Sponsored by: Accenture & Avanade | Enterprise Scaling and Value of Generative AI and Agentic AI
Summary
- Agentic AI definition: Agentic systems enable software programs to perceive, reason, execute, and adapt—moving beyond generative AI to autonomous, goal-directed decision-making.
- Enterprise deployment strategies: Implement practical approaches for deploying, managing, and optimizing agentic AI frameworks across large organizations, measuring ROI and driving adoption.
- Real-world case studies: Learn from enterprises successfully implementing agentic AI at scale, showcasing measurable business outcomes and competitive advantages in the AI-first era.
Sponsored by: Accenture & Avanade | Enterprise Scaling and Value of Generative AI and Agentic AI

Talk by Venkatesh Rao, Americas Data & AI Lead, Accenture (HQ) (Sponsor Speaker)
In this talk, we will explore the transformative potential of Generative AI and Agentic AI in driving enterprise-scale innovation and delivering substantial business value. As organizations increasingly recognize the power of AI to move beyond automation towards true augmentation and intelligent decision-making, understanding the nuances of scaling these advanced AI paradigms becomes critical. We will delve into practical strategies for deploying, managing, and optimizing Agentic AI frameworks showcasing how autonomous, goal-directed AI systems can unlock new efficiencies, enhance customer experiences, and foster continuous innovation. Through real-world case studies and actionable insights, attendees will gain a comprehensive understanding of the key considerations to architect, implement, and measure the ROI of large-scale Generative and Agentic AI initiatives, positioning their enterprises for sustained growth and competitive advantage in the AI-first era.
Databricks Named a Leader in the 2025 Gartner® Magic Quadrant™ for Data Science and Machine Learning Platforms: https://www.databricks.com/blog/databricks-named-leader-2025-gartner-magic-quadrant-data-science-and-machine-learning
Build and deploy quality AI agent systems: https://www.databricks.com/product/artificial-intelligence
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements
FAQs
What is agentic AI and how does it differ from generative AI?
Agentic AI goes beyond generative capabilities by enabling software to perceive, reason, execute, and adapt—creating autonomous, goal-directed systems that make independent decisions.
How should enterprises approach agentic AI implementation?
The video covers practical strategies for deploying agentic frameworks at scale, including management approaches, optimization techniques, and ROI measurement frameworks.
Which customer data platforms support the agentic AI era?
Customer data platforms built on the Databricks Data and AI platform provide the unified data foundation required for agentic AI systems. Agents need real-time access to accurate customer data, and architectures that combine data governance with agent execution enable safe, compliant autonomous decision-making.
What is an AI-native customer data platform?
An AI-native CDP is designed from the ground up to support AI agents and machine learning, not just analytics. It provides low-latency real-time feature access, complete data governance, and seamless integration with agent frameworks on the Databricks Data and AI platform.
How can organizations prepare their lakehouse for agentic AI?
Ensure your lakehouse on the Databricks Data and AI platform provides real-time data access, complete governance through Unity Catalog, and integration with agentic frameworks. Build agent-specific features in low-latency formats, implement comprehensive audit trails, and establish clear policies for what agents can and cannot do with customer data.
Full transcript
Right. So, good afternoon everyone. Um, Weni Rao from Accenture. I'm a managing director in the company. I'm responsible for running our what we call as the data and AI practice. U which seems to be the more popular area these days. I've been with Accenture for 19 years. I've done various things and over the last three years I have been running our integrated data and AI practice. So what we will do over the next 40 minutes or so are you know we'll split this into two parts.
Uh 15 minutes you'll have to tolerate me rambling all by myself and then I'll be joined by four esteemed colleagues here from the industry and we'll start making sense of it. I mean the theme of the discussion today is largely about look we we have all been hearing about AI then geni you know agentic AI what does all this mean in the context of enterprises and you know is there is there some touch and feelness to this thing around enterprises and how things are scaling and what's the reality and you know where is where is the pendulum swinging right so that's that's what we want to do um look I mean I I have a few slides I I'll I'll go across that but but before we get into some some content what I want us to reflect upon is the technology has come a significant way right and these kinds of conferences are where you have this conversation so there is no ifs and buts around you know is the technology there is it real is it stable is is can it deliver to the promise so we are going in with the underlying premise that companies like data bricks and and the peer set will be able to deliver the technology at scale right so if that is true then what essentially is happening in the world of enterprises right so that's that's what we'll dig into so let's jump right in I don't know I mean that thing is active listening I don't know what it's doing so it's got nothing to do with me all right so um I mean essentially where we start is I mean what's what's agentic right I mean for me to simplify the conversation, it's it's going into four realms, right?
