What's Next For AI & Machine Learning with Santosh Tiwari, AVP of Decision Intelligence at Hexaware

(Intro Narrator at 00:00:01) Today, we're talking to Santosh from Hexaware about the current market trends of AI and machine learning. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:15) So how did you fall in love with technology? What's your journey like?

(Santosh at 00:00:18) Yeah, so my journey in IT career is very orthodox, I would say, because I switched different disciplines when I started. I started as a design engineer in a firm which was basically building mechanical designs for manufacturing companies, and I used to leverage AutoCAD as a software to build those designs. Then, you know, I identified opportunities to automate those designs rather than doing it manually. And the way to do that was using some programming languages and a script, the steps out and automate the steps. That led me to get into the software development arena and then switched my field from design engineer to software developer.

(Santosh at 00:01:08) And then I got into software engineering for a lot of e-commerce website development and a lot of custom app development, and then I got an opportunity to work on business intelligence at a project. It was a very new field at that time. So I explored a lot and started figuring out how business intelligence plays a role in a corporate organization and what it means. And then that was like my tipping point towards data and started loving data and leveraging data, how to tell stories to business leaders so that they can take informative decisions. And since then, it's been more than seventeen, eighteen years I have been in the data field and evolved as the field evolved, from reactive data analysis to proactive data analysis, building data platforms, played various kinds of roles, right from BI developer to architects to taking leadership roles on managing teams for delivering data projects.

(Santosh at 00:02:19) And then I pivoted into a go-to-market kind of role wherein I leverage all my background and knowledge in the data arena to position our services to our clients. So that's the role I'm playing currently. So that's been my evolution in that industry, and I love every bit of it.

(Joel Beasley at 00:02:44) I'm curious to know, so when you moved from individual contributor to first-time leader, right? Can you tell me about how you got that opportunity?

(Santosh at 00:02:54) Absolutely. So I was playing the role of business intelligence developer and then elevated to become a technical lead wherein I was guiding the team, not necessarily managing people management, but guiding the team and mentoring them from technical aspects of execution of the project. And then very soon, I joined an organization wherein we had to transition a business intelligence workforce and business from UK to India. That's where I took my leadership role and, you know, built a team around it, took the team to let it understand, understood the business, and then transitioned that business to India, captive of that organization. And since then I've been building and leading teams around data and analytics and worked in different kinds of organizations, right from captive banks to product companies, and then switched to service companies.

(Santosh at 00:03:59) And it's been more than ten years I'm in the service sector. And I love the service sector more than any other sector because it's very exciting, a lot of learning opportunities it offers. So since then I'm in the service sector and playing different kinds of roles and positioning services to our clients.

(Joel Beasley at 00:04:21) And when you made that transition to start leading people from individual contributor, what was the most difficult lesson that you learned?

(Santosh at 00:04:30) So not difficult. However, the most interesting lesson, I would say, is putting yourself in the shoes of a team member and then figuring it out. What would you expect if you are a team member and then deliver the same as you are at the receiving end? So one of the things I learned, or probably honed, is the listening skills, understanding your team members, figuring out, you know, their points of view rather than imposing your points of view and ideas on them. So that was the most significant learning in the initial phases of my career that I learned and figured out very quickly, how valuable it is for the very good of the execution of what we are doing.

(Joel Beasley at 00:05:20) And do you currently have direct reports? Like, do you currently lead a team?

(Santosh at 00:05:23) I do lead a team currently. By virtue of my function, the team is very lean. However, I had been in roles in the past where I was managing delivery and managing teams as big as eighty members globally. So that taught me a lot in terms of how to deal with bigger teams and diverse sets of people in diverse and different geographies and how to align with their sentiments, their culture, you know, work collaboratively and productively.

(Joel Beasley at 00:05:57) For communication on larger teams, like I haven't ever managed a team of eighty people. How do you do that? Do you have to have a communication person? Do you get really good at writing and articulating your thoughts? How do you communicate to eighty people?

(Santosh at 00:06:11) Yeah. So typically, we create a kind of a second layer, leadership within under me for leading a specific track and technologies. So majority of the traction or interaction goes through them. However, we always make sure that being a leader, you are approachable always no matter what hierarchy the team members are in. So I made sure that everybody within the team, I'm always approachable if they have anything to share, any concerns to raise, and then you keep total transparency within the group.

(Santosh at 00:06:52) And apart from that, regular communication channels to discuss. And not only the learnings within the professional, also social interactions in the form of, you know, wherever we can meet in person, gather everybody and talk in person or call them virtually. But regular cadence, regular interactions is the key. That's what I used to do and, you know, make sure that everybody is open, everybody is free to talk. Nobody's getting penalized for any ideas which are not taken forward. Right? So all ideas are good ideas, that kind of culture. So I think that was my mantra to deal with a larger team.

