Latest / TeqTalk with Jas / JP Morgan's AI Delivers Billions While Most Enterprises Are Still Running Pilots

Transcript

JP Morgan's AI Delivers Billions While Most Enterprises Are Still Running Pilots · TeqTalk with Jas

• speaker-0: Thank you so much, Arna, for coming to the show. Really appreciate your time. ⁓ before we actually go ahead with the topic of the day, why don't you introduce yourself? speaker-1: Thank you. Thank you so much, Jess. So I'm glad to be here. And my name is Arnab. I live in New York, Queens in New York City. And ⁓ I'm currently the outgoing CTO at VNS Health and have around 30 years of experience working in financial services, healthcare, and pharma in different roles, mostly focusing on architecture and data. Also ⁓ currently pursuing a doctoral degree in ⁓ AI in the specialization system engine. speaker-0: Great, thank you. Arnama, we have reached an interesting point in the AI journey. Every enterprise now has access to the foundational models, and yet we are seeing dramatic difference in the outcomes of the organization. That tells me that the competitive advantage is no longer AI itself, it is the enterprise behind it, it is the applications behind it. So I have quick two examples to share. Years before Gen AI Boom, Zillow built AI-powered home buying business. It ultimately had to shut down after recording more than 500 million in inventory write downs and losses. And if you fast forward to today, we have ⁓ one of the examples I'm taking of JP Morgan that has ⁓ rolled out governed AI platform for more than 200,000 employees. And the CEO says that it's already delivering billions of dollars. In annual revenue. So different generations of AI, the same enterprise lesson. Leading AI programs at Tech Focus, one thing has very clear to me that model gets the headline, the news, but actually enterprise determines the real outcome, the real ROI. So let me start here with you. Why do you think AI is exposing enterprises' strengths and weaknesses? more than any technology wave has ever done before. Okay. speaker-1: So let's look at it from a from a perspective. That is, as we had seen, the major transformations that we had seen, for example, Y2K, that was a huge endeavor that many companies had undertaken. So of course the companies went back to the legacy codes and fixed a simple problem of a digit debt field into four points, right? So but that was the work that was done, which was the the objective. But what came out of as a byproduct was the understanding of the applications that the companies had. So we gave them a an an opportunity to get a full hand, arms around the the all the assets that the companies had at that time. I think the same can be talked about AI as well. We saw the same thing with cloud as well. If you look into another big transformation, that cloud it it eliminated a lot of manual tasks that had to be done. In order to rack and stack the servers, provision the servers, which had very little alignment with what a business like banking or insurance or healthcare would be doing. So what cloud gave us was infrastructure as code. So we were able to build out infrastructure very quickly, deterministically, with a high probability of success using repeatable patterns like infrastructure as code. Similarly, as we come to AI, we see two things. One is the the bright shiny objects. Of course, whenever we see a new innovation, it can be well described by Gardner's the the the the hype cycle where you have an innovation trigger which creates a peak of inflated expectations, then there is a trough of disillusionment, then there is a hope of enlightenment, and there is a plate of productivity. So we go through this this seasonality or those those those transient times before we get to ⁓ a deterministic, a prob ⁓ a sustainable value driven by the by the capability. I think with AI, to answer your question, we see that it's not only the intelligence layer, it's it's a long illnessed story because there are data, there is infrastructure, there are a lot of pieces that need to be there which companies could work with in manual steps before. But now that we are bringing machines in, we are bringing a totally new player into this into this in the ecosystem who need A different type of way of working in order to give the results. So agents don't get tired, there's a lot of productivity issue, benefits, there's a lot of benefits that will come with this, but at the same time, it takes away the human intelligence, which we should not underestimate. So it is giving us an opportunity to look into certain things that we were able to do without, certain certain rigorous ⁓ capabilities which human beings were able to improvise what they were able to do by themselves, now that model would not work, and we will need those things to be to be designed properly. The reasoning frameworks, the RAC, or the or the new protocols which are emerging, new architectures which are emerging, in order to have the machines work together with with humans in that case. If I can add one more thing to this, there is also, I think we need to understand AI from the speaker-0: Yeah. Yeah. speaker-1: And I'm clicking from your question, Q from your question, that is AI has been around for quite some time. So deep learning, machine learning had been around. So companies were able to predict the the the the customer behavior, the the market penetration, so all those things they were able to predict. What we get new with this is generative AI, which was a new concept, and then agentic AI. I think that that distinction needs to be understood very clearly and what we are using it for. speaker-0: Yeah. speaker-1: Are you using it only for repeating certain tasks? Are we using it for augmenting? Or are we looking at full autonomy of replacing certain activities with machines, with agents versus units? think that context has to be set. speaker-0: Do you think that this time is more like the audit