So we we have the ability for a software program to perceive, to reason, to execute, and to adapt. The perception started like, you know, when when generative AI came about, we should we could clearly see the ability of an algorithm to perceive and reason. You know, it then started to evolve and it started to reason. You you could query, it came back, you query again, it came back with something else. So that ability started.
But then for enterprises especially for those of us who deal with large corporations for us what's really interesting is when the algorithm can actually execute something on my behalf so that I can I can go do something else and then adapt uh along that with that execution is the is the adaptability capability right so that's where the needle starts to move for us within the four walls of an enterprise right so that's that those four things are are what we saw evolving from an accenture perspective where we see that okay the needle is now uh you know slowly moving definitely certainly moving from PC's to scale uh examples right um so again if I if I jump then like how do we how do we see this shift happening right so there is there is a whole lot of AI ambition out there now we we try to segment this again as a as a good consultant right I mean we are trying to bring some pattern to be able to understand this this whole motion like you know the whole motion of enterprise adoption of AI and how does this happen so if you look at the left side we call that realm as AI infused right so there is a whole lot of conversations there's a whole lot of activity where you can infuse AI into the transactions that you're already doing then on the right side you have the spectrum of AI native right so in the AI infused I mean it's no prices for guessing right so if you have deep domain knowledge I mean if you have brownfield processes etc. you you infuse AI into this.
I mean you are not yet going into a fully autonomous agentic system to replace whatever it is that you're you're having right now. Right? You infuse AI it will still be human assisted. You will still have enough guard rails and you will you will you know you will leverage AI to either get productivity or sharpen whatever it is that you're doing. But as you move to more AI native to the right side, you are starting to talk about you know creative content generation, R&D, PLM where you know that under certain guard rails you can actually let the algorithms go native like you know it can run autonomously on its own.
So as you start looking at you know where do you put the AI initiatives it actually helps to have some kind of a rubric where you can start thinking about in what kind of activities do I infuse and in what kind of activities do I go and implement a fully native solution. Again I remind you all right I mean the underlying premise is that you know we are taking technology for granted. We are we are assuming that big platforms like the one the reason why we are here for they will continue to deliver the technology and that capability of the technology will continue to evolve right so we are not questioning that what we are saying is let's bring some method into how do you segment what kind of activity will you scale what happens then once you start bringing AI into any kind of activation then you start reimagining work you start to reimagine workforce and you start to redesign the workbench and what we mean by that is you know essentially the first one is the easy one right so you you start to look at I mean it's the oldest adage right I mean if if you put garbage in there is garbage out so there is no prices for taking a broken process and putting any amount of technology into that so you have to start re reimagining work in the era of AI it's as much true for AI infused as it is for AI native right so unless you are able to fundamentally rebaseline and reimagined work you the impact of AI will not be there right I mean it'll be it'll be interesting it'll be entertaining but it won't deliver the value uh that you're going after the second big one is reshaping the workforce now here again right I mean there's there's and we have all spoken about this for more than three years now is the whole impact of AI around workforce there are multiple dimensions to This conversation one is clearly like you know how do you skill and upskill your workforce to leverage what uh the conversation we are having around AI but more importantly you will need to infuse a lot of skills in your organization to be able to manage these complex systems of intelligence that you would have deployed either on the infused end or on the native end right and that workforce many many enterprises that we consult certainly are not get ready with that workforce transformation and it's a journey.
