[Music] [Sadagopan Singam] Good afternoon.
It's so heartening to see a large set of participants here post-lunch. So very grateful to you for being here. My name is Sadagopan Singam. I'm based out of Silicon Valley, just across the street from Adobe headquarters.
I'm part of HCLTech, and I run the Global Commercial Business for HCLTech. Part of this ecosystem for more than 20, 25 years working on the digital ecosystem.
And joining me today is my colleague, Shilpa. Why don't you introduce yourself? [Shilpa] Yeah. Hi, everyone. I'm Shilpa. I'm part of HCLTech Customer Experience Practice. I lead all the fun things in customer experience, personalization, AI, automation, analytics. So we are here today just to speak about few of those topics. I'm sure you've heard a lot about these topics yesterday, today, and tomorrow. And we'll try to keep it interactive. Please feel free to ask questions whenever we are on any slide that you see fit. Thank you, Shilpa. So with that said, we look forward to make it as interactive as possible during the session, post-session, and obviously, on a continual basis.
I think the theme of this conference is agentic and sitting in the Silicon Valley, I do know how much things are progressing around the world. And also, when you look at the interest in this theme and the early adoption and the degree of success in the pilot initiatives or even fully-blown initiatives, we are here to share some of these themes.
I think the first thing that we want to start by is defining what we think is agentic AI in the context of both broadly, as well as in the context of digital marketing.
So one is, we do believe that the best use case for agentic AI obviously lay in the front office. So we've seen our research probably 18 months back when all of these GenAI things were evolving up. 70% of the use cases for early GenAI adoption will happen in the front office. So that was very clear to us even 18 months back. So now as we begin to tighten the applicability criteria and see where this fits neatly. We realize the digital marketing space is a space where it is pregnant with opportunities, and it has got a very good upside there. And therefore, it becomes important for us to see how this fits in overall and where the convergence is happening there. First is with agentic AI, we are seeing that ability to uplift the customer experience goes to norm exponentially. It happens because partly it takes control of both the interactions, the process that come together, plus the ability to take autonomous decisions. So this is real to be seen. We are seen in many opportunities, at least I would say, in half a dozen opportunities. We can see a factor of five or six times improvement from immediately felt there. And we'll explain in terms of how this all comes together as part of the session itself. And when you're able to enhance that type of experience in the digital platforms, it's going to have massive effect to all of the stakeholders who are experiencing those effects of leveraging the digital platform. And that is where we believe that the greater upside is. And how do we get there? We get there fundamentally because we have four things running in parallel. For this agentic AI is to feel the degree of versatility and the empowerment it provides to enterprises. First is being contextually aware. And this becomes very, very important. The same agentic AI can work in different formats, in different functions depending on the context. That we believe is extremely important, which means that it's highly self-modulated to the opportunity at hand to the context that we are discussing. The second is the ability to detect multiple patterns. Sometimes we tend to go for familiar patterns, but we see that these agentic AIs can look at multiple patterns and clinically look at the possibilities and rule out or take traverse the path which will get the best possible outcome in the given context. So that becomes the second differentiation. The third thing is it's multimodal. So it is not going to be something to be integrated in an app. It has got to face the real life in terms of leveraging everything that's available across channel, across different types of aperture spaces, as well as different types of multiple agents. Eventually, the overall architecture will mean that we'll have multiple agentic AIs coming together to drive the enterprise functions. Therefore, the ability to move across multiple modes, including leveraging multiple types of LLMs or multiple types of secured enterprises. So these becomes our secured trusted networks. So all becomes extremely important to provide the best-in-class service to the enterprise. That's where the multimodality is going to be helpful. And fourth is adaptive learning. It should not be repeating the same thing again and again and again. This is not a recording device. It has got to keep changing with time. It has to learn. It has to self-learn. It has to self-heal. As well as drive better outcomes time and again. So these become very critical as we begin to incorporate this in the digital marketing space. But none of this will become possible unless there is a vision set out in terms of what to accomplish. For example, should we lift 10 pounds? Should we lift 20 pounds, 50 pounds, 70 pounds? So these are all things that are directly correlated to the overall vision that we have in terms of what we want to accomplish. This is where HCLTech fits in. See, our philosophy in the last 18 months has been anchored around a particular framework called Total Experience. What has happened is, we were coming out of the pandemic and post-pandemic, we saw humongous adoption of technology process, investment sampling, and the customer expectations and the customer's customer expectations have substantially changed. And that when we began to do