time? Those organizations have built their data foundations well. They are really creating good outcomes. However, when the data foundation is not good, output is not real. So w what do you think about that? speaker-1: Let's take a simple analogy. That is, let's think of a business which say you're building cakes, you you form a bakery and you want to make cakes, right? So what do you need? You need an oven, you need recipe, you need ingredients, and you need some tools, utensils, knives, plates, so on so forth. And you also need a chef. So five things, right? You need those five things. Now, if you apply it in the context of AI, you have the The oven, which is the infrastructure, is the cloud infrastructure. Then the recipe is a large language model. The ingredients is the data, the different types of data. And then you need software engineering techniques, MLOps, in order to promote the code, in order to manage the code. And of course, you need governance, which is like that of a chef. So you can have a recipe, the recipe can be tweaked, the cloud ⁓ is sort of matured at this time. But if you look into the type of data which a company has, that is where the big differentiator comes in. There was an MIT report which came out last year in July. So Nanda stands for the network agents in decentralized architecture, in which they're taking an agentic view of the internet, right? How the agents would interact with them. So from the Nanda project as they're building it out, they did a re ⁓ a detailed analysis, a detailed report based on which they found out that only five percent companies are actually going to be able to make the PL level impact with EI initiatives. Others are stuck with pilots. But at the pilot they're using with synthetic data, they're using a very small, narrow use case, a narrow model, and you can easily show success. But when you're dealing with real life issues, we have data coming in from forty different sources. The definition of customer is different in different different data databases. The definition of product is different. So that's where the real life impurities kick in, the real life challenges coming in. It's a complex problem. And that is where the the initiatives are fading. I think rules companies which are really invested in their data, not only in data management or data storage, but also all the layers that go around it, like data governance, data quality, metadata management, and also the new features that machines require in terms of RAG. knowledge graph, ontology and so on so forth, they're being able to reap a windfall of benefits. speaker-0: True, I agree. And I remember twenty twenty-four, everybody was busy doing POCs. And there was so much talk about I am doing this, I'm doing that. Adoption was really high that point of time. Now it has gone down because the time has come when people are have started scaling. The POC time is kind of over for most of them. So now the adoption is declining a bit because the reality is here that how much you can really scale. speaker-1: Right. You have a heightened expectation, so which is all euphoria, all hype, and then that then you have go through that doors of reality where you sort of figure out that not everything that was promised can actually be delivered in the context of your business, right? Not in an ideal world ideal world, but in a real world, and that is where you go through a ⁓ trough of disillusionment and then real expectations set in. And I think that is key. One more thing I'd like to mention here, Tess, that In case of AI, it's not only one capability. In fact, we're talking about a set of capabilities. We're talking about ML Apps, which has sort of reached the the the plateau of productivity. We have generative AI, which is now at the rough of disillusionment, as you rightly said, that it is now setting into the mainstream world, but many other things which are on the peak of ⁓ of expectation, inflated expectation, AGI, for example, ambient AI, for example. So there are a lot of which are being being discovered, being innovated, and they have a long way to go. So it's a set of different capabilities and they're at different maturity levels or different time, six months to two years to reach the plateau of productivity. That makes it more complicated. speaker-0: And whenever there is any shiny object like Gen AI or before that cloud, rate of phasing that object is pretty high, and then slowly it gets down. So that is the kind of nature we have seen. And we have lived through the cloud transformation, and I keep seeing the same enterprise pattern. Organizations adopt the technology first and try to redesign the operating model later. For many of the organizations, I have seen that. During the cloud period, I've seen companies lifted and shifted applications to data center without actually simplifying the enterprise and having the data foundation underneath. With Gen AI as well, many organizations are doing the same thing: deploying co-pilots, chatbots before actually thinking that customer journey, escalation path, governance, human interaction, human role should continue to play an important part. AI optimized the interaction, but enterprises still Or the relationship. speaker-1: Let give a few examples to what we just said, practical examples that we have dealt with, right? So you you took the example of cloud. So before companies would engage into a cloud journey, it is extremely vital and important to keep an inventory of all their applications and put a lifecycle management in front of it. And like a buy sell hold, right? That is, there are some applications which are going to actively invest, certain things that we're going to tolerate, and certain things we are going to retire. we're gonna exit out of it, right? So if you know that you're going to exit out of it, there's no point putting that in cloud. So that kind of ⁓ an analysis, that kind of judgment, decision, should precede the cloud