(Joel Beasley at 00:07:41) And before we started recording, we were talking about how you're at a company event right now. Right? Doing some of that communication. Is that correct?

(Santosh at 00:07:49) Oh, that's correct.

(Joel Beasley at 00:07:50) What are you learning? What's the cool thing happening at Hexaware? What communication is going on there?

(Santosh at 00:07:55) Yeah. Very interesting. So when the pandemic hit, right, everybody went virtual. And even before pandemic, we are a global company. We work from different locations. Many of the colleagues, we never met, but we work very closely, and we know them face by face, by video calls and all. However, when the pandemic started, Hexaware took a lot of initiatives to make sure that all the global workforce is well-connected and informed what's going on within Hexaware, what's going on in the outside world, how it is impacting us, and how we can work together to take on those challenges. Some of the initiatives were from our HR teams to engage the workforce globally, learning different skills, for example, you know, whether it is cooking or haircut or music sessions. Those kinds of sessions very frequently have been organized, and people participated wholeheartedly on those sessions to make sure that they're doing something apart from the work. At the same time, they are connecting with colleagues and not only colleagues, their families as well.

(Joel Beasley at 00:09:09) Explain the haircuts thing again.

(Santosh at 00:09:12) Yeah. So a few sessions that the team organized wherein they invited celebrity hairdressers, and then they asked them to teach different tips and tricks for haircutting and designing. So remember that in the pandemic, people stopped going out and people were doing haircuts at home. It was very challenging. And, you know, because we are all not skilled at it and people were getting shy to come on videos because of the nasty haircuts. So that was one way to, you know, hone our skills, at the same time give them confidence that it doesn't matter.

(Joel Beasley at 00:09:53) Oh, yeah. That's where my beard came from, actually. I never had a beard until COVID, and everyone was out of the office. So I figured, hey. Now is the best time if of any to just go through the awkward month or two of starting a beard. Yeah. So I get that. I think I wore a hat a lot. Like, if you look on past shows when I couldn't get to haircuts, I started wearing some hats. So you're exactly right. But it's super cool that you guys integrated that into your company meetings. That's kind of cool. I like that. I haven't heard that before.

(Santosh at 00:10:24) Yeah. So, but this is one example. And there were a lot of sessions like regular yoga sessions to make sure we are fit and fine, not only physically, mentally as well.

(Joel Beasley at 00:10:37) I am curious. When we first started talking, you mentioned the evolution of your experience and your journey, and you had said something that sort of stood out to me about where now you're spending some of your time figuring out how to position go-to-market strategies, how to position services to clients. That's something as business leaders we all run into to some degree, whether we're in a conversation when we're just kind of hearing about it happen or we're actually actively driving that strategy. What's something important to think about when you're thinking about go-to-market and positioning your services to clients?

(Santosh at 00:11:11) Yeah. I think the key aspect is to understand what's trending in the market, where the industry is leading towards, and which are the tools and technologies that are really helpful for addressing business problems. So by that thought, you know, since I've been in a data background, now I'm very much focusing on the decision intelligence area and part of that practice. So we formed this group with that notion. And how do we leverage AI ML technologies to address and solve complex business problems, either automate them or augment them to expedite business decisions by bringing foresights to the decision, right, rather than just human intelligence or human experience and knowledge.

(Santosh at 00:12:07) So that's where the industry is going if we look at it. So we built a whole practice around it, a lot of offerings around how do we enable our customers in different domains to enable them to accelerate that journey. But to your point, when it comes to GTM or go-to-market, it's very important to understand the pulse of your customers, their business, and their business problems, and establish yourself as their trusted advisors. Once you are able to do these two things, you understand their pain point as if it is yours. And if you have to solve that for yourself, what are the measures and what are the techniques you will use rather than thinking, okay, how much money can I make out of it?

(Santosh at 00:12:56) So very often, it entails cannibalizing our revenues. However, overall, if we look at it, if we establish ourselves as a trusted advisor and help your customers to address their business problems in the most optimized and cost-effective way, that's the best go-to-market strategy. Then you will be repeatedly called for advice and you'll be referred to different customers as well, and then business automatically grows. So that's one of the key mantras that we follow. Majority of our business is repeated business from the same customers, and we do get a lot of references. New businesses, what we acquire is majority from the references. So that kind of modest opportunity we follow. Another aspect is being in sync with the journey of transformation. What that means is, you know, you position your service, you convince a customer to take your service. And once they grant you a project, then most of the organizations are disconnected when it comes to implementation.