a root cause assessment in terms of what could we do to make sure that it becomes very endearing to our customers. What we realized when we looked at across industry around the world, what we realized was there are multiple strands of experiences, enterprises take to customers. Sometimes it can be our customers or any stakeholder for that matter. It can be an employee experience. It can be a customer experience. It can be an user experience, or it can be experience of different kinds that enterprises provides to their stakeholders. And also what we realized is when we began to assess the maturity and the adoption levels of each of these layers of experience, what we realized was these were all in different levels of maturity. Some was probably very good. Somebody were catching up. So these different strands had its own characteristic. And what we realized was the best-in-class enterprises began to have an approach where all of these multiple strands of experiences come together in a harmonious way to provide a synergistic uplift to the overall experience the stakeholders can experience. And that becomes the defining norm in terms of the best place to leverage all of the digital platforms and get the best experiences to the customers. And therefore, our framework in two enterprises is essentially anchored around what we call as some total experience framework. When agentic AI and this total experience moments come together, we are talking about a tsunami happening there in terms of the exponential power that both these inherently bring to the table. And when these combine, we are actually seeing a very humongous upside in terms of what enterprises can achieve. And very shortly, in a matter of few minutes, we will get there to see what are the possibilities. But before going there, I want to make sure that we have a very good view of where to start on agentic AI. So sometimes everybody starts with the end result, the end state in mind, and they want to work backwards, which is very good. But several times, it may be punching shorter than where you can go unless we understand how this all come together and therefore set the right position in terms of where we can actually reach progressively. So the first thing that we see is, we want to make sure that like any other fad that comes and goes. There must be an ROI anchor any of the agentic AI investments in general. And also, we believe that as this movement has moved at a rapid pace more, the pace of adoption has been much more rapid than what we could do imagine basis, whatever we had seen earlier trends catching up. We are now suddenly come to a stage where it is not only horizontal applications, it is actually becoming more and more verticalized. So we are now seeing in real life, about many verticalized AI agents that are coming into the fore. For example, let's say, utility customers provide to their customer or the experience a bank provides to the customer, they are fundamentally different given the sensitivity of the information there. Therefore, there's verticalized agentic AI agents will become very, very important to keep in mind when you draw the vision for the enterprise.
Also, as I mentioned, the ability to do multimodal becomes very important because the way it is going is you will be able to bring any LLM. You'll be able to bring any model. You'll be able to build any reasoning engine. You'll be able to bring any type of trusted network to come together. Therefore, this multimodality becomes extremely important. Therefore, when you set out a vision on the agentic AI adoption for enterprise, make sure that we factor in this adequately so that we'll be able to get the right set of an upside there. And we also realized that, change management becomes very essential because the decision-making and the speed of processing will call for an understanding and an adoption that will not going to happen in the normal course. Therefore change management becomes very essential. Also, the fact that it doesn't mean that we take the human oversight out of that. For example, I was just joking with a colleague of mine in our own HCLTech booth. We invite you to visit us as well. So we are trying to have a digital way of drawing a caricature. We are trying to compare it to manual process. And somebody was telling me the digital caricature process perhaps had more manual interventions at the back end than what the actual manual caricatures had carried last year. So that means that you don't completely take the human out of the equation. There is going to be intervention. What it is? What is the judicious blend of that human intervention with what agentic AI can drive is an art by itself, but we need to make provisions for that. And also, responsible AI consistent with the enterprise standards becomes very essential. All of these have can be brought together in a very seamless, efficient, elegant way. And that's the submission that we want to bring it to. Also, we see that either we can act at the infrastructure layer or the application layer. But I do believe that we need to act at both levels to get the best possible set of returns for the enterprise. And also in doing this, I want to quickly talk about what is it that we have seen. So because, this space has suddenly evolved at a pace that it is very difficult to put a finger on one thing and say this is what it is. Therefore, we created an organization within HCLTech only to focus on how the effects are being felt across the world, across enterprises, across geographies, across organizations of varying degrees of maturity. And what we see was essentially from a total experience standpoint. We see that, you can do either bottom-up or top-down strategy. This becomes very, very important. See, one thing is these tools are so handy to essentially absorb inside enterprise.