journey, right? So similarly, in this case as well, with the case of AI, the processes had been you have we in many cases we find multiple variants of a business process based on a product and the companies grow organically or inorganically. through acquisitions, there are multiple processes which can exist, multiple for multiple products. So first, in order to standardize and get benefit out of it, we need to standardize the process first. So this B that B D A T from a business starts with business layer, business process layer, then the data, then the application, then the technology. Usually we put the cost in this case and not go through the right level of order, I think. That is something which is extremely important because as you rightly said earlier, that's a bright shiny object and the boggy back end. Usually the bright shiny objects get the visibility, attention, and the the the boggy back end, if I can use that term, because of the high investment involved in this. Take the example of a restaurant, right? In a restaurant you usually make small changes in the seating area depending on season, summer, in spring, in winter. There's small changes very minute changes, those things happen. But a change in the kitchen is done once in twenty years because that's a high degree. So similarly, I think what we do is we go towards incremental investment for businesses and put the focus on the front office and the back office gets sort of sort of ⁓ overseen. There's an oversight on that. And ⁓ and that comes and bites you behind, bites you back because of the investment gap and the speaker-0: So what do you think or what could be the reason that companies, enterprises constantly adopt new technologies first, then they actually think of redesigning the operating model around it? speaker-1: What are the specifics? Let's go to ⁓ I I'll talk about two theories in this case, if it's that okay, okay, starting with the first principles. So Clayton Christensen, very well known, has a fantastic book called The Innovator's Dilemma. And he has given real life examples when some companies, really established companies, they were doing their job so well with their established products, with their established customer base, they actually miss that what is the new thing because the market gets changing all the time. So they cannot position themselves. With the new products and services as they emerge. And and their part of their profit is taken up by the new entrants, right? So for that, Clayton had ⁓ had a had a concept, disruptive innovation, and so there are there are different different dynamics that play over there. And we have seen many examples of that. We have seen Blockbuster and Netflix, we have seen BlackBerry, Nokia with with iPhones, with Apple. So what his viewpoint was that I think it's prudent for companies to start up as sort of have a spin-off. Which focuses on the pure play part of technology and not disrupt the mainstream business, the mainstream business model. And once they find significant ⁓ level of confidence with the new t hypothesis testing, the investment models, with operating models, that's the time when they integrate the new technology. I think that's a very prudent model, which we can see that it is it's a combination of innovation and adaptation. So that's what I was talking about, the second theory. which is Curtin's adaptive innovation theory, in which we have we human beings, we think in two ways. One is that is we try to innovate, but when we innovate, we have a narrow or a very limited amount of scope that we work with. But when we talk about existing systems in company, existing processes in the company, then adapting those systems overnight is not possible. That co that requires huge amount of investment, huge amount of funding. And the right analysis has to be made in terms of total cost of ownership. So I think the model that Curtain talks about is you have innovation on one side and you have adaptation on one side, and the right the business value is created when there is significant collaboration, significant value I would say cross collaboration between these two teams. And we can really look through the entire architecture, systems of engagement, systems of records, which had traditionally been built. Now we are putting a new layer in between the system of intelligence, which is not a tool, not a specific technology. It is everywhere but nowhere. It is nowhere but everywhere. So it's like electricity, right? So I think we are getting into that sort of a pervasive kind of technology where we are building an intelligence, embedding it within the process. And I would talk about Lean Six Sigma as ⁓ as a very strong accelerator for that. So there are opportunities to to take a step back, look at it from a design or architecture standpoint, and then have a well-informed strategy by which companies can take it versus just chasing a specific tool, because that may give you some short term benefits, but the long term or the delayed gratification is going to be compromised in that case. speaker-0: Yeah. From your experience, tell me how would you evaluate an AI investment before actually greenlighting it? I've read that your your mantra is meaningful use of information, not just technology for its own sake, right? Treating AI as an enterprise capability, as a system, and actually you can use that in practice. So, what is your way to evaluate any AI investment before you actually say that okay, let's go ahead and start building? speaker-1: So, this is very close to my heart, the question you just asked. This is my dissertation topic as well. So, about the AI readiness, what are the risk and the success factors that the company face as they do the AI readiness assessment in order to harness the power of AI, right? So a prudent way of looking into it is in terms of the capabilities. So, what business capabilities are you delivering today, at what maturity level, and then