It won't happen overnight. It's not a skill that you'll just go and recruit on LinkedIn, right? So it will take some time and effort and concerted effort to actually get the workforce to pivot in that journey and redesign the workbench like you know how do you provide the tooling to enterprises right I mean it was um there were and in this journey over the last three years there were co-pilot moments right where organizations declared victory in their AI journey by saying that look 60% of my organization has access to some form of co-pilot and I'm using co-pilot more as a as an adjective here not a specific uh a product uh but but that category of AI assists right everybody has something on their uh workstation and then that alone doesn't guarantee that you know you have made the journey you have to have the right kind of strategy around tooling you have to have the right kind of strategy around enablement and that's what we mean by redesigning the workbench right so now um as you look at this journey again you know from a traditional enterprise if you are getting into that agentic AI kind of mode, you know, you are starting to get from a more deterministic kind of u managing and governing an enterprise to a more variableized like you know self-arning kind of an organization right so but today enterprises are built on very strict rigid very deterministic rules right so it is pre-wired processes I mean yes we have parallelizing uh processes you have a lot of activities but it's still within the four walls of a structure and from there we are getting into a a system wherein there are humans acting on a set of rules then there are agents the AI agents acting on a certain set of rules then there is the feedback loop which is retraining your agents not just your agents but also the humans who have to intervene within the four walls of the process so there is going to be fluidity as you start going from a very dis deterministic construct into a agentic construct and you have to start thinking about this journey with a lot of intention around how you will manage through this change because this is not just about technology.
At the end of the day, organizations are there to do what that organization does, right? So if you are a oil and gas company or if you are a bank, you are there to do oil and gas. You are there to do banking. You are not there to invent AI for the rest of the world, right? So you you still have to make sure that you will manage the organization with that fluidity. So again um this is going into a bit I don't know why the slides are looking like that.
Yeah. So um if you look at the agentic ecosystems I mean what we are essentially seeing again this is a a distillation of a lot of learning that we have had implementing a few thousand projects now you have a hierarchy of agents right I mean there's a lot of talk about can I download this agent for that activity can I do this like you know I have so many agents I have this I have that agents alone are nothing but like a set of code right I mean in in in vacuum it's interesting but it doesn't solve anything on its own.
What we are really seeing for solving for complex problems are a complete agentic ecosystem. If you look at the right side, so we are talking about the full hierarchy of utility agents, microility agents, super agents and orchestrator agents. And like the adjectives on on those phrases are those agents are supposed to some of them are supposed to orchestrate, some of them are supposed to, you know, make decisions based on the activity that the utility agent goes and does.
And those decisions have to be managed um and governed with a full stack of responsible AI agents and that's really really important right I mean if you have isolated cases of you know different departments and different functions within your enterprise doing various random things right I mean I got an agent from this particular software or that particular website and I download it yes it does some activity for you but is that still governed under the responsible AI framework Is that autonomously governed around the responsible AI framework?
What about that emotional state like you know how does it uh manifest in terms of the brand promise that your brand stands for? Does the you know within the four walls of the enterprise are those conversations within the four walls within the framework of what's allowed in your organization when it talks to an external organization does it you know convey the promise of the brand that you are supposed to represent right I mean those are very very important things to be had that's where we say that look I mean agent alone is not the be all and end all don't get too carried away with the hype I see lot of PE money going into uh these startups which have you know this agent for this or that agent for that you have to think about it's not where the agent is coming from it's how you tie it all together into you know a complex machinery that works within the constraints and the four walls of what your organization is supposed to stand for yeah again um coming back to this right so essentially all of AI is fine but at the end of the day your organization does what your organization is supposed to do right so if you're a manu manufacturing company.
This is the value chain of the manufacturing organization, right? So you you buy, you make, you store, you move, you sell, right? That's what you want to do better with AI. It's not your your job is not to create the next generation of language models or or something else, right? Your your primary purpose is to get this value equation, you know, uh enabled using AI. That's that's the that's the superpower that's available to all of us. now and what we are seeing actively happening in the marketplace and this is where accenture is actually putting its bet saying look AI is coming to mainstream and you know it's really really scaling for us because when we are having conversations like this and when we go down and say if you take a mega process you know call it a a procurement process and an order to cache process or a R&D process then you have conversations at the next level, the component level, but then you go all the way to the activity level and that's where you start designing your utility agents.
Right? So when you start implementing agentic systems which go all the way from level zero to level six, then you are actually taking a part of the organization and enabling it with AI. Right? So that's and given that we are having multiple conversations with enterprises across the world at scale some of these conversations that's what brings us back to the thesis that enterprise scale AI is real it's there it's for all of us to leverage we are we are fully committed to the fact that technology will be available yes it may it may have some gaps today but it will over a period of time a short period of time you know that will that will evolve and remediate on its own.