It's easy fact, it's easily digestible. Therefore, several organizations take a bottom-up approach. But I think there are farsighted organizations which would also look to do top-down. But what we really suppose to know, it is too early to say which one is right, but we definitely see these two can seamlessly come together. And we are seeing examples that you're going to describe. One of that is part of another session itself. Also, we believe that from a digital marketing standpoint into the realm of the domain that we're all interested in, we do believe that from a wall-to-wall end-to-end perspective, there is a efficiency gain of something like two-thirds overall, if implemented elegantly with the plan with a reasonably well-settled vision. So this is the first thing that I would say is a message that we want to share authentically out of our own experiences. And also, we believe that it is coming out of three things. One, you got the cost angle. So we are seeing close to 50%, 50 plus percentage in terms of an optimization possibilities and measured or an 18-month period. It is not something that you're going to realize day one. Or an 18-month period, we are actually seeing this type of benefit as well, which means that you get that much more money to essentially plow back for additional initiatives, or you can make it a self-funding initiative. So we call it as the break-even point where the initiative pays for itself. And that is a scale that we can possibly get in 18 months. The third is in terms of, is it wide enough? Is there a pocket that is resisting? Is there a pocket that is not interested, that doesn't see the value? And we are very, very positively surprised that we couldn't find many who actually were willing to try. So we're just kudos to the digital marketing community and the leadership and some of which is present here to absorb these type of changes much, much faster. And I think another thing that has happened is because of these types of technologies being put in place, the fundamental way the customer's customer or stakeholders interact with our customers are completely changed. So essentially, agents can actually even resolve decisions faster. That means the number of queries that they need to give to the system actually comes from. That is dimension one. Dimension two is because it's multimodal, it can support search from different types of channels. That's the second part. And the third part is because of the way that, agentic AIs are beginning to get at the front of working on issues, half the searches that used to go to traditional channels are actually getting routed through agentic AIs. So this is very, very important because if this is the level of engagement that we're going to see the stakeholders have with our customers, with enterprises, it means that we need to engineer this very well and have a mechanism to review how this is all performing and also be able to meaningfully capture at an overall level the comprehensive level of impact it's having on in terms of performance and in terms of providing solutions to our stakeholders. And how do we do that? I think what we see is in all of this, the first thing is we need to get the right content creation modal. So one of the thing that we have seen is in the way that organizations have built up their digital marketing paraphernalia, we see that the traditional ways of creating content, setting up the infrastructure to create the content, that led to definitely change.
The traditional way of putting structuring, putting that taxonomy, putting that metadata and tagging everything, that's all great. That discipline really helps. But the pace at which the business is moving on the type of queries this generative AI can essentially solve for customers means that there's a traditional content creation mechanism that will probably had to loosen up a little bit. That's the first thing that we noticed when we saw organizations scale up on these initiatives. The second is we also realized the data, as well as the data about the metadata analysis, as well as the traffic and the transactional data analysis of them. That was always very difficult. Therefore, given the scope of what the agentic AIs can do, the ability to create synthetic data becomes very essential. Therefore, that is something that we need to invest in before we go into a full frame, agentic AI deployments.
Third thing that we noticed was in the traditional world, we saw the degree of personalization these agentic AIs can bring to customers. It can't be replicated in the traditional world. Soon, if you don't embrace this, the point here is if you don't embrace this, the differences and the delta that we have on these attributes will begin to show up. Therefore, it's not only about creating the future, but even protecting what we have will become an issue in terms of the impact we can have with the stakeholders. Therefore, it will be essential to see what are the different attributes on which we will actually see this make a big difference in terms of adoption. And the fourth is the product development cycles in terms of essentially moving from one stage in digital marketing to going into the next stage. That had a certain cycle. That had a certain release management, the principles. And that had a certain frequency. Those will substantially change when we move to the agentic AI world. Partly because the careful steps that we take to make sure that we cater to every invisible scenario is automatically digested by the agentic AIs. And then the manual processes, as well as the limited capabilities in analyzing the data all will actually, these limiting factors don't apply in the agentic AI world or the capabilities they serve for the end customers actually mean that these limiting factors you need to overcome. So this will become very important as you begin to design your agentic AI. I'm going to leave a lot of time for discussions here. Therefore I want to make it relatively short in terms of what we want to cover here. So very quickly, I think how do we go about doing this, right? So one is, we are now seeing a broad section of industry use cases from a consumer standpoint in terms of how they can adopt agentic AI. So this is now almost getting to a codified standard, which means that there are enough examples, there is enough depth in terms of what can be accomplished here. So this is something that is actually right out there. So that is dimension number one that we need to look into. Second is we also need to clearly sequence in terms of what we should prioritize for the early adopters, what should be those success stories, what should be the sequence in which we need to build success, what initiative success can feed into the next initiative success? Now this is something that we need to definitely bake inside in the overall plan that we have while deploying agentic AI. Also, the automation use cases to make sure that we are able to leverage the native core capabilities of agentic AI to its fullest extent, which means that we need to redesign the experience, the process, and the platform's ability to drive. All we'll have to come together in doing this.