looking into what Technology dependencies, be it infrastructure, be it data, be it architecture, be design, API, information security, all these things taken together, which are data privacy and security, not to forget that, right? Very important in this space as well. That is how these capabilities help in making sure that the investments that we are making, how they would integrate in a way that it's a well composed solution. So in order to do that, like you start ⁓ building a a house, you start with the foundation, then you build the ground floor, then the first floor, then the second floor, not the other way around, right? So I think it's important to build the foundation there first. Because that forms as the as a strong pillar based on which further scale out would happen as your business grows. speaker-0: Yeah. So I know I I completely agree that we need to have a foundation layer because you're going to put things on top of it. But how about the companies, those who have already got legacy of traditional data, multiple systems, fragmented data, how they can really get started? This theory seems to be simpler when any startup is coming up when they have nothing or maybe minimum amount of data, they can get the foundation much faster and take it ahead from there. But in case of ⁓ regulated industries, right? If you talk about healthcare or life sciences or ⁓ even finance, how could they really get started? A couple of best practices would really help. speaker-1: Sure, sure. So this is a very valid burning question, as we just said. There is a very recent Mackenzie paper which talks about this. That is what kind of strategy companies can take. So they talk about those similar on the similar line that in in the established companies, existing companies, again you can tie Clayton's theory of innovators' dilemma and also in his recent paper on this as well, they're very well aligned in a way that ⁓ For large organizations, they just cannot go through over changing everything overnight, which would huge funding disruption. So they are they have no other option but to take an incremental approach. Increment and learn. So the Sun cost which is there in legacy systems. So that's what we did a few years ago when we went on digital, right? If you look into the pattern, the pattern is very similar. That when we started with digital, they were batch mode of operation between systems, digital report twenty four by seven operations, event driven architecture. So what we did was we built ⁓ web services front end in front of the legacy systems. So there were cobalt systems in the back end or kick systems. So we built a web services layer using K2E or dot net. And then we were able to build event based architecture by which we use queues, JMS queues or or events in order to To do a sort of a real time, give a illusion of real time usage to the customers, although the backend systems were still operating on the batchboard. So I think that adaptation, that facade is what we would see in this case as well, but it would go beyond the business logic to the data layer. That is, we will we will we'll be able to put some additional, I would say, massaging on the data, keepability through a through a robust data governance process, data quality process, by which we can improve the quality of the information, which is key for the machines or as we go for automation as much as we can. That is looking into the the the the the the real time data or the fidelity of the of the data, the data quality factors, dimensions that I try to mention. That those things can be done in addition to the sources. So it puts additional, I would say, force or additional additional weightage on the warehouses or Or created data, data virtualization, newer concepts as they are emerging. With new infrastructure, they are becoming more and more realizable. So there is some additional ⁓ processing that can be done in data hubs or data virtualization, which would keep the back-end investments as is, and an incremental investment, not an overall, but an in incremental investment, would let us focus on getting the data into that central repository, into the lake house architecture, which is evolving and there are a lot of new capabilities coming with Apache Iceberg, open table formats by which you can take advantage of the of the low cost cloud storage, cloud, cloud, cloud compute versus premium multiprocessing SP architecture. So there is a cost benefits deflationary ⁓ advantage that's coming with this. Companies can take advantage of it and put a strong data quality governance, which possibly in many cases were ignored in the past, but now there's no choice. You really have to build that in and make data higher quality and ⁓ easily, as I said earlier, as you said very rightly pointed out, meaningful use of information. We're not talking about data anymore. We're talking about information. Then we'll talk about knowledge and then we'll talk about this time. speaker-0: All of a sudden, from a couple of years, data has become so important. Previously, it used to be a kind of a boring topic. It was not fashionable at all. Before humans were the integration layer. People just ⁓ prepared over the bad data they used to rekey, reconcile, spreadsheet, apply judgment wherever it was ⁓ disagreed. And now, since the paper is gone, AI is doing all of these, things are getting exposed now. The weaknesses are getting more and more exposed. You have helped organizations in ⁓ technical debt reduction and simplification. So give me a real moment where AI exposed a weakness nobody was really interested in knowing, but you took it out and got that moment alive. If you have any example. speaker-1: I can I can give a few examples. So as much as we say that we have manual processes, right? It is taken in a negative connotation, right? A company we are doing manually, people are doing things manually. Of course there is cost, there is a scale issue with