So let’s roll back the tape and start with you as a kid. Where did you grow up and what were you like as a kid?
All right. So I was born and raised in Kathmandu, in Nepal. I grew up in the middle of a civil war, and what that meant for me was I microdosed on stress early on in my life, right? And that gave me a certain resilience, which became my default operating system.
I remember being a quiet kid. Unlike my profession would kind of indicate, I was actually a very creative kid, especially in a country like Nepal. There’s no early specialization, so either you are a good student and you’re good at everything or you’re a mediocre student and you are mediocre at everything, right? So basically being a good student for me allowed me to become more of an all-rounder.
So when I came to the U.S. for a liberal arts education, it was very natural to me to study something like physics with something like religion, right? Because again, we didn’t really have an option of just specializing in something as I was growing up. Sometimes I reflect if, let’s say, the need to exist in a capitalistic society was not a constraint, I would probably be more on the creative arts side, predominantly around writing or music.
But life has transpired in a completely different way.
Suman, I grew up in Indianapolis in the 80s. And I often say boredom breeds creativity. Do you feel like you saw that as well, that the lack of stimulus? And I fear, I have kids. I don’t know if you have kids now, but I have kids and they’re so overstimulated. I’m just curious if you think that’s a negative on creativity.
Oh, 100%. 100%, right? Because I remember days where it was just me in my room and I had read all the books that I had. There was to read all the novels in my mom’s closet, so there’s nothing else to do. So as a kid I was forced to entertain myself, and a lot of my early explorations with math were just me trying to figure out ways to hack things or do things in a different way. That was the entertainment I had.
And then I got two kids now. My older one is almost six. And despite all efforts to just minimize stimulation, the world we live in is so chaotic, right? I just worry that boredom is increasingly scarce of a commodity.
It’s yeah, it’s a luxury boredom. And something that’s really not wired into our kids’ schedule at all today, at all. So we’ve got to rethink that.
Standing Up Organizations
And if we move into the corporate experience that you’ve been able to have in your career, and it’s been pretty consistent, right, all around data science and… You have now developed an expertise in standing up organizations. You’ve done it four times and not for little organizations, but for Fortune 50, which brings a whole layer of complexity and regulation.
Walk us through what that experience was like. How did the mandate come in? And what were the business problems that were really looking to be solved by standing up this org? And how did you initially think about this?
So when I first entered the industry, data science was not a term, right? So there were a few things that were happening, which resulted in this capability being in demand. So the biggest thing was there was a shift from traditional informatics and healthcare analytics, which is called informatics. And there was this desire to move from software systems like SaaS to more open source systems.
Second was this need, things like Hadoop and distributed compute were just beginning to be a thing. And the world of big data was within reach, right? So there’s also this desire from large enterprises to tap into that promise.
And then the third thing that was happening in parallel was, yes, organizations always did classical regression and basic statistics. But now suddenly some of these, what is today known as traditional machine learning, was also within reach.
So you combine some of the hardware offerings, the ability to use novel algorithms, and the desire to move away from legacy software into more open source frameworks, right? That’s how the field of data science was born. It wasn’t like the organization set out to say, I want to build a data science function necessarily. It was more around, hey, there’s opportunity here somewhere that is at the intersection of some of these things that are underway. And we should probably tap into that with the right talent. I think that’s how I remember the dialogue transpiring in the early days.
I think when you’re trying to establish a net new capability, and that’s true with AI today as well, right? Like what’s called data science today was called AI earlier. And now this takes a whole different meaning. But fundamentally, the essence of standing up a net new capability, both from a tech, process, and talent perspective, a lot of things translate.
So the biggest assumption that we made when we first started was that if you train people on these new tools, somehow they’ll organically shift into the new way of working. And the truth is a lot more nuanced, right? The incentives, the inertia, all of these things play a big role in this.
Now, second assumption that again is a very naive assumption in hindsight is somehow the end users, the people who are at the receiving end of some of these byproducts of these new capabilities, are organically going to be excited and going to be consumers of the new tech.
Now, unlike consumer tech, where you are trying to think very hard about design, you’re in many ways unconstrained by, let’s say, the regulations and the inertia of the organization. Doing this with an enterprise comes with a lot of moving parts that you need to navigate with constraints of freedom. And you also need to think very carefully about the adoption and change management of this.