Some of the tenets that actually help drive is one is obviously, omnichannel. This is something that we have spoken about multimodal to take the information, as well as to omnichannel to disseminate the information. So that is the first criteria. The second is now we can put customer first as a principle because there are several resolutions the agentic AI will have to autonomously take. Therefore, we need to have a certain set of principles codified there. One of that is customer first. Therefore, it's able to resolve those issues very quickly. So I think in practical terms, I'll tell you what it is. For example, if there's a bank trying to employ agentic AI, what we have seen are two metrics. One is, in the US particularly, a call for a bank typically costs under $35.
Now what we've seen is when you begin to do the agentic AI type of solution there, we've been able to bring it to something like $4.
So this is a real number I'm talking about, $35 to $4. So this is one data. The second data I want to talk about is, for example, the credit card, the ability for reconciliation, the compliant dispute resolution mechanisms. What we are seeing is in the earlier systems that existed prior to this GenAIs and the agentic AIs, what we are seeing is a bank will typically set a limit saying that if somebody were to call and say that, "I have a dispute," the discretionary power of the manual agent who was operating that used to be something depending on the bank, it could be $150 or an up to $180. Those will be the limits. Now we are seeing through GenAI and agentic AIs, the confidence level inside the banks have gone up. Up to $500 to $600, decisions can be taken in favor of the customer.
See, this is a big change. Look at the experience that can actually carry forward to your own customers. If you repeat that across multiple stakeholders, that's the type of change that we're going to see. So there's an order of magnitude improvement. One is the cost is coming down by probably a factor of five or six, and the ability, the confidence in the system is also going up in a substantively. And that's a big change, which is what we call as both innovation and business value driven. The moment you begin to measure and optimize, you will see the type of outsized benefits that we begin to get out of deploying these types of solutions. And in doing this, I want to talk about what are the different stakeholders through which we can really tie it all together, right? One is we see that, obviously, we can do for your customers. You can do for your partners also. What we have seen is the ability to respond, for example, what is the test of a true partner? There are multiple ways. I can take this as an example. The test of a true partner is when the partner can actually really stand in for you and provide the same level of experience and expertise and comfort that you do on your own, right? So what is preventing that? One is two different organizations. But more importantly, the process, the ability to take control of a set of activities which you do was versus what the partner can do, and the ability for the partner to directly dive into your set of process for any type of resolution. So these become in a very, very difficult to accomplish now in real world, partly because of the technological barriers that you have. And agentic AI will break it. I'll give an example here. Some of the heavy duty Midwest engineering companies, when they go and sell some capital in terms of big machineries, one of the promise that they make is within 36 hours anywhere in the world a downtime will be resolved in favor of the customer. So because we are talking about a paraphernalia that runs for millions of dollars, therefore, this is the type of commitment some of the Midwest companies, engineering companies typically provide to customers. Now all of these companies do not have the presence at every nook and corner. Therefore, they rely on partners. Several times what happens is in order to fulfill this criteria, if you invoke a partner to go and service that, a partner may go and service there, but what the company is getting in return is one-- Yes, the criteria to get it done in 36 hours is fulfilled. That is non-negotiable, but the cost associated with it dramatically goes up. So it's partly because the partner is not able to leverage the system that the main principal company had. And that communication gap, that information gap, that data gap is something that agentic AI can plug in very fast, which means that to provide a seamless, borderless across the channel experience in a consistent, repeatable, scalable, reliable way is becoming a clear reality with agentic AI. So that is the second dimension that we see.