that, seasonal issues with that. But it is sometimes it is mind-boggling that how much a human being can do, use their intel human intelligence and they can fill in the gaps, they can use their experience they can use there in order to fix a lot of issues. It takes time, it comes at a cost, but they can get a lot of things done. Now when we talk about machines, now we are talking about the minimal involvement of the human being. So we are expecting everything to be done on the happy path, right? Everything should be done deterministically as they're they are done. So now more effort has to be put in order to make sure That the data quality is up to the max maxed out. Schema drift, which is a very important thing that happens, that is when a database is created, over time, there is a drift that happens. And unless you have a proper metadata management capability built in, those things rely on humans to use manual processes to check, check and balance that which goes back and forth between the technology data stewards or data engineers. with the business users, with the operators. And those things take time. So now we are getting the benefit of the of doing things faster, doing things quicker. But what that requires is that the the the data, underlying data to be pristine, to be properly qualified. And if you look into the the newer capabilities which are coming along in terms of ontology, in terms of knowledge graph, so the RAG, which is usually used as retrieval augmentation generation, Which extracts data from standard databases. Now, in the case of knowledge ref, in in terms of this, now we talk about multiple additional capabilities in terms of nodes, edges, and the triple objects, right? Subject predicate. So we're putting in context. A simple example, I'll just take a very simple example without getting into the complexity. That if I tell you that I crashed a car, and I also tell you that I crashed a party. You can understand the difference between the two because you can crash a party, you can crash a car, or to a machine. That's not not intuition, it's not something they can have intuition around it. So I just give a very simple example. Now think in terms of multiple data elements, data lineage issues, then the sources, defining them differently. Lot of effort has to be put in order to understand the ontology and the lineage of the data through processes. Which the companies could have lived without over the past. But now it is exposing, as you rightly said, that is, it's not enough to store data and retrieve data through primitives. Now it's about making sure that we build some rigour and robustness around it so that all the data that we that we ingest, we should be able to be prepared to go to higher level of complexity, higher level of redundancy, and make them more sustainable. in terms of scale out in terms of the different usage. speaker-0: What does AI ready actually mean when the end user is a clinician in the field, not actually any analyst at the desk? What do you think should be changed first? speaker-1: So I would say again it starts with LinC I I will I I will go by the once one theory which is of the Links Sigma, which starts with the customer, how it is going to be used. Not how it's going be produced by the technology, but how it's going to be used by the user. Now, very soon we'll possibly need to have another variant of that, which is not a human user, but an agent, right? Or a hierarchy of agents, if you will, right? So that's that's a different conversation. So let's let's just assume in a very simple case that you start with the user. The experience of the user is extremely important. Typically, when we talk about agent, agent DKI, we talk about a goal or a plan or an objective, then certain resources, rules, and some guardings, right? So what is the limiting factor within which the agent would be able to work? And if it goes beyond that, the agent would call it out as something beyond its scope and it will not hallucinate or give an improper guidance to the user, which is very important in a regulated industry, right? So guard rules is as important as the response or the action. So let's take, let's assume that all those things are taken care of, right? You do a proper design, you start with the user, you make sure that you understand and capture the user's needs very elegantly, right? Listening has to be there, empathy has to be there, proper understanding of what the user was doing before. Now you're taking the analyst out and we're putting an agent over there or a clinician in the field with not having adequate support staff to support them or walk them through if there are any unhappy paths, right? So that's where the proper thinking, design thinking comes along. So that's where you start with brainstorming. And typically, as we build systems, so the journeys have to be understood. So all the capabilities that we had with agile development, with customer-centric design. in terms of human centric design, in terms of customer journey mapping, all those things are there. We need to go through all those things, right? So they are not not disappearing. They're not going away. The design aspects don't go away. But so it it is it is I would say it is a more more defined understanding of the of creating real definitions and then building the design in a way that it would work and it would not be brittle. So think in many cases, as you mentioned, with fragmented single tools doing different things, building this is very important thing that an unwanted thing that we come across when you want to make updates. Adequate testing has to be done. So if we build it in a way which is seamless, try to have an orchestration layer. Try to make sure that all the privacy and security is not an afterthought, but they're embedded into the design. Architecture is there in all the layers, infrastructure, integration. data, all those factors have been put in. It is just just we're just