How AI Changes Workforce Planning with Vijay Swaminathan

Welcome to the future of Less Work, where we explore how work is transforming one conversation at a time. I'm Riko and in today's conversation we'll dive into the data of work. As organizations try to redesign work, roles and processes for AI, what we're really doing is looking into the data that describes what these jobs and processes are really doing. And this is harder than we expected it to be. Maya Ha-Moment in preparing for today's conversation was a sudden realization of how different, for example, strategic workforce planning is going to become. Used to be about headcount matching needs used to translate into org changes, hiring needs, headcount allocation. But what happens to all these when we can add intelligence without adding people? Or we can add people with contracts without hiring them or allocate work between humans and machines? My guest today sits at a very unusual vantage point. Vijai Swami is the co-founder and CEO of DROP, a talent intelligence platform that helps enterprises shift from reactive decision-making to proactive lead growth. And I invited Vijai for this conversation because of his perspective in linking between talent intelligence, AI and work for spanning in new ways. Vijai, welcome to the future of Less Work. Thank you, Nith. Super privilege to be here. Thanks for having me here. So let's start with strategic workforce planning. What does that mean today? I think the, I mean, you touched a few interesting points in your opening comment if we can a little bit anchor on that. AI has been evolving for the last, if you take from 1940s to now, from the Turing Test Times to now. This is about, I would say, about 70 to 80 years, right? So evolution. And the last post-chat GPT, the large language model components of AI has just taken off in a very big way. And as this was happening on the AI side, strategic workforce planning was also going through a less noticeable innovation within the companies, where largely companies were using it to match supply and demand. The supply demand gap has been the primary focus of strategic workforce planning. But with cost pressures and innovation pressure increased in enterprise. They were also looking for various ways of bringing in how do I reskill my enterprise? How do I bring in more innovation into the enterprise? But the problem was the strategic workforce planning was not getting enough attention because typically it's a small function with 3, 4, FTEs trying to model big organizations. Right? So budget is always a problem for SWP functions. But with AI coming in, I feel like there is a new sort of energy into SWP function because the problems SWP is solving for. Or the same as the problem that the AI transformation efforts would require, right? So I think I'm going to step away from HR speak for a second at the basis of this, right? Organizations are allocating work in people. And the people need to have the right skills at the right time and the right structures, right? So conceptually for a second, strategic workforce planning is about looking at the organization and asking, how do I build it and allocate the resources in a way that accomplishes what I'm trying to do, right? And that's what we've always had to balance that. I think it's just that the pieces that suddenly fit in the equation, they used to be just, you know, people work in some boundary conditions around skills and structure. And you're right to point out and I didn't in my opening that first of all, we've got skills that are, you know, changing on us very quickly. But we've also got different ways to bring in people. We can, like you said, we can reskill them, we can hire them, we can train them, we can then bring non-human capabilities. Correct, correct. I think now we can do lot more with less assets per se, right? Less human assets. And how do we do that? So that has also become the preview of the SWP function these days, where they are evaluating not just the supply side, but also the demand side. So in many ways, SWP with AI, I believe, has come to, you know, a bit more energy, bit more excitement is happening. And the broader enterprise integration is beginning to happen to your point. SWP has always been viewed as an HR exercise, right? So, but now the broader enterprise integration as a SWP as an enterprise function is beginning to happen. So for me, SWP is all about tied to intelligence, AI transformation and innovation enablement. That's how I see SWP transfer here for you. So what's the data challenge that we have these days in the organization when we try to match, you know, the right people with the right work and the right skills in the right question. Fantastic question. I wrote a paper over the holidays. One of the things that we have always, from the industrial revolution time frame, we have somehow anchored on to this job description as a central asset for understanding what people must do. And as a result, we have developed sort of a fairly, I would say, rigid taxonomy of the year of the occupations, year of the job families, year of the job rules. And as a result, it's an asset that has evolved for a very long time. And even if you look at enterprise at CM tools, whether it is work-based, success factors, oracle, let's see them and so on, they have anchored on this asset of job profile job description quite a bit. And that's how they have grown. But the problem when we did the research is either the job descriptions are outdated or people are doing something else. I mean, how many times we have heard I was hired for something else, I am doing something else. So one of the concepts that we