tasks that require someone to read, reason, make minor adjustments or judgment calls based on, you know, multiple documents or data sources. So if that work is rule based or repetitive, like order exceptions or compliance checks or, you know, responding to routine inquiries, then agent KI can take it over and at least assess communicable. And the last is, the error impact. Some processes are especially vulnerable to human errors, like data entries and things like that.

And when those error happens, the consequences can be costly, things like also, like, delayed shipments. Right? Missed SLAs, lost sales, or compensation claims, etcetera. And in those cases, agentic AI not only increase the efficiency, it also reduces the risk and protects the bottom line. So the take takeaway over here is that focus on use cases that are high volume, multisystems, human heavy, and are very prone to error or error sensitive. That's where your agent API creates the most immediate value, at least for now, and sets the stage for the scale transformation. Alright. So now, you know, let's talk about what really matters to our stakeholder, which is the value. Right? When we apply it in data care thoughtfully, we typically see back across, like, three common, you know, dimensions, which is productivity, process efficiency, and quality. Quality.

Each of these areas are contribute the each of these areas contributes to, like, tangible business outcomes. So let's let's walk through what those are. First is productivity. Sorry. First is productivity. So agentic AI helps reduce the time humans spend on routine or repetitive tasks. This is the time, we can give back to our business, whether it is for the higher value strategic work, or, like, customer engagements or innovations. Right? And in real terms, it basically means that more hours are returned to the team's back, lower full time employees' tie time is, spent on per process, improved, employee satisfaction and engagement. There's also higher revenue per employee, and it can even increase, upsell and cross sell opportunities that TST refocuses on, customer growth. Next is the process efficiency. This is about the speed, essentially.

Agent API can really shorten the time, or the cycle time, whether that's, you know, resolving a customer query, approving a claim, or, you know, completing a pricing updates. And these metrics can be tracked by, you know, better, customer responsiveness, fewer lost sales due to flow processes, higher net promoter scores, fast inventory turnarounds, or, again, more effective upsell and cross sell through the real time responsiveness. And the third is quality. This is very agenda key. I, you know, reduces friction and risk, by minimizing human error in repetitive tasks. It prevents downstream issues, that cost time, money, and trust. And we typically see that in, you know, things like a drop in compensation claims and, complaints, less time spent managing issues, fewer noncompliance violations, fewer manual interventions that lead to delays or cost corrections. Right?

So, you know, in short, agent API helps you do more, faster, and better, and it delivers triple benefit, freeing up human capacity to accelerate, to build, and improve accuracy, all while driving measurable business outcomes. And the best part is that these, KPIs are not just technical. These are c suite, metrics, which basically incentivize our top level 10 level leadership to invest more heavily into the agent in generative AI, processes and transformation. Alright. So let's talk about the five ways CPG can use agent AI, and they can use it right now. So all of these use cases that I'm gonna be talking about are being used as we speak in the CPG industry, and they have been, you know, promoting growth revenue, growth essentially, due to the productivity increase. So, you know, let's talk about all five of them. First is autonomous product innovation agent.

These agents can, continuously, like, scan social media, social trends, consumer reviews, and competitive activities to detect unmet needs and can generate new product ideas. They can simulate ingredients, in a combination, predict success, and even recommend go to market timelines. The impact from this is that the brands are, dramatically, you know, speeding up the innovation cycle. They're reducing the r and d time or r and d cost as well, and they're also launching products that are more likely to succeed because they are backed by live real world data and not just, you know, intuitions. Right? The second is demand sensing and inventory agents. These agents, use a wide variety of data from point of sale to weather, event calendars, and promotions to anticipate demand shift in the real time.

They then cannot anonymously, you know, adjust the supply chain plans and reroute log log log if needed. The result to this is, you know, reduced stock out, less waste, lower build holding cost in a smaller linear supply chain that can respond in hours, and it does not take days, for them to get respond. Third is AI brand managers. These agents go beyond personalization. This can be used into in marketing. Right? They go beyond, basic personalization. They go into hyper personalization. They craft and deploy content across channels dynamically adjusting based on what resonates with each microsegments. The business impact out of it is, that it's it gives a significant boost in the engagement, conversion rates, and customer lifetime value, all while lowering the acquisition cost and reducing manual marketing efforts. Enterprises in in CPG are spending millions and 10 tons of millions of dollars in marketing efforts, and and are not able to scale as much as they want to.