doing the right due diligence and being disciplined in our design, in our deployment, in our planning, and so on so forth. So all the things which we should be doing, I think they are reinforced in this case that we have a have a definitely a need for speaker-0: Many decades enterprises have invested billions of dollars in building system of records, ERPs, CRM, EHRs, data warehouses, designed to capture transactions with precision. So now AI is asking those same systems, like how you mentioned, nobody is telling you to replace those systems, but AI is asking those systems to reason, explain, recommend, or even take decisions. That is the difference in the job now, right? As per you, what are the architectural changes that are evolving in this age of AI? So speaker-1: We given the the the answer through a context of what we faced say few years ago, around eight or years ago. So if you think about it, our data centers, our mainframes, they were all built for one purpose to do transactions. They never thought of the user experience, right? They were built with robustness, security and reliability in mind. So then we went into the era of digital where we went directly to the customers and we had experience, customer experience as and twenty-four by seven operations as an imperative, right? So with that we realized that it may not be possible or may not be cost effective to build those capabilities into the old legacy systems. Rather, build systems of engagement which would focus on the properties of maximizing customer experience, which would by nature be nimble, be agile, And we would be needing to make multiple changes to them based on the seasonality of the of the business, right? Or business improvement that would happen. So an approach that we had used in this case was that keep the back-end systems as is, the investments, in many cases some cost, give them as is, and then build a facade, as I mentioned earlier, build an integration there, and then build applications, mobile applications, for example, which give a very good look and feel, very good experience. w of course without compromising on security, using APIs, APIs became more powerful or more more popular than back batch promote data operations. So we saw an ecosystem being built out or capabil ancillary capabilities being built out. I would suppose that the same thing would happen in this case as well. That is, you would have intelligence. It is it'll be sort of faced approach in embedding intelligence into the back end systems. versus starting with the front end systems. So that's where they would start first, but slowly and slowly they will make their way into the into the back end as well. That's the journey, that's an evolution that we will see. It may not be a revolution to to begin with. That's where with maturity, with testing out, with reaching the plateau of productivity, that's where we will see more and more of innovation happening, beginning with the front end, and then slowly making way to the back end as well. So that's something which we have seen multiple times in the past. It is the pendulum swinging from one end to the other, right? So that is something which the proverbial thing that we would possibly see in this case as well, that is we would be we would be seeing that it'll start small and then it will permeate into the into the other parts of the architecture. speaker-0: Let's talk a little bit about long term value instead of short term hype. Like you have mentioned before. Whenever there is a shiny object, everybody runs after it and then try to build on top of it. Every technology wave has created winners in the short term, and that is what even Gen AI era is ⁓ doing. Very few actually create leaders in the long term. And the companies we still admire today, like Microsoft, Amazon, JP Morgan, Netflix, they have constantly working on making sure that they have the best architecture, enterprise, and engineering discipline, long-term capabilities that outlive every technology cycle. So what principles you think remain timeless despite of rapid technology change? speaker-1: Yeah. I would say I would point to one person who I admire very much. His name is Warren Buffet. So he doesn't chase the market, right? He doesn't go after the bright shiny objects. If you look at his investment pattern, he invests in Coca-Cola, he invests in stocks, which are mundane, which are boring, but they are used again and again and again, right? I'll give another example from my real life. That is, I usually go to the dentist once a year. 30 minute point. But that's not what keeps my teeth healthy. I brush twice a day. So that regular discipline, right? So so the the first principle that you meant ⁓ that you sort of mentioned is that we sometimes we have to balance ⁓ instantaneous gratification with the long-term gratification, right? So we as human beings, our immediate reaction is that is, we go for instantaneous gratification. What we see now, it's important, it's very important, right? So most leaders lived by most companies they live quarter by quarter, right? So that's something which we'll be cognizant about. But at the same time, look at Amazon, right? In the in the two thousand, in the dot com crash, they lost eighty percent of their value, of the the investor value. But still they were very focused on the not start principles that they had. So they built it was an internal data data IT operations company which built the real large the retailer And also now that the infrastructure, the web infrastructure we have created. So it's because they had that they had the the belief, the faith of the leap of faith in terms of sound architecture principles that they built on. They built everything without making shortcuts. They had a strong governance on their architecture. They built everything loosely coupled. So they used all the they built everything from the first principles. They're not making shortcuts. Maybe they made