So I think a people-focused view of this, like the fact that the people who need to build and be part of this capability are organically going to do that, is an assumption that doesn’t mostly land. And then the fact that the consumers are going to be excited about this is also an assumption that doesn’t really always transpire.
Now, maybe an underpinning framework beneath all of this is, if you’re already an organization and you’re trying to start a new entity within that organization, what that means is you need to create space where there was no space before. And you need to elbow certain people. You need to create new ways of working. You need to create new rules of engagement. You need to just occupy space. And that is not a tech problem. It is a branding and a comms and a politics problem.
So all of these are things that the first couple of times when you’re trying to do this, you don’t think about because you’re so focused on bringing the right people in and building the best possible algorithm to predict this best possible thing. Over time, what you realize is, honestly, those things don’t need to be as complicated. You could build a basic version of those things. But what is truly important is giving a nod to the ecosystem that you exist in and the environment that you exist in and creating a sustainable way in which you can progressively exist and thrive.
Adoption and Change Management in Large Transformations
Suman, there’s so much that we can sort of double click on, but I think top of mind today, and something that you mentioned, is adoption and change management as being so incredibly critical to being able to drive impact and embed AI successfully within the organization.
How do you do that when it’s involving possibly rewiring, eliminating, reorging, like on a larger transformation level, because there’s so much fear and fictional narrative out there around jobs are going to end? And there’s just a lot of concern around that. So, having it been an incredible lesson, how do you approach that? And how do you think about that?
Going back to this theme of it’s never about the tech, it’s about the people, right? I’m a big believer in first understanding the power dynamics. Where does the power center lie? And what are their incentives, right? So forget change management of the end users, for your setup, for your capability, for your solution to exist, you need to have a good read of how decisions are made, how funding is secured, and how people’s incentives are defined. So that I think is one element of this.
Second is, especially true for AI, AI is not a thing by itself, right? Especially in an organization like ours, AI is an enabler. It’s not our core product. It is an enabler of a P&L. So just framing the value of AI in the context of a win for your end user, I think, is a non-negotiable. I think that’s how you start the conversation, right? AI for the sake of AI is never going to take off, regardless of how good the solution is.
Now, being somebody who other people want to see succeed is a very fine art. Understandably with AI, there’s a lot of spotlight, right? There’s a lot of eyes on it. So it’s very easy to hog the spotlight and then forget the people who are actually bringing the revenue in the process. I think those systems are always short-lived because then the people who are actually holding the P&L and bringing the revenue in are always questioning, what is this individual doing, right?
So be somebody who your peers and your stakeholders want to see succeed is the biggest lesson I think I’ve gathered over all of my experiences so far.
You also need to make sure that there is this aspect of co-creation, right? Change management is not something that you do at the end. Change management is something that should be part of the build process itself, right? The biggest way to drive ownership is to give people something to do in the process.
And the way we have hacked this over time, I think my philosophy on this has evolved, but these days when I create product teams, some of the end users come and sit in my product teams as a member of the development squad. Because then you’re shortening that distance between ideation and product design. You don’t have to go and check requirements or gather requirements or do UAT in a separate bucket. You’re already shortening that iteration cycle. And now if you add all the vibe coding and AI lens to it, that becomes even more pronounced, right? Because now you’re going from ideation to prototype within a few hours, if not days.
So these are all things that we try to bake in as part of the general umbrella of change management.
There’s also this issue of product stickiness, which is not really change management. Your product needs to be good enough that people want to use it. Obviously, you can change manage the hell out of a crappy product, but the thing is nobody’s going to want to use it if there’s nothing sticky about it.
We have a separate design stream that continues to think about how do we make the products lovable and not just viable, right? So going away from this traditional tech mindset of MVP. Your product could be technically viable, but nobody wants to use it. And that’s not serving anyone, right? So we’ve been spending a lot of time and effort in thinking through that as well.
And the last thing I’ll say is there is no bigger change management driver than an executive leaning down and saying this is important, right? So this credibility by proxy is probably the biggest driver of adoption. Obviously, all the other things need to be true, but if you don’t have that top-down mandate saying this is how we want this particular thing to be adopted, you’re not going to see any traction.