(Santosh at 00:14:06) Implementations can go wrong, specifically in the AI ML space, which is a very complex field of execution. So it's very important to stay connected, to stay on track for implementation, make sure whatever ROI was discussed and, you know, aligned, we are meeting that goal. And there would be changes, deviations, how swiftly we address those deviations. Those are the key aspects, and that drives customer satisfaction and a smile to our customers.

(Joel Beasley at 00:14:42) Now when you were talking about the AI ML technology to solve business problems and helping with business decisions, every year here at the company, and we've been doing this for about five years, we take a list of all of our customers and we sort it by, like, who are these top 20% of the people that we want a hundred more of, and what are their attributes. And every year, it's subsequently gotten slightly more sophisticated, and we're just doing it with a spreadsheet, you know, some basic stuff. But then I thought this year, I started looking for software where I could put in all my deal information, and it could tell me about time to close. But here was the problem. So your CRM will often offer you a thousand different reports. Right? But I don't need a thousand different reports. I need some AI system to run through ten thousand possible things that could be impacting my deal flow and then present to me in an order of priority the most important things I need to know that happen to impact my deals. And then that way, I can make business decisions on it, whether I change my territories or I do something with the sales team. And so is that along the lines of what you do?

(Santosh at 00:15:59) Absolutely. And that's one of the use cases as you just pointed it out. Similarly, any business domain, there are so many decisions. In fact, you know, there are around 3 billion decisions. Business decisions are taken annually. Majority of those decisions are still manual today. And it comes from the wisdom of the leaders on that business area based on their experiences and based on their biases of past experiences. However, those are not very data insight-driven business decisions. As a human being, we have an exposure to the areas that we have been exposed to. But when you apply a data-driven approach to that, it can combine the experiences of not only yours, but everybody and everyone else who has worked in that field and the business operated over years and combine it all together and, you know, create an insight and fuel your business decisions with the output of that foresight.

(Santosh at 00:17:11) So that's where the industry is going towards. And, you know, a lot of business processes, there are technologies like RPA to accelerate the business steps automatically. However, AI ML basically helps to identify what are the next best steps based on what worked better in the past, what failed. Just like for your example, you know, for CRM, which territory should I go for? What are the parameters that are influencing? Why did my sales traction not work out in XYZ area? How do I segment my audience? Right? Things like that are suggested by AI algorithms today based on how your traction had been in the past. So those are the areas that DI is addressing, and it can go to very complex use cases as well.

(Santosh at 00:18:08) Like, we are advising one of our manufacturing companies to do a predictive maintenance of their big machines, which saves a lot of time versus doing it manually. And it takes a lot of time even in diagnosing the issues in the machines. So we built algorithms which are basically looking at all parameters of what fault could be and identifying for a service engineer very precisely where the fault is so that it can be fixed.

(Joel Beasley at 00:18:36) Yeah. That's becoming popular. I saw one guy who, like, cell phone towers in Hawaii. Like, to get the parts out there, it takes a long time, and it's expensive.

(Joel Beasley at 00:18:48) And so they'd often have to charter private flights or whatnot to get things out there. So they were working on this predictive maintenance algorithms to detect when they're going to fail earlier so they can get parts out to them before they actually fail.

(Santosh at 00:19:01) Yeah. So a lot of cool stuff, and I'm so excited that I'm part of this journey and leading that native service from Hexaware.

(Joel Beasley at 00:19:10) So back to my use case, obviously it was very specific. It was sales related data, and I, of course, have unreasonable expectations that I can just send it data and it be incredibly brilliant and give me insights back. Are we there yet? Are there off the shelf tools that will actually do that with a legitimate form of business savvy and intelligence? Or are we more at the point where if I were a large company as I was having this issue, I'd go hire Hexaware and they would do consultation, figure out exactly the data, massage the data, then build systems specifically for me versus me just putting it into a giant one size fits all system and getting very personalized results. Is that where we're at right now?

(Santosh at 00:19:54) No, not yet where you can just ask your question and then the AI will answer in this paradigm. We're not there yet. However, there have been efforts by all the large CRM companies, whether it is Microsoft or HubSpot or other leading products. They all are trying to build these kind of services on top of their existing platform. But having said that, it will never be one size fit all business requirement or all scenarios. Even the sales process is very different from organization to organization. The magnitude and expectations are very different. So it will never be one size fits all. It will have to be customized and tuned to the way the business function works in a given scenario. So to answer your question, we will always require some kind of customization on identifying the parameters which are important for the organization and the business they are into and customize the models based on that. So we may have base models, if you will, and then you will have to improvise that based on the business scenario we're dealing with.