shortcuts. But they kept a track of it and they remediated those once they went through the through the business cycle. So I think what that created is again a foundational architecture, foundational principles based on which they could actually deliver long-term value. Short-term value came that would come out automatically. But I think the long-term not-start capability, that is key for a company which is serious about survival over decades, right? And we have seen, as you mentioned, we have seen Morgan Chase, Netflix, several others. Netflix came up with a very different architecture. They took advantage of the of the new streaming capability that that was new, net new, and they monetized it. So similarly, other banks, some insurance companies, they have also done the same thing. That is, they have really thought it through and they just didn't build something which would meet the needs of today and would not be able to sustain. into the tomorrow. Of course, changes will come, but those changes have to be adapted, the changes have to be factored in, and it it it is it is it's a wise decisioning that goes around the core platform. So I think what Amazon built was a platform, Microsoft built with Asure is a platform in a fabric, which literally stand the test of time, with incremental innovation happening on those. But the key, the foundational part doesn't change that much. And they can very easily incorporate new technologies, new innovative technologies, adapt to those and make the platform more capability oriented, richer by future capabilities. Because one challenges that we have is usually the front end gets more feature rich or overengineered if not checked, and the back end gets less investment and more technical debt if not governed. So I think those companies which can do the right balance between these two. They can literally stand the test of speaker-0: the statement when you said that leaders should have more balance in their innovation with operational discipline. speaker-1: So this is a difficult question because this is something which is evolving, being built and experimented with, right? So in many cases, the operating model gets prompted by the AI governance. So the governance process, which is very malleable at this time, which is being shaped, right? With the new technologies, with the new realization, with the new prioritization of activities that are being built. So typically what I would say is that That for companies it is a test and learn cycle. That is it is difficult to get everything right in the first go because as we had seen in the past, a classic example is that of big data. So when we started with this, the company started aggregating all the information. There was a huge throughput of data which came along, and we realized that all not all that data were important. So we went to smart data. From big data to smart data. That was the the the the plateau of of productivity which happened, right? So in this case, I'm going to assume the same thing is going to happen. That is, as we have some use cases, limited use cases, which go through a full generative to agent AKI implementation, there will be organizational learning, institutional learning, which people will learn. And this would be And the radars have to be really open for for learning from those. That is, I'm going to going to strongly believe that there will be learnings which we'll need to factor in into our future initiatives. And there would be certain cases where the test of hypothesis is going to fail and we would have to redo them or re make some course corrections as we go forward, right? So I don't have a specific answer, ready-made answer for this. I would s assume that in regulated industries, not assumed that that's possibly a strategy that we took in my my private company, is that those where there is higher regulatory pressure, we would stay away from making them our first test and learn topics. We would rather work on those within the periphery of the organization, where the risk of of data leakage or the risk of the ethical use of AI would be minimal, safety issues would be minimal, and Because there are a lot of things that we don't know today, which are not really technology problems, which are societal problems. But now as we are working with AI, we're entering into that that boundary where it's not only technology, but it's also how society is going to use that data, how the EU acts, how the different policies which are still evolving, risk management which is evolving, how those would be would be putting their policies for that. speaker-0: Yeah. speaker-1: But for simple things like that that expense reimbursement, there you can have a very granular, very well-defined policies, and based on which you can get some efficiency by which people can submit expenses through a mobile application. Those things can be auto-approved and it can save some cycles. I think these are some of the tests and learn that companies can take. Very simple processes where the risk of customer challenges or customer issues or ethical issues. can be can be minimized or eliminated and we have a sort of a safe harbour in within which we can we can test out and learn. Most important is learning as a part of it. That is where the data is failing, where the LLM, we need to make sure that we provide more information in terms of again the different technology arguments that that I don't want to talk about that here. But how we can make use of the of the data. more efficiently, more effectively and govern the result. Not only in terms of the output, but also in terms of the outcome. That is, are we really impacting the the PL, right? Are we really getting shareholder value? Are we really getting speed to market? Are we really getting lower total cost of ownership? Economics is an extremely important part because when we initially do live in the data domain, when we deploy this into the cloud, I've seen some of those