So those are all the things that I think we try to bake into our ways of working.
Non-Negotiables in Org Building
Do you have some non-negotiables? Having done this now four times, you say, look, I’m not setting out on this journey unless the following two, three things are set in stone. Because organizations have a tendency to kill things by a thousand cuts or have organ rejection or, you know, not invented here. All that fun social stuff.
Yeah. I think the biggest thing is that if there is no ELT, meaning CEO plus one sponsorship, there’s no sustainability. So I think that is a non-negotiable for me.
There needs to be at least a three-year roadmap. If you’re thinking about hiring people, changing ways of working, building things in parallel, getting it adopted, getting metrics and value out of it, you need a runway of two to three years so that you’re not living on a daily basis. We do stage-gate it so we can come up with concrete milestones, typically every semester or every six months, right? But the runway is for two to three years. So being given that space to build is also a non-negotiable.
How about access to resources? How do you make sure you have unfettered access, not just to human talent, dev talent, but also data? Or someone might say, oh, this is my data. Why do you need it for your project? I mean, I could see that becoming a thing.
In my experience, that is a byproduct of executive sponsorship, right?
Right. Once you have that ELT support.
You kind of flash that card.
Yeah, exactly. Exactly. So I call it the king stamp, you know, like back in the Silk Road days, right? If you had the Mongol stamp, you could go around and do your trade in an unfettered way. So you just need that to move things along.
Change Management and the Impact of AI on Workers
So, in addition to having the king stamp, it’s the classic get the ELT or very visible leaders leaning in. Then there’s the bottoms up. But I think the old metaphor is everything dies in middle management. Are there other things? And I also love the not just viable, but lovable. Are there things you see? This is all in the change management theme, right? Carrot versus stick. How do you get people to adopt, incentivize them to do so?
Obviously, we’ve seen some of the tech companies say part of your annual review is going to be an assessment of your adoption of the AI tools that we’ve rolled out across the enterprise. If we get down to the nitty gritty, other things that you’ve seen or done that helps those rollouts be successful beyond the stamp, beyond those who are already going to lean in, but you’re moving massive organizations, right? And it’s people.
Yeah. I think three things I’ll say to this.
One is, and this is a phenomenon that we benefit from, AI specifically sits at this intersection of excitement and fear, right? People are excited about the possibility, but also fearful that their jobs are going to go away. That mindset, I think, is prime for change management. This is what we found. With the right language, you can get people to adopt things a little bit more easily than maybe with previous technologies.
So that’s the base case that we benefit from, specifically because everybody wants to do something with AI because there’s this idea that this is going to make me more ready for whatever the future is. That just comes with the territory and we’ve benefited a lot from that. So it’s not a change management tool, but it’s just the reality that we exist in that benefits us.
Now, besides that, what we do is we spend a lot of time upfront doing all of the business cases, right? Like venture capital people, you can resonate with this. We come up with all of the productivity, efficiency, consistency metrics that we project. And then we say, okay, this is when roughly this investment is going to break even. So effectively, we run like a startup. We can come up with a case, like five-year projection, 10-year projection, what is this going to lead us to, right? And then what are the assumptions that we are banking on? And we’ll periodically pressure test those assumptions with our group of sponsors.
And then once that is defined, then we come up with metrics that become our goals for the semester. And goal setting is probably the biggest way to drive organizational change because as our ELT members have those goals, and then it trickles downwards, then at the end of every six months, there’s this mad rush to our shop saying, okay, did we hit our goals? Because only we are the ones that can track all of these metrics. So just having this top-down alignment on goals, I have found it to be a very powerful process for adoption and change management.
Then the third piece, which I think is a slightly softer twist on this, is it is very easy for me to go and give an update to my sponsor group to say, okay, these are all the cool things that we are doing, right? But it is a lot more powerful for the leaders of my end user functions to come and say, okay, here’s what we are finding with my team.
So I do my best to put them in these seats as much as possible. I’ll do all the work on the backend to arm them with the stats and all the numbers, right? But once they go and do this, one, it’s visibility for them, so that’s great. But second, now they are accountable for it. Every month they have to show up and say, okay, this is what’s happening with AI adoption. This is how I’m personally leaning in to drive this for my group of 1,000 people.