The reason why I'm saying that is even those larger problems where physical activities are involved, if these can be solved through agentic AI, in the case of digital marketing, where you typically deal with all of the content information, communication, collaboration, the opportunities are bound. And that is the point that we're trying to make here by saying that from a customer journey creation to ability to understand the customer insights and act on those insights and be able to provide a much superior service at a much lower cost and therefore, increase the overall experience of the customer. And this becomes a distinct reality in the agentic AI world. An example of that we will actually see in a few minutes from now. The third is we can create multiple models, for example-- Because of the sophistication and the productivity increases, these technologies can bring to the fore. You'll be able to deploy these solutions much faster and keep experimenting, which means that you are afresh that you have in terms of what you drive through digital marketing itself can happen at a much faster pace. And that becomes very important because as this is a world where we look at what the neighbor is doing. We look at what the enterprise across the road is doing. And that movement sets in and provides a velocity that's going to be very useful for enterprises. So I think therefore, I'll also do it in a consistent way, which means there are standards enforced. There's a way that we scale it up. There's a way that reliability is measured there. There's a way it works with synthetic data so that the scale up and the rollout can happen in a much more predictable way, which becomes very important for an enterprise in terms of how they plan their activities, what they promise to their stakeholders. Having said that, this is absolutely my last slide before I head over to Shilpa to talk in specific terms how organizations are embracing.
We see that it becomes very important to map the customer journey here. Sometimes the first principle, the cardinal principles through which organizations interact with the stakeholders, in this case, the customers, should not be understated. Just because there is a power of agentic AI, there's a brute force through which we'll be able to enumerate every different path. That may not mean that the classical approaches that we take towards on a customer journey need to be in a short circuited, but in fact, this can be reinforced. That's a clear point here. So the first thing is in terms of the awareness, if you look at in the customer journey, be able to create the right type of personalized interaction, personalized communication with the customers, our ability to use interactive assistance to provide the wider palette of catalogs that they can browse. Our ability to create hyper-customized emails, or communications, or messages to customers are creating the next best set of offers for the customer so that it becomes, for example, I do know global corporations today. When they want to do, let's say, in this case, let's say, a quick QSR, a quick speed restaurant. If they have a mobile application, we see that even today there's a latency of something like an hour. Supposing they want to do real-time offers in the way that these enterprises and architecture is set up. If they want to say, let's say, in Atlanta, there is a pub, at this hour I want to essentially create some offers so that more customers can come into the store. Today, many of the QSRs don't have the ability to do there is a lag of an hour. Even with all of this technology, you are able to operate out data at the centralized level with a lag of an hour. That's the type of architectural limitation that come in. Therefore, all of this can be broken through agentic AIs. So with that much ado, let me invite my colleague, Shilpa, to talk about specific terms in terms of how we have done it and throw this floor open for questions for a greater length of time. Shilpa? So we just saw here how marketeers' journeys are essentially changing. And when we look at it, GenAI changed the marketeer journey in content very quickly. But what agentic AI is doing, it's changing the entire journey. It allows us to do agents for each step, as well as doing this entire journey automatically or automating it. Going forward, we see that from a content or a customer experience, consumer experience perspective, it's not just content, but there's a lot to be done in campaigns or journey orchestration, there's a lot to be done in personalization, especially in experimentation and segmentation stage. So what GenAI allowed us to do is just do variations. But when we bring in agentic AI in experimentation, we can actually run automatic experiments, put in the results of those experiments on web pages quickly as within a day or two or three without any marketeer interference. So it actually accelerates every step of the process that is there. Not just generating content for that experimentation but actually running experiments and then launching them when they are successful. In analytics, there are multiple, multiple applications. We can identify KPIs that we should measure through agents. We can also apply those metrics on the pages through agents. We can then without dependence on reports to be generated and visualized, on runtime, these agents can give us the information that we need to the business users, not to the analysts, not to the marketeers, but to business users and executives in the language that they understand. So if an executive is looking at it, an agent can do a report in revenue terms. Whereas, if a marketeer is looking at it, they can do the report in impressions term. Whether if analysts are looking at it, they can do the report in KPI terms that they understand which first initially took a person actually preparing these three reports, doing it on a daily, weekly, or monthly basis. But now anytime anyone can log in and get it in the terms that they need. The real-time visibility can be achieved. Same use cases can be applied in commerce, in services, customer support. The multiple applications, I think agents are now there-- We've been trying to connect customer support to CX for a very long time, and we are trying to connect customer data platforms. That's the use cases where it came from. When an agent logs in, they should be able to see where the customer is. But agents allow us to connect that now very quickly. There are lesser need for integration because agents can automate that process and bring in the information to whoever needs it whenever they need it.