numbers. Numbers can be very high. ⁓ So that cost benefits analysis, that TCO analysis has to be done. Which when you're doing the initial part, you possibly have missed those things. So it's the full picture. Getting the full picture end-to-end is important. And that's where I think architecture, finance, HR, legal, privacy, all these departments need to come together in order to make sure that we make the right decision. speaker-0: Yeah, and that leads to what I'm thinking right now is every major technology wave has eventually become invisible. We no longer talk about cloud first companies or internet first companies. We simply expect those companies to be ⁓ adhering to those technologies. I wonder if AI will ever have that time and it would follow the same path. Perhaps in a few years we won't distinguish between AI and And non-AI companies will simply distinguish whether organizations built on that capabilities to continuously learn, adapt, and you know, make better decisions like how you mentioned and those that cannot. speaker-1: This is a very interesting point, the question that you asked. So I follow Ray Karzwa, who has to who talks about singularity or AGI. And also you will know Harari who talks about AI from a very different historian's perspective. So there's a concept of singularity, that is when machines, AI, and humans are going to integrate, right? Now what will happen when singularity happens? That's not going to follow the same same paradigm that we are living in today. So a lot of uncertain, unknown is possibly is on the ⁓ is at on store for us. So that's where I think singularity and there's a fantastic book by Ray Haswell on this. So it is very difficult to see the thing I would like to mention here is that when we talk about technologies that we have talked about before cloud, data, ⁓ agricultural innovation, industrial innovation, in all these things humans What in the center? speaker-0: Yes. Now we are trying to replace speaker-1: Now we are replacing humans. So what that would mean, we don't know. Right? Maybe it would be very simple, but maybe there is some element of uncertainty that we will we will be up for some s for some surprise. Don't know at this time. So so that's I think if we take the concept of it's a concept at this time. The concept of singularity, what would happen? There are predictions that are that are made by many forward thinkers who are making a lot of lot of very well informed or very very good predictions. There is a very good paper by by Amadi of ⁓ of Anthropic who talked about the by love of grace, five basic five points where AI would change a lot of things that we do in our lives, very forward thinking ⁓ view on on the and and that is something which we built ⁓ anthropic on. So I think what we would be there would be some surprises, I would assume that. That is For the first time, humans which have ruled the planet for thousands of years. So possibly we'll see some competition, right? Which we don't see from other species. We may see it from here. And how we deal with this, it is not a technology issue alone. So that's why socio technical, right? Economic, all those things, legal, for example, the laws will change significantly for that. Privacy rules will change significantly for that. So I think we are talking about a new ecosystem which is going to evolve. And ⁓ it's not fiction anymore, it's reality. But ⁓ as we say that that fiction str the fact is stranger than fiction. So possibly we'll see something like that in our lifetimes. speaker-0: For sure. And I'm very positive about it. So Arna, last question for you. What advice do you give to the leaders and CIOs, CTOs, CEOs who are currently shaping this AI strategy? speaker-1: I would say that it is important to look into the foundational capabilities. So the change we are better positioned to look at it in an informed way through an evolution rather than a revolution. So yes, changes, destructive changes will happen. There will be natives which will come along. But for established companies which are which have a business, which have a set of ⁓ customers to support, I think there would be a a method of incremental enhancement, test and learn. I think that is a model that the Kinsey paper talks about that as well. That is for established companies it is possibly a good approach to adapt, make a change, adapt to it, and then have a proper transformation roadmap in order to manage it, like we manage any transformation, right? Galvert model, curtains model. So any any established model would work in that case. And ⁓ in some cases of course there will be the dark horse, there would be the the disruptors which will come along, possibly they would be natives which would be using AI. But again, time will prove those things, those test out those hypotheses. But I would say that take a well informed, step by step strategy, minimize risk and maximize production and efficiency and effectiveness. I will highlight on one thing which I follow closely. That is Plans are useless, but planning is priceless. There will be disruptions, there will be changes, changes will happen, things will emerge. ⁓ But if someone has the discipline of planning, then the probability of having success is higher than someone who is not not able or capable of planning. So again, Mike Tyson Tyson's famous quote, all strategies work unless you are punched on your face. So I think it's it's important to to to build that capability. speaker-0: ⁓ nice. speaker-1: the discipline and be prepared, be ad be flexible and adaptable as changes happen. Capture those changes through the filters of risk minimization and perfect maximization. And of course there's there's a big ethical piece to this as well. So ethical that's important in this case. speaker-0: Thank you so much for joining Tech Talk. It was pleasure having you. Great insights and I'm very sure that audience is going to love it. speaker-1: Thank you so much, Edison.