But the fact that we are having these deep conversations around enterprise tri process transformation you know cross enterprise transformation that's what gives us the confidence that this thing is actually real and it's it's moving and it's scaling that said let me invite uh my guests here to the panel here I mean from Trronox again let me sudipto from AON taken from Shell and Kiran from Satah have please. Yeah. Anywhere.
Anywhere. Oops. Your camera. Why don't we do this? Uh Rose, why don't we start with an introduction and as you introduce yourself, if you can also tell us like in your organization, what have you seen scaling? um like a serious one with AI. Yes. So yes, sure. Hi. Uh thanks for having me having me here. This is Rose May. Uh I'm the VP of enterprise data and AI as well as business planning for Tronox. Ronox is a vertical integrated u mining and a chemical company have operations across six continents.
Um in the company um certainly I'm leading the strategy of how to deploy data AI drive value. Um on the other hand uh one of the major responsibility I have is also in terms of uh business planning where to invest um and then with that um responsible for the enterprise optimization modeling as well. which is also called the brain of the company. Um so in terms of the scale because we just started uh last year actually toward the end of the year um to pilot a AI use case in commercial and then was able to demonstrate uh the rapid prototype in within two months and with that certainly build the confidence of uh of the techn technology.
Um and then with from there we start to build data platform this year and then now this month actually our commercial use case will move to production. Um certainly in terms of the scale uh we will plan to reuse those AI agents built to the other functions uh and then we're bu as we building the data foundation for example the finance area and to end planning and the operation will be the first use case this year as well. those are uh certainly as part of the business planning function uh is um familiar with the value where the value is in the company.
So those are the big three categories that we see can scale but reuse the data as we build and the reuse the AI agents as we build. Uh so that's how you scale it and then in operation of course uh if we build the first prototype successfully then the quickly replicate to global assets that's that will be how we will go. Um so just uh with that I give yeah sure thank you. So as an introduction Tees I am the chief data and AI officers in in Shell.
I don't think I need to explain what Shell does. um oil and oil and gas but we call it energy energy company. So my responsibility is couple of uh areas. One is of course data data foundation to make sure that we have a proper data foundation to help to scale the AI and another one is basically data products specifically utilizing the data bricks in this space and of course the AI and AI innovation is my main uh the focus area I must say we call it under the innovation just in 2024 just for you to give you the scale that we have over 125 PC's done in 2024.
So it's huge. Of course, I would like to say that most of them didn't give a big values. So we have to kill but some of them specifically made big impact in our business and specifically what we are looking for if the use case gives us gives or makes shell faster safer and safe is you know being the safe is quite important for Shell and also if it will give us to you know to achieve our vision towards the low carbon you know division.
Once we have done those things, so we have found a couple of big use cases, big ticket items like a couple of them I will give one is we we call them as a yeah we we call them like a we give them a personas like AI personas one is the trader agent who is basically a digital co-orker for us and who is basically getting the real-time market data then make or execute to trades or even sometimes s you know uh put um um to give some information to the traders to change uh to hedge these execution the these trades basically and of course this has been done via um agents but within the certain risk parameters created by the human.
So human and digital coworker works together. Another example could be around the uh the marketing area. The marketing uh the campaign optimizer we call it the marketing agent. Again we call them all agents like from the matrix. Um but basically what it does is basically get the real time data uh from the the users the behaviors and immediately give backs the campaign or give the offers during the during the selling uh in our retails.
So imagine couple of years back we were doing this already but we were doing this for after a couple of weeks time to you know reach out to our the customers. Now we are doing during this the selling item. So another one I will not go dive into it but customer operation is the main one to increase the operational space. uh and there's one more thing that I would like to add. This is a probably the biggest uh one that we have been doing for years with the traditional AI and ML around subsurface piece but now we are bringing it towards with the agent.
So basically like subsurface data think about the cysmic data that we you know get from the entire earth to find the best place to drill and this is a super super important you know the item for a for a company like Shell and what we did is basically get this data but give this to the agentic AI to find out the best the places like the like the the we call it is the um how do you call it um uh radi radiologist radiologist agents.