So all of that I think I’ve found to be, you combine all of these things like goal setting, this general excitement and desire to be AI fluent, and the fact that the leaders can be made accountable through certain strategies. You can move the needle a lot faster on change management as a result.
Change management is one thing. Changing individuals is another. And one of the things that we’ve talked about in the impact of AI and the workforce broadly is this shift from what I call being a conductor versus a violinist. And what this means is, as a conductor, you’re conducting lots of different things, whereas a violinist, you’re an expert at one thing.
And at least what I’ve seen in my own experience and also a lot of our companies is the people that are really exceeding and excelling are the ones who have multiple agents, who have multiple things working for them. It’s almost like a manager versus an individual contributor.
And it’s one thing to have change management. It’s another thing to retool a workforce. So this is stepping away from this idea of innovation within and just talking about the impact of AI and workers. I’m curious if you’re seeing that as well. If you fast forward 10 years or think of the advice you’re giving your children of what are the skills that are really valuable, are you seeing this yet? That the talent within the organization that can multitask, in essence, can be running multiple agents simultaneously, solving lots of problems that they previously didn’t have the bandwidth or resources to solve, are the ones who are breaking out and future-proofed? Or is that not something you’re seeing quite yet?
Future-proof is a loaded term, especially when you don’t know what the future could look like. I’d maybe give a slightly different take on this.
I get asked all the time, what can I do to be AI fluent and future-proof my career? And my answer is always, there’s nothing that is out there as a non-tech person that you could not pick up within two weeks. What is out there, like Cloud Core, like GPT Codex? I mean, you can figure these things out with two weeks of dedicated effort.
So it’s not keeping up with every new advancement that is out there that’s going to make you future-proof. What’s going to make you future-proof is your ability to understand your workflow, right? And figure out what portions of that are easily agentifiable.
I hear the word low-value work a lot, but I don’t know if that’s the correct framing, right? Because if it was low-value work, people probably wouldn’t be doing it. But there’s work that is easily agentifiable. So having a point of view on, okay, these are the things that I would like to take off my plate. So next time my organization rolls out a cloud co-work or some Microsoft version of Microsoft Copilot that actually works, I can shift over that to this and then move on to what is next in my pile.
So having a POV on what does me operating at the top of my license mean? And what are all the things that I would like to do? And what are all the agentifiable work that I have in my portfolio that I would like to slowly start transitioning off? I think that is a better mindset to have.
A second analogy I give often is, I feel like we’re in the Palm Pilot phase of AI. It’s not really iPhone yet. It’s also not really the flip phones anymore, right? So was Palm Pilot a necessary condition to move to iPhone? I don’t think so. Did people who use Palm Pilot actually become better iPhone users? Probably not, right?
So I think it’s more about your value prop is your understanding of the workflow and where the inefficiencies are. And the more you hone in on this because of this impending wave, I think the better you’re going to be in navigating and quote unquote future-proofing, like whatever the future might look like.
Predictions: Autonomous Agents and Future Workflows
What percent of work in an organization, this is a very broad question, do you believe could be agentified today, if that’s a word? And will be agentified in a couple of years from now?
I think in terms of workflows and not individual bodies of work, right? Because for something to actually drive the needle on top line, bottom line, you can’t just agentify pieces of it because then you still need, it might actually end up in more work, right? So what are unique pieces of workflow that end-to-end can be handled by some autonomous software? I think is how at least internally, we think about the value here.
Now, this usually comes with connotations of headcount reduction and let’s say people losing their jobs, but that’s 100% not the view that we are taking.
There was an example in IKEA. I think they rolled out this agent called Billy, which was named after their popular bookshelf, from what I understand, which was able to handle, I’m just making these numbers up, but it’s in the ballpark, like 57% of all the incoming queries right off the bat. So the customer support staff was not required anymore because the agent was so good that it was decidedly mitigating all of the customer intake requests.
Now, what IKEA did was they asked the question, okay, what is the remaining 43%? What kind of questions are people asking that Billy cannot answer? And turns out a lot of those were interior design questions, which had subjectivity built into it, matter of taste, and maybe agents aren’t good at it just yet.
So what they did was they redeployed their kind of 57 right back into a new revenue-generating stream. They created a new function for interior design and that became a multi-billion dollar revenue stream for them.