sort of pioneered, and there was an MIT study on this, is the, in enterprises, work is largely hidden. It's not visible. It is hidden in multiple process maps, standard operating procedures, contractual requirements, context, and interconnectivity of various job rules, like in a network, through one way. And the biggest challenge that SWP planners will have is, all my data assumptions about what a particular job rule is doing, accurate. And if that is not accurate, how do you translate that into proper tasks and, you know, workloads and then say, these tasks can be automated, these tasks cannot be automated. So you're basically saying, as we bring in AI, and some of it isn't even AI, it's just simply AI is pushing us to automate processes, which is still not even. Yeah, not even AI. Not even AI. It is understanding the context of what the company is all about, right? So SWP planners need to sort of have a lot more consulting ability to understand, hey, within my company, what is the uniqueness of a supply chain analyst, for example, right? So rather than just looking at job descriptions, what are the additional things within the context of a retail company or a manufacturing company that they are doing and bring in more insights into what their responsibilities are? So I think that is not necessarily AI, but. Yeah, I'm trying to figure out the for what, right? I mean, part of what we're saying here is, as you bring in technology and reallocate work between technology and people, you need to understand, as you've said, right, the details of the work so that you can make, you know, judgments and reallocate them. It feels to me like the place for this is probably the owner of the work. So nobody other than maybe the manager of the team or, you know, owner of the process would be able to get that level of detail. That will never appear in job descriptions. Fantastic question. I think what strategic workforce planning needs to attempt to do is create a centralized repository of an accurate description of what is happening in an end. Enterprise, right? So what is the purpose of job description? So you have a centralized repository of across all the jobs. This is what people are doing and have that as accurate as possible. Now with the arrival of LLMs, can we read all the process maps? Can we read all the strategic documents associated that? is embedded within a particular department with a process owners, can we centralize all that and in an LLM way add to our task library, right? So that is why what drop I've come up with is okay, we are not going to just take job descriptions, we're going to read process maps, standard operating procedures and other vision document that the CEO might have published to say that supply chain analyst is expected to do this, is doing this now, but it is expected to do this in the future and start getting a library of task correctly. Then you can say out of these 10 tasks that the supply chain analyst is doing and expected to do, what can be done by AI and what should be done by human, right? So that vision sort of comes in very easily if we do the homework of accurately documenting what are the expected task out of a role. For example, if you think organizations are able to document their tasks or do we just need to find other ways to track what's going on and figure it out? It's a combination of some functions have better documentation, some do not, right? So for example, in technology, right? So there is a lot more awareness, there is a lot more documentation of what the new age front and tail and should do because in UI has completely changed in the last two, three years, right? So it's no longer the dashboard, click and drop type of UI, we know a natural language UI. If you look at DevOps, for example, another research that we have done, it is no longer just making sure infrastructure is available, there is heavy emphasis on cost optimization because what you pay for machine is now at the central of every infrastructure optimization problem. So all of the evolution is a bit different for different functions, but by reading CIO priorities, by reading CFO priorities, by reading the process maps that engineers are following or resources are following, we can bring in incremental task intelligence to the on top of the job descriptions, right? So I think that becomes an ongoing journey to your point, it's not going to be like, oh, everything is there, we just put LLM and comes in, but can we institutionalize that process so that we know how work is changing and that becomes a central repository of tasks and skills and then on top of that, we apply whatever AI assumptions are in terms of what machine is going to do and what humans are going to do. So that I've seen in your research a description of work kind of split into builders, orchestrators and synthesizers. What is that? Yeah, I think that's a great question. I think when we were all kind of shocked with the Rival of the chat GPT, I think the a few section of the team where we were already involved in AI, quickly realized that there was a mad rush across the globe to get a lot of what we call as the foundational model of a I talent like people who build AI, build AI inside the company, right? So, but very quickly, enterprises realize that talent is not needed, like not every company needs foundational model building capability, right? So what we need is ability to use the AI that is rolled out by the foundational models and also integrate various AI capabilities that various vendor products are putting out like work days releasing, success factors releasing AI, multiple products, tech stack that I have already released