So basically it does the all the work for us and then of course there's a human in between to define if it is the right place and then we do the dling and this is a billion dollar value generation uh item over the five to 10 years with that I will oh yeah you already have it hi I'm Sudto I'm uh globally responsible for data and AI platforms at AON for those of you who do not know AON we are a fortune 200 company second largest insurance broker.
Uh we are in the business of managing the risk and volatility for our clients. So if I'm a enterprise, there are two types of risks that I encounter. One is my business risks and second is my human capital risk. So we are in the business of managing the risk for our clients. So uh we are a regulated industry. So being a regulated industry uh when it comes to generative AI uh there are of course there are certain constraints that we have to be mindful about but uh barring that we also look at it as a opportunity for us to uh actually accelerate our innovations in because jai gives us that opportunity.
So there are two areas that we have seen success uh in our uh acceleration of uh LLM and generative AI and all the new forms of AI that we are seeing. one u adaption when it comes to the chat bots and the uh search engines. Uh when it comes to a regulated industry, we cannot like we have to be very extremely sensitive as to what data the chatbot is uh accessing. So we have to provide it in the context of or in the ring fence of the data that we are allowed to allow our users to have access to.
But given those constraints, we have seen that capabilities like enterprise search, capabilities like chat bots, contextual chat bots has been extremely uh valuable for our users be business users, beta technical users and we have seen significant adoption. So we did launch our enterprisewide uh chat which we call it as GPT user widely across the organization. So that's one area that we have seen uh adaption uh and I would say a lot lot more pull but the expectation from the users out there is low because we are not complying to the fidelity of the response which the user needs to get.
The other area that we have seen significant value in Zen AI that we are uh where where it is embedded inside our business cases maybe very similar to AI infusion where let's say if I look at insurance one of the typical process of insurance is a proof of insurance everybody needs a proof of insurance now as a user be my be an uh consumer of insurance uh be a producer of insurance or be the broker I don't need to know how this uh certificate of insurance gets generated.
But behind the scenes, if I'm able to use uh jai or llm to accelerate creation of the certificate, processing of the certificate, I'm increasing the time to value. Now whether uh whether I'm leveraging ji or llm, nobody cares. But it's a value that I see leveraging those technology because it helps me to accelerate the time to value. Similarly, we see uh in in an insurance there are multiple other process steps. Let's say uh uh one of the steps is the brokering process where you are brokering between a client and carrier.
Now, as we look into the brokering process, we have to process multiples right if we are able to automatically process documents uh with LLM of course with human in the loop that again accelerates time to value. So, so the use casesdriven conversation around GNI we see a lot of adaption with our clients with our business uh as well as even the technology organization because it accelerates the time to value. So that's how we see our journey.
We are still uh we are mindful about the about certain governance aspects around hallucination around uh the uh the explanability of the gener but even with those constraints in mind wherever it's a use casedriven conversation we have seen a significant benefit with us as a as the data organization or with business or even the clients. Good. Hey everyone. Um Msore. I'm the chief data analytics officer at Sutter Health.
Um how many in the audience actually are in from healthcare or just okay some any of you are local? Okay. So hopefully you'll know Sutter Health. Uh we are actually one of the top 10 top 10 uh healthcare systems in the country and uh we're headquartered here in Sacramento uh not too far from here and uh we operate about 24 hospitals um hundreds of surgery centers, ambulatory facilities, uh labs and uh employ um or contract with about 12,000 physicians across the system.
And um we actually serve a very unique geography because in about a 100 mile radius of where we are, we actually have some of the most um expensive zip codes in the country and also the least um from a uh access standpoint uh you know financially challenged zip codes in the country. And so that speaks to diversity of patients. And when you talk about scale, I mean that is a very key aspect that we think about. So my role is to enable the best information, data, AI to our ecosystem and my stakeholders are obviously the three and a half million patients that we serve uh because we build tools for them.