So having an AI brand manager can really, really scale and and, you know, reduce the cost of it. Fourth is sustainability, optimization agents. These agents continuously monitor packaging decisions, you know, manufacturing, you know, to use, logistic emissions, and they recommend great and greener alternatives, simulate trade offs, you know, and track ESG compliance in real time. And the impact of this is twofold. Meet they meet sustainability target while enhancing brand deals fast, and they also reduce the risk of, regulatory penalties and consumer backlash. So it can contribute to your cost cost avoidance as well. And finally, the, dynamic pricing and promotion agents. These agents, analyze competitor pricing, you know, consumer sentiments, real time sales data. Then they autonomously adjust price and promotions, across regions or channels. Through this, you can gain, optimized margins, increase promote promo ROIs, and, you know, faster, reaction time in volatile marketing environments, all while freeing up pricing themes, need to focus on strategies as opposed to spreadsheets.

So together these five use cases demonstrate how JETiKARE can be deployed today, not just to automate, but to elevate, the intelligent across the entire value chain as well. Alright. Let's talk about some of the risks and challenges. I mean, as any with any technology, any powerful technology, you know, it's it's always bring opportunity to to use that, technology much more responsibly. So it's important to recognize that while the systems are you know, the systems can, like, plan, reason, act, they must do so within controlled, ethical, and, you know, aligned boundaries. This slide basically breaks down the risk and challenges and, you know, that we must manage and the strategic consideration that we must, you know, keep in mind as we scale, each and every care responsibility. I'm not gonna drain the slide because you must have heard a lot of the sessions around, you know, responsible AI and governance and everything.

But some of the areas that I wanna highlight here is that, you know, autonomy. We need to ensure that agents act within the intended scope. Without guardrails, there's a risk of agent making decisions that is outside their scope. And, and and we also have to make sure that there is a process laid out in the system to reverse those, or or at least have a human human intervention to, you know, just change those those decisions if those decisions are not are basically outside their remits. Then there's bias and fairness, transparency and explainability, which are basically the part of responsible AI. We have to make sure that, AI models, can basically inappropriate any, bias into the data because, it all comes from the data itself. So they can have bias into their decisions as well, and they can be unfair, to some of the customers and employees.

We have to make sure that human are in the loop for that as well, and they're using and they're using, AI much more responsive, responsibly as well. So how do you how do you manage these challenges while still moving fast and, you know, innovating? So it comes down to, like, designing the system very, responsibly from the start. So, you know, think of it as strategy alignment. AI should serve your business strategy, not the other way around. Right? You have to define clear goals and boundaries for, you know, every agent. Then there's data strategy. Your agents are as good as the data they're trained on and they're exposed to. So ensure data quality, diversity, and governance, those are gonna be the those are those should be the top priorities as well. Third party integration.

Agents often interact with, you know, multiple system and external systems and APIs. So those integrations must be secure, scalable, and auditable. And, obviously, the continuous learning and ethical framework should be in place, so that, you know, both the technical and human loop are in the feedback, and then they are also building policies around transparency, accountability, privacy, and fairness. And then and we should also revisit them a lot. I'm working with, you know, I'm working in creating a governance two point o because earlier we used to create governance around generative AI, but now it's becoming agentic AI governance, because now agentic AI is gonna be taking action. So we have to think from that lens to, you know, modify our current governance framework. Alright. So as we close, I want to zoom out from the technology for a moment and come back to what this is really about. Right?

This code, by Sam Walgreens is the best. It's no longer about having the right answers, but asking the right questions. We are basically moving away from a world where, you know, our time was consumed by sedating tasks that are repetitive, manual, mentally draining, to to, you know, mentally draining work that left little to no room for imagination or growth. Right? So now we are moving from a system that is that rewarded ability narrowly defined by what you used to know or what you basically know essentially, to one that now rewards agility, curiosity, and adaptability. Then moving from collectors of the fact, you're just basically trying to remember everything, to being the connectors of thought where we are, you know, our greatest strength is recognizing patterns, understanding bigger pictures, and solving more complex problem in new ways.

And lastly, I really wanna highlight that agent AI doesn't really replace human human advantage. In fact, it amplifies it. It frees us to focus on what makes us distinctively human, which is empathy, strategy, innovation, and judgment. So as we design these systems, as we integrate them into our businesses, let's remember that the goal is not to, you know, not just to have faster execution, but, you know, smarter thinking, better decisions, and more meaningful work for people behind the processes. Alright. That wraps up the session, and I'd like to open the floor for questions. Let me see if you have question on the chat. Alright. Any questions? Alright. So I'll thank you, Shazia. I'll close the session now, but if you have any questions and you would wanna connect with me, feel free to reach out to me on LinkedIn.

I'll be more than happy to, you know, take take a deeper dive in the session in the, in the session and also in the topic itself, and talk more, you know, on on this, agility as well with you and especially in the consumer and, retail sector. Thank you so much. Appreciate the appreciate your time for the