(Joel Beasley at 00:21:13) And so how do companies currently interface with Hexaware? Do they go to the website and they reach out and they start a conversation with a specific problem?

(Santosh at 00:21:23) Yeah. There are various channels that they approach us. The website is one of the ways. But apart from that, we have been making recommendations in terms of where our services are better than other organizations or our peers. So we have created a niche in our services, and automation is in our DNA. That's how we differentiate from other service providers. Any service that we offer to our clients, there is a certain degree of automation and platform led approach always built in that. So that's how we created a differentiation, and we have been recognized in industry. Our platforms and tools that we take to our clients, those are recognized and gotten patents. So our customers also realize the value of that and increasingly been called to address their business problems.

(Joel Beasley at 00:22:20) What's the most exciting thing maybe outside of Hexaware that's going on in the tech world that's got you really pumped up?

(Santosh at 00:22:27) Yeah. So being a technology person and specifically in the field which itself is very exciting, I keep making myself abreast with what's going on in the market. Right? One of the things in the AI space that is interesting is the creative AI aspect of it, which means there are models, there are techniques developed which are helping artists to augment their work by use of AI. So for example, there are models available wherein you can create the whole painting based on just a few words of input. So there are frameworks like DALL-E, very, very interesting output. I have played with that myself and not being an artist, but I can create a lot of nice art with that software. Right? So that's where AI is leading towards. And not only pictures, right? It's being used to create music, create videos and all sorts of artistic stuff.

(Joel Beasley at 00:23:38) Now I'm just curious. I don't think that you're a super detailed expert on DALL-E, but I know very little, so you probably know more than me because I haven't even played with it. I've just seen the articles where people type salmon swimming upstream, and it's like fillets of salmon. It's not the actual fish. Do those models, are they at the point where they're trained on a specific set of data and they can only do things that they know about? Like, maybe they know about water streams and salmon. Or are they at the point where you could maybe make a pop culture reference to a musician or an artist and they would understand that too? Do you know where that's at?

(Santosh at 00:24:13) Yes. Absolutely. I think DALL-E specifically is being trained on different kinds, different forms of art performances and different genres of arts, primarily painting. DALL-E is focusing on that. It's been trained on a lot of historical data on various artists across the world and their style of working, their style of arts. So if you give an input that, hey, I want to build a portrait of myself or somebody, a famous personality in the style of Picasso, so it will understand how Picasso's style was, and based on that, it will transform that picture and present the output for you.

(Joel Beasley at 00:24:59) Oh, so you can give it a photo and it can manipulate it?

(Santosh at 00:25:02) Not necessarily. DALL-E doesn't take photos as input. However, it has been trained with the pictures of all famous personalities across the world.

(Joel Beasley at 00:25:10) Got it.

(Santosh at 00:25:10) So let's say if you want to create a transformative picture of a famous personality in a different style, in a different setup and background, in a different era, for example, World War II or in a Picasso style, so all these are parameters that you can keep adding as an input in the form of text, and it will use those learnings and build a picture based on that.

(Joel Beasley at 00:25:37) Oh, that's pretty cool. Now Elon Musk, you know, he went around, I think, five years ago and sounded the alarm and then realized nobody really cared about AI taking over the world. So then he started Neuralink and the different AI projects that he works on. I think it's called OpenAI. What do you think is the most likely path in reality to happen with the emergence of AI? Are we gonna get crushed by it? Are we going to evolve next to it and with it? What do you think the most likely path is for us as humans?

(Santosh at 00:26:07) Oh, well, yeah. There is always fear and debate in industry that AI will take over human beings and our decisions. However, my take is AI will get smarter as we allow it to. So having said that, proper governance is needed. So let's say any company which is building any AI product has to be governed, what kind of intelligence they are infusing in. And so having said that, AI will always augment human beings, never replace them. That's what my take is. And as more and more AI applications and technologies are developed, the platforms are developed, even the tools are developed when I say machines, more and more tech users are required to manage those and operate those. So if you look at the fear that AI is taking everybody's job, it's a myth. AI is, in fact, going to create much more jobs than we ever had. The only difference would be these jobs are getting transformed to very technical rather than nontechnical jobs. But apart from that, it's not gonna take anybody's job ever. There'll be an aspect of wrong usage of AI, and that's where the whole governance is required, very stringent governance. Any new technology, there is a negative aspect of that. When Internet was introduced and became public, there are wrong uses of Internet, but there are so many positive uses of Internet that we are harnessing and living with. Same will go with AI as well.