Um it's the 12,000 physicians that we serve. So we actually enable them with AI. Um and then my team uh which obviously is on a journey of learning AI. I mean again I like to boast that my team uh about 160 um folks have a combined 1,000 years of sudden knowledge amongst all of them. And so if you think about the institutional knowledge that is resident in the team is amazing. And this is actually something I'll talk about later too is around workforce and workforce development right subject matter expertise is really important. uh the aspects of AI and everything can be taught over time and I I know it's not an easy curve but it is a curve but knowing how to apply it in the context of our world is really really important.
So in terms of scaling I'll just quickly summarize you know one of the things we've recently done is rolling out this uh capability called ambient documentation. So you go to the doctor's office uh you sit down and you start to tell the doctor everything that's going on with you. Well, most of the time the doctor doesn't even look at you, right? They're typing notes with their back to you on a computer. And this is a challenge because the fundamental practice of healthcare requires empathy and requires a patient to physician connection.
And this was this has been lost over a long period of time ever since the electronic health record came out. But more recently, uh tools like ambient listening where you have a device that's actually listening to the conversation. Again, you're consenting to this, right? So, there's an important element of consent. Uh but there's also security. Um but you're able to then capture in real time what's actually happening.
And then how many times have you gone back home and realized oh my gosh I forgot to tell the doctor something or you got this really complex after visit summary that you can't make head or tail of. We're actually at scale able to translate that conversation into easily readable uh text and also electronic health record that we can stand by. So couple of examples like that but we'll talk more as we go. Sha thanks a lot.
Thanks for those introductions. Uh Rose, let me start with you with the first question. Uh I I'll pass the mic or maybe you take that. Yeah. Um look, I mean you you spoke about, you know, multi-prong transformation journey in your organization and the question always comes to business value, right? So what are the KPIs and metrics that you track track? Let's start there then we will get into the AI part. But how would you guide this audience around how the business value measurement and tracking happens?
Yes. So earlier I mentioned we started uh commercial AI use case as a first um and then uh we're moving into production this month. So very excited to look for that. Um I would think the value from three perspective. One is it really augment um our sales marketing people and uh um from executive support of course and then leadership support. Um so what the solution we put in with aenic AI architect and then all the or uh like you introduced earlier how we have corporate AI agent and then pass tasks to the different AI agents to perform tasks.
Um so with that um one thing they augmented for example we will be able to globally generate sales report uh on the fly and pulling information from all the different sources and support our sales uh in terms of meeting the customers so they can spend more time understanding market dynamics um and then uh uh more tailored focus on customer care um to improve business topline. So with that um the KPI is coming twofolds um we consider both time and the money.
So one is augment the sales marketing um group that they can drive a more topline and then profit and then growth of the company. Uh that's coming from the money side but on the other hand the time is really important critical for this whole data uh AI development. So think about how much time um they can save from the mandate task in the past or the even just find the right data right so multiple people looking for different data or coming with different results.
So um so take all the things out of uh the people who need to make decision with precision and then with speed. Um the other part is um with this process we automate uh the data pipeline, data quality check uh and then a lot of analytics dashboards with that take out the time the employees need to spend to support sales marketing activities. Um so and with that also the other aspect of that is uh we wi with the solution we put in now it really simplified the IT or system environment uh investment uh so from cost reduction perspective but also timing because we are one background of Tonox is still in the middle of ERP consolidation so for our commercial team to get to the result they currently have.
Normally they will need to wait till the ERP consolidation complete. So we shift the timeline now two years earlier. So get to the global visibility and then uh support what they need to need to have in terms of the tool to make the right decisions to grow our business. Um that is from timing perspective is significant saving and bring the value up. Um so the last thing I would say ba based on because we're investing now data platform um built it's the data foundation because uh AI really feel from data um so with that the productivity gain for our small core team um is significant reduce the time and effort as well to get the value.
So, so overall in the data AI space I would consider both time and then uh money side is uh how you consider where the value is. Super. No, thanks thanks for that. I mean absolutely we need to we need to start from value and you gave a very empirical way in which how you measure it but there is no AI without data. So Sudepa if I come to you how do you think about data availability governance data residency data security all things related to data if you can talk about that.