Now looking at how to enable this architecturally, a lot of products, what Adobe is doing now is they are bringing agents within their platforms. But all of us understand that organizations are not just one product. We have so many products out there. We have a data infrastructure there. And we have then multiple MarTech products distributed. Along with that, we also have to make decisions on which AI model suit best where. So we have to decide for every agent that we're using at base which AI models will they use? Will they use multiple models? How will they connect with one another? And then build an agent ecosystem on the top. So you will have an agent, for example, for audience management. There will be which can be an Adobe internal audience management agent. And then you might have a custom agent built to move that data or the audiences to the customer service side. So it will be multiple, multiple combinations of agents that will need to be built enabled through this data marketing, a marketing ecosystem that delivers the end goal, which is as much automation in this entire ecosystem as possible.
Now whenever we implement something, the question comes how do we measure the effectiveness of it? And as we started, we saw that there are three core user groups for us. The first is the consumers we are running these programs for, then the marketeers, and then the developers or the users of the platform. So the first measurement for the consumer or the end customers are engagement KPIs. So again, continue to measure those engagement KPIs because that's how we know whether they are getting personalized experiences, are there abandonments happening, how long are they staying with us, how much time they're spending with us, how much satisfaction do they have? But for marketeers, the KPIs are different. We move to more performance-based KPIs. What is our time to launch? If we took a week to launch a campaign, or can we do it within a day? Can we do it lesser? What is our overall marketing ROI? Did we invest in the agent or experimentation agent? How quickly and how many experiments did we run? What was the ROI on it? Did it actually impact our revenues? And then, very important, content velocity, all of this personalization need all of this content. So how quickly can we generate that content? How quickly can we get it out? How many pieces of content were generated within a day, within a minute, or any time? Then the third is the same KPIs cannot be applied to our developers and the platform users. So they have operational KPIs. We need to see how did the application after putting in the agent performed. How many bugs did we get? How many tickets did we get? How many issues did we get? How much time or effort did it take to solve them? So application performance becomes important. Code quality becomes important because we need to trace it. We can't just keep putting out code there over code, over code, over code without traceability being there. And then test velocity. How quickly are we testing the code that was put in? Even if it's agents, how are we getting any efficiency out of there? So we have to understand that even if it's agent, we have to have different KPIs for different type of agents that we are deploying. We can't just have engagement KPIs and measure them and see, "If we are doing more impressions, we are running more campaigns." That's good enough. We have to look at all other KPIs as well and go down at the user level to look at them.
Now just to understand-- I'll just go back here. So essentially, these are just some examples where we are already doing AI. And if you see, there is a big spread here in terms of both B2C organizations going into agentic AI and already adapting these use cases. We are also seeing that this is across verticals. So these agents are not limited to just B2C retail use cases or simple use cases. Complex use cases within organizations like life sciences, healthcare, because compliance is a very big place where agents are playing a part because they want to automate those processes. Then financial services, where there are a lot of regulations, they are looking at bringing in agents to have their customer service, to onboard customers quickly because the processes are so repetitive. So all areas that we have here, we are actually seeing adoption of agents coming in. And these are some other use cases where we've implemented AI to essentially starting from generation of content, selecting of segments, marketing to those segments, measuring performance. And we've seen up to 40% efficiency gains in terms of bringing in more revenue. We've also seen amazing improvement in time to modernization and about 80% reduction in content generation efforts. So if we do it right, we identify the use cases correctly. If we see where we can bring in and connect these use cases together, there's multiple, multiple benefits for all involved. Yep. Right. Thank you, everyone, for attending the session. And if you want to scan-- [Music]