multiple AI components. So it's becoming more important for us to data engineer and understand what the MCP servers are, APIs are and see how to best use the AI capabilities that is coming out of my vendor products and also the foundational models. We call them artists. Are you talking about orchestrating other tools out there? Other tools and also foundational models, right? So because each company cannot build a foundation model by itself, they have to use one of the big models. But using that model also requires some serious talent. So that's why we call it as like the builders and orchestrators, right? The building talent, it's a small talent that is available across the globe and you know you would see that the big talent, whether it is, for example, in Israel or San Francisco or Bengaluru, India, these are very small section of the talent. So, so you're not talking about that within the organization. You're saying within the organization, most people wouldn't be building. Correct. Within enterprises site, unless you are like hyper-scaler, there will be a lot of builders like in Google or in my thoughts. What do you think leaders are oversimplifying the people's side of AI transformation? That's a great question. I think today when an innovation like this hits, we tend to get carried away, right? So I think the biggest thing is the tasks that are left to human, after AI transformation, there is a lack of understanding of what, how complex that is. I'll give you an example. For example, when in an agent to agent, in an agentic world where multiple agents are exchanging data and then they are solving for a particular problem, the biggest role that the human can play is verification, for example, right? So, but verification is not a simple task. We are now learning that verification is a multi-layered process. Like in the previous world before machine-scain, verification is actually a very simple task. I have done everything and all I have to do is some checks and balances. There is a checklist. I need to make sure that checklists are checked. But in AI world, verification is not a checklist. You need to check for buyers. You need to check for governance. You need to check for expanding the data assets, for example. You need to consider different cultural elements if you are a. So maybe it is a checklist, but it's a new one. It's a new one. It's a new expanded one, right? So, and also, it may not be an entire checklist because there could be a lot of unknowns. Like we are still learning where the models could go wrong and go right, right? So, I think that part is severely understood by leaders, right? So, it's not understood well. And as a result, we may be tempted to say, you know, I don't need enough humans. I just one person can do it or two person can do it. And in the biggest problem that leaders have is to think AI is just LLMs. I think that is another big sort of a challenge because when you roll AI, you are considering multiple technologies. Like if it is a machine shop, you're considering IoT, Edge computing, robotics, and so on. So, all of this require a lot of context to be added into AI and that context is still owned by the humans. So, you could get carried away and think that, hey, you know, I can cut X number of FTEs or I can operate and that could go very wrong very quickly. I think this is something that leadership needs to be very careful about. Yeah. So, so I think we've been talking about this in a number of episodes. I think that we think of this as a, as a math problem, right? I mean, you bring in tools and you can replace people. And there's an assumption there that you understand exactly what those people did as they were doing the steps that you identify. Exactly. And you're basically saying the part that remained behind is different than how a human would describe what they were doing before because because the combination is different. It doesn't cleanly cut. Correct. Content writing, for example, right? So, it's not LLM came and definitely improved, but it is not like there were no technology that helped us in writing even before. So, the complexity of the writing is storytelling and if you remove the human element completely in brand material very quickly, it's going to sound very lethargic and more like that. Yeah. And we see this, right? I mean, you can tell when AI has written text as a poster when people write their own text. >> Exactly. Even when email comes, I'm sure you can now deduct it in two seconds. Hey, this is machine return, this is human return, right? >> Yeah, absolutely. Interesting. So, if you look forward, you know, a number of years, they used to be able to say 10, but now even five doesn't sound like it's music. >> Where do you see this going? >> Yes, if you look at the usage of data, the way that we understand how data represents work in the organization and what that translates into. Where do you see us going? >> Great question. So, I am clearly not able to forecast like three, five years out, because things are changing. At least for the short term, I'm on the cam that there is actually still not lot needs to be explored and understood. Purely from a readiness standpoint of making enterprises truly be able to take full advantage of AI. What I mean by that is everybody is marching towards an agent tick world, right? So, in different capacities, we want a lot more autonomous way of working. We want the human agency to be solving for bigger things, bigger, high growth things. We want the human agency to be working on control, governance, buyers, all those legal things, privacy things. But we still operating in a previous generation way of how we look