Yeah uh it's a interesting topic uh being an insurance company governance is always the starting point for all our conversation. So we are highly regulated. So we have been uh highly governed when it comes to data uh in the past but with uh LLM what has changed is now unstructured data documents multimodal information like video uh uh images they're also becoming first class citizens like uh transactional data or structured information table data So which makes us to rethink the governance framework and this is where when it comes to cataloging when it comes to stewardship in the past we would always look at it more from a transactional data more from a master data lens now we have to start looking into or now now we have started looking into our entire data stewardship process our observability process our data lineage process as well as all the governance and the controls that we want to put in from a um at a catalog level from a from a multimodel a multimodel form which means people can start uh running an LM query and make sense out of an image which we could not have even imagined in the past.
So that adds additional uh technical complexities as well as regulatory complexities for data. Second consideration that we that comes um is around the uh the human in the loop dimension. So we know uh like if you if you look at even Tesla right Tesla autopilot has been around for quite a lot of long years but before any individual who is a who is driving riding a Tesla driving a Tesla gets comfortable with autopilot it takes time.
So maybe even if your LLM is 95% accurate and your ENO or your rate of errors is u 90% still you would be less comfortable in a uh machine. So you are more paranoid if a machine makes mistake than a human making mistake. So we are aware of it. So that's where all our use cases of LLMdriven document processing and insurance is all about data and document. So we process a multitude of documents from policies, quotations, uh exposures, uh claims, loss runs and of course LLM helps us to accelerate uh the work which previously an RPA bot would have been doing or even a human would have been reading and processing.
But this also brings an additional level of governance where you have human in the loop for all the different processes. Some in some situations we have seen that the human is not making too many corrections but it's the comfort that yes there is a human overseeing it both from our from our side as technologists people who are governing that yes human oversight as well as the dimension that if a machine makes mistakes how do we take action against that so that's the that's the second dimension that we are grappling with part of it is a technical dimension how do we automate it but most of it is the human comfort effort but we know eventually those humans will uh uh like that human effort on the human in the loop will reduce over time like Tesla autopilot.
The third part of it uh which uh which is also very uh important and relevant is previously when we look at like initiatives like oh I need to do master data governance and master data management I would be looking only to my transactional systems. Now we have to look into documents as well to do ma master data management and master data governance because people can uh read some of those information convert it into unstructured data and then linking information and tying together all the entities across needs uh uh needs the multimodal uh dimension and thoughts.
Same applies as we are creating like with any industry we operate in five like there is a there's a clear set of processes which we have for example we have an onboarding process we have a pre-binding process which is like a placement process and then we have invoicing claims and loss runs uh and then you have the cash collection process each of these processes entails a data product behind the scene traditionally those data products constituted only structured data transactional information.
Now each of those the rem of those data products have expanded with videos with takes which with blob objects. So so it's it's like uh different form factors of data is everywhere which we have to manage through all the traditional lenses of governance as we have seen. But if we are able to manage it as long as we are able to establish the trust in the data we can think of use cases which was not possible in the past. searchability on videos which we have in a in a single folder.
We could not have even imagined that with a simple few lines of code we can now do that given that we have the agentic uh uh like workflows in place. So it's so but your AI is going to be as accurate as the trust you have established with the data and it's a journey. So we don't expect that we it's a single bullet as the AI is evolving as the variety of the data is changing. We see this this journey evolving and we need to have a road map that eventually we'll find out what's the sweet spot.
So it's it's a it's a journey with experimentation but as long as we are able to be comfortable with the level of governance that we have at this point of time I think uh it opens up a lot of possibilities for businesses like us. Sure. No, thank thanks for that answer. It's a fairly detailed one. Um, taken uh Shell has been on an AI journey for a long long time. I mean, even before Gen AI era now, what are some of the challenges that you see as you look at the new agentic and genai projects?
Challenges in scaling them in your organization. Yeah, good question. So I would say the first one and the biggest one most probably the hype versus art of possibility. So being in the shell we are pretty much exposed to all the startups and all the vendors where they come and then make a very good the PC's or the demos and our business immediately jump on it say we should be doing this why are we late etc. And then we need to really I need to specifically calm them down and show them the art of possible today.
So focus on the value and focus on the what is possible today. I need to remind them the blockchains. I need to remind them to the metaverse and so on and so forth. Of course AI here to stay. AI is a big big transformative power. We believe in it but where to put our money is important one. So the reason that I told you like we have 120 over 120 PCs shows this actually. So there are so much demand but where is the value is not known yet.