(Joel Beasley at 00:27:48) If you were to go back to yourself back when you were programming AutoCAD and just getting into it and you were to give yourself some advice about the future of AI and where the market's going, what would you tell yourself?

(Santosh at 00:28:02) Yeah. That's an interesting question. In that time, you know, AI was not mainstream. Nobody was thinking about it. Though the concept was there decades ago, organizations like NASA and all were using it in some form and fashion, but it wasn't mainstream. However, if this kind of open technologies were available that time, I would think these designs, and today, if you look at it, nobody does or creates those mechanical designs anymore. Right? The technologies like AR, VR have taken over. You have 3D model building aspects which were not there earlier. And now those 3D models are directly fed into CNC machines and 3D printers have evolved. Nobody thought that 3D printers would be a reality. Rather than CNC machine to carve out a specific shape of a component of a machine, you will use a 3D printer to print it. Right? So those designs are completely irrelevant in today's scenario.

(Joel Beasley at 00:29:13) Now I'm watching the time. Did we cover decision intelligence? I mean, we talked a little bit about it. Tell me a little bit about what decision intelligence is. And then if people that are listening, if they're interested in finding out more about this or working with Hexaware on decision intelligence, how they can do that.

(Santosh at 00:29:28) Absolutely. So decision intelligence is about solving complex business problems leveraging AI and techniques. Right? And infuse foresight in the process of business decisions, if you will. Imagine an ML model guides an individual or an application by predicting the next best action or step based on historical events or patterns, et cetera, and help them to take the most appropriate decisions. So that's where the decision intelligence comes in play. This most often helps in augmenting or automating business decision-making processes and then derive better outcomes. So efficiency improvement is one. The differentiator in other technologies like RPA and traditional business intelligence is, what we were doing as traditional business intelligence was more reactive in terms of identifying what happened in your business and then infer from that and take actions manually. However, decision intelligence is a much more proactive approach wherein AI/ML techniques are being used to understand the insights from the data and then create foresights, what could be the next best action to take, and then help to take appropriate decisions. So that's where it differentiates. And it has a lot to do with leveraging various disciplines of AI/ML, whether it is computer vision or NLP, statistical modeling as well. So all these disciplines, sometimes we combine all these together to address a business problem or solve a problem which was very complex to address earlier. But leveraging these technologies, we can address them in a much more automated way and drive efficiencies in the business process.

(Joel Beasley at 00:31:27) And if people want to work with you on this and learn more about it, where do they go?

(Santosh at 00:31:31) Yeah. So they can contact us for a preliminary discussion to understand what business problem they're dealing with. We offer consulting services in this space wherein we have specialized consultants who understand deep mathematics and statistics and AI/ML techniques. However, they come with very great business acumen to relate with the business problem and then marry these techniques to figure out how these techniques can be leveraged to address complex business problems. So that's the best way. But apart from that, we do come across a lot of different companies who know what business problems to be solved, but they don't know how to do that. And that's where we get engaged with them and build their strategies in terms of technology landscape, in terms of what techniques to be used, and then help them to build those processes. And then the third area is, which I believe is becoming very prominent specifically in large enterprises, wherein they either have built some models or they're in the process of building and piloting a lot of models, but they're struggling in how to operationalize that in a large scale and embed that in a business process scenario. So that's where we help them to operationalize in terms of integrating the output of the model to the business processes and the applications so that the business users can get that insight embedded within the process rather than going and looking at some reports or dashboards, if you will. And the whole aspect of maintaining the model's accuracy and output over a period of time. Models, when built, we do hypothesis testing. We train them, and we figure out, okay, the accuracy level is acceptable for a business scenario, and then we productionize that. But over a period of time, the model output deteriorates when model drift and data drift come into play. So there is a consistent need of reevaluating and auditing the model and figuring out whether the output is still valid for a scenario. And if not, it requires retuning, recalibration, and retrained and then redeployed. So that whole life cycle of model calibration is called MLOps. And that's where a lot of frameworks and tools are available in market, and it's still a growing field. A lot of work is still being done manually, but that's where we help our customers to leverage the best breed of the tool and experiences, how to automate the majority of those steps and not get caught in the wrong decisions and insights given by the models and thinking that, you know, the model was giving better output six months ago, it should be good again, but it won't be.

(Joel Beasley at 00:34:47) Thank you for all of the insights on AI. What's the website for your company?

(Santosh at 00:34:53) It's hexaware.com. Within that, we have a list of different service lines. We have one service line called Digital Core Transformation. And within Digital Core Transformation, we have aligned decision intelligence as one of the service lines within DCT.

(Joel Beasley at 00:35:12) Oh, amazing. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.