at work, right? So, we are still not able to come to terms with work not being rigid. Like, if you say micro shifts, some companies are panicking, right? Like, the work is still imagined as eight to five in many traditional companies. They are saying, "Oh, we are in the office. We are not even able to cross that mindset barrier." So, for us to be completely autonomous with AI, there is at least two to three years, there is still tons of work where human agency still dominates, right? So, at least that is the camp I believe in. We will have a lot of pressure coming from leadership. CEO and CEO, because CEO is going and telling, "We are all going to be agent tick." We are going to see a force coming on. Yeah, and they want to see the ROI on the investment, right? I mean, clearly. But you're saying something interesting. You're basically saying, part of their ability to successfully transition more work from more human control to more AI autonomy, requires stepping away from the structures that were so tied to in terms of the industry. Job and the profession and the org chart and the function. Right. Actually, I was talking a lot of CIOs and CTOs. Many of them are not necessarily ready to plan for lower headcount even, right? Like they are from their perspective at the execution level leadership. I'm not talking about the CEO and the CFO, the execution level leadership, like a CIOR. They still are not planning more than 10, 15% optimization while you hear in the industry, like, "Oh, technology should be 45." Because they are not able to let go of their control, right? So because a smaller organization, these are all the industrial age mindset we have, right? Smaller the number of people you think you have less control, less influence. So every year, we are designed to think that if I have 1000 people in my team, I'm a very powerful person in the enterprise. How do you unlearn or relearn that? That's going to be a very, very big change. So, and then if I am not hiring, suddenly, my existing team members feel that, "Hey, maybe we are shrinking. Maybe how come we used to hire 100 engineers every year?" Now, we are not hiring. So all the metrics of what makes me important and then what makes us successful need to be redefined. Correct. Our workload model by design is people's centric. Like I was talking to one strategic hospital specialist. Now, one other thing they were saying was very fascinating when they say, "Say you want to open a hospital," right? By design, the design is actually okay. Why number of patients are expected per year? So that will require X number of triage nurses, X1 IC nurses, X2 doctors. It's a head-con model that is ingrained in us for century, for almost century. So it's changing that. We are not thinking in terms of, "Oh, the triage nurse can be actually a robot." We are not thinking like that. That's a huge change that will happen, but it's going to take a long, long time. Very interesting. We are going to be actually just be incrementally thinking. Okay. All right. We do need 10 triage nurses, but maybe one can be a technology that helps all that. Yeah, or we'll give you a software that will help them triage. Exactly. I think sometimes the way this happens is that somebody opens a clinic and brings in that triage robot and suddenly they're more efficient. And that's kind of how some of changes sometimes happen. Wonderful. So as we wrap up, I actually want to ask you for a question instead of an answer. I always end with a request for, "What do you think?" Is the question we should all be asking ourselves about the future of work? So, I mean, I kind of direct this response to kind of my fellow SW planners and talent acquisition leaders is really what is the true nature of work that is happening in the enterprise, either at a function level or at a department level, whatever term that is applicable. Like for example, marketing, right? We are rather than thinking there is an SEO analyst, brand, collateral builder like that. What are the 10 or 15 workflows that they are doing today? And can those workflows, how do the 10 or 15 workflows, can we automate two or three? Right? So, in terms of workflows as opposed to job roles and job descriptions, because even though you are hired as a job role, with a job role, you hardly look at all that. You actually say, "Oh, I need to go make a SEO posting. What are, how do I do that? I may have to interact with multiple people." So I'm actually connected a function in a networked way. I interact with multiple other people. I read many things. I'm barely thinking about, "Oh, is this a job? Is this there? Was this there in the job description?" So, I think SWP planners especially need to understand these workflows and that's where we're going to focus quite a bit on. Yeah, I think Microsoft calls it the move from org chart to work chart. Okay. And it's a very well-served, very well-served. Yeah, describing those things. I totally agree with you. Wow, that was really interesting. So, Vichai Swamy, thank you for joining me. Thank you so much. A blast work today. Yeah. So, let's talk to you. I know you are such a well-respected leader. So, thanks for your kindness and taking some time with you. Thank you, thank you for sharing your knowledge and for more about Vichai Swamy and Drowpeel. I find all the links in the episode description. If you've enjoyed today's episode, follow the future of less work where you take your podcast and remember your best people do not work because they have to. They don't work because they measure or you tell them to they work because by working for you, they're doing what's important to them in life. Till next time. [Music]