So that example, and this is obviously like all things AI happening over the past few months, almost transpiring in real time, I also want to give a nod to the fact that when we say workflows are automated, it doesn’t mean that people’s jobs go away. It means that there’s new revenue streams.
And that’s what I mean by operating at the top of your license. If everybody has a point of view on this is what me operating at the top of my license means, then when inevitably that agentifiable workstream gets removed off your plate, now you have more people working on things that are maybe more revenue-generating. And that’s how the enterprise of capitalism sustains. That is how I see AI and capitalism working hand in hand.
I agree. Doing the same with less is one version, or you do more with the same. And same meaning cost of labor or people at work. And so expanding the vision to be able to do more is what this ultimately allows us to do. But it does require retooling of a workforce, and not everyone’s going to learn graffiti. Not everybody’s going to embrace the tech as others do, and there will be some dislocation.
Business Case of AI: Top Line and Bottom Line
Right now, so many of the conversations we’re having, it feels like, let’s talk about how AI can impact and save us money as opposed to revenue growth. We’re on that pendulum if it’s a zero-sum game. In your last few roles, where are you seeing the emphasis, the attention? Top-line growth or bottom-line cost savings?
Build your business case on top line. Nothing beats top-line improvement, right? But especially in a regulated industry with a high risk profile like ours, the risk-reward paradigm does not justify putting something out there that’s going to jeopardize somebody’s health, right?
So start building that muscle internally. All of our use cases in our first pass were all internal efficiency focused. The framing we had, which is very deliberate, was what does the future of work look like for these key commercial personas and how do we build that using AI, right?
So there is a top-line play in the sense that by being able to get information faster, by being able to create content faster, maybe you make better decisions and that results in a top line, but it’s a top line by proxy. Whereas the fact that you’re getting decisions faster is efficiency play in the short term.
So start with that. Build a business case on top line. Build that credibility. Build that muscle. Build that confidence. I don’t think the agentic tech frameworks of today, especially for healthcare, are good enough to be scaled at mass with the general public.
So as the field evolves, keep a close eye on it, run some controlled experiments, but don’t go overboard on top line just yet. Because again, in a regulated industry, the risk-reward paradigm needs to be monitored very closely.
Hard Lessons: Competitive Versus Collaborative Games
Just to wrap up, obviously the podcast theme is hard lessons. So if you, given everything that you’ve known and experienced over your career, had to go back and talk to your younger self, what would be one or two hard lessons that you would want to convey to your younger self that really just changed the way that you think and operate today?
There’s this idea, like a game theory idea, of a competitive game and a collaborative game, right? I would say corporate America in general, regardless of what function you are in, is always a collaborative game.
So the more you can index on that, and there’s some game theory tactics of when do you collaborate and when do you not, I think that definitely has stayed with me as the winning combo. How do you frame your win in terms of somebody else’s win?
Especially for a function like AI in an enterprise where it’s not really the core product, you always need to be anchored on how is somebody else winning as a result. And to the extent that somebody can be a P&L holder, then you’re already on the right track. So I think that’s the biggest lesson I will communicate.
The second thing is, especially when you’re building a new capability, you need to start with a seed team that is rock solid, right? Because the rest of the culture, when you scale from five to 300, the culture is just a kind of extrapolation of what those five people bring to the mix.
So be very careful and stringent and don’t compromise on how you bring those five, six people together in the beginning. I think that would be another lesson that I would impart on myself.
I’ve already talked about this spotlight thing, right? Especially with areas like AI, it’s very easy. All eyes are on you. Just to be on the spotlight all the time, right? Be very stringent in what gets your face time versus your end user’s face time because chances are you’re occupying space where there was somebody else before.
And you need to over-index on collaboration and just being a good partner and being somebody who other people like to see win. So again, these are not AI-specific learnings. They’re just general corporate America learnings, but I think they apply equally to AI.
That’s it for this episode of Hard Lessons. If you enjoyed the conversation, follow the show on Spotify or Apple Podcasts and visit sifoundry.com for more on corporate innovation and emerging technology. Hard Lessons is brought to you by Silicon Foundry, trusted advisors to Fortune 500 companies.
Podcast: Play in new window | Download