So the another thing pretty much linked to this one is basically this innovation part of it the PC's. So uh we realized we have P cycle of like six months and for me is not the P anything beyond six weeks is not P. So there there is this tendency of say you know very much difficult to say no or to kill the projects when when when when they started. So I brought this specifically uh into our processes like beyond six weeks you need a big um the justification to continue if it is really value and then we are very easy now to say no and kill certain PC's and we have to do that otherwise you spend your money on different things and everything but everything has to be backed by business if it if business is not involved then we are basically not doing The last thing I think is pretty much what you also said on the data space.
There's this saying you won't you you don't raise to the level of your technology but you will you know uh drop the level of your data. Data is super super important. I I'm I'm coming from the very much data background like in 25 years I've been in the banking industry as well and it's always the one that is being neglected. So everything is everyone is basically focusing on the hype focusing on the this shiny thing that we need to follow but everyone is forgetting the data part and that's a big challenge.
So most of those PC's cannot make to do production for the reason that we don't have the proper data quality to make these agents work as we wish to be working. We see some uh the agents they are started uh but we couldn't scale because the underlying data like in the mobility space as an example wouldn't good enough uh to make it automatized. So those are my three uh the points. Super. No thanks thanks for that. So we spoke about the key metrics the value we spoke about data we spoke about scaling AI but in the end we know all of this works with the human right.
So talent. So Kiran, why don't you talk about you started there uh talent is the most important thing in the equation. Why don't you talk about talent in the era of AI? Yeah, talent is super critical, right? Especially when we're in the healthcare space. Um it's all about relationships. It's all about people. Um you know, one of the things I found right is we all use chat GPT and all of these tools right at home. We are consumers of these products. the moment we walk into the office, suddenly everything changes, right?
Uh our systems don't talk to each other. Our our data is really bad. Um and all of that and healthcare is actually sort of at the lower end of the totem pole in many cases in terms of maturity. And we have an incredible amount of work to do. Uh and it starts with talent. And so for me, the way I think about talent, right, is not necessarily trying to find the best engineering person out there or the best, you know, technologist out there.
I mean those are important and we need that but folks that can actually understand the health care space a lot more. I mean there are many gives and takes in the space of healthcare and we really have to think about um the way we operate in a very different way. So I'll give you a couple of examples right? So we talk about agent tech all the time. At the end of the day, you still have to tell the, you know, the agent has to still figure out in the context of the situation what to do.
And in many ways, um, folks that have never done that in the past where they haven't actually mentored, uh, a program to go and learn. I mean, at the end of the day, an LLM is still a baby. It's learning and you're teaching it and you're giving it context and you're giving it u, you know, the environmental variables in terms of how to operate. That's a skill actually and it's a skill that needs to be honed and developed.
So if I give a very uh you know smart baby to a programmer and say grow this and make this better and that's a skill that you have to develop and that takes a little bit of leadership it takes a little bit of vision takes empathy it also takes collaboration u so I stress to my team all the time you know I really want you to grow in these spaces and that's why sometimes the best um AI folks on my team are the ones that never grew up in the space of data or AI they actually grew up in healthcare because they understand the finer aspects of human uh the empathy um and the connection you know with the patient uh and the other thing I'll also say is you know healthcare is very risk averse right so we don't want to take big risks it's we're talking about patients lives but technology around us is asking us to take risks I mean we talked about 90% you know is is that good enough right and so teams are conflicted all the time about what is the right balance of risk that you have to take and I think as leaders we have to figure out you know the application of technology with the right level of risk is really the sweet spot and how do you implement that is going to be uh a key.
So we are doing that. Uh we are growing our folks all the way from internships to college uh recruiting. Um but at the same time we're bringing folks from the very high end of the scale in other industries because we know if you've come from outside of healthcare maybe you have a different perspective and yes we can slowly ramp you up on healthcare but um you'll bring a perspective that is very unique and different and that could be valuable.
So absolutely I mean it it it really takes a whole village to groom this AI baby into something useful. Uh so with that said folks like if there are any questions or comments for this panel we can take some now otherwise we will wrap in a few minutes.
Learn more about the Databricks Data and AI platform.
The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.