Avoid "Burning Star Syndrome" using Smart AI Investment Strategies with Kunal Sawarkar

Speaker A: Welcome to the Lean AI Podcast, where we're flipping the AI conversation on its head by focusing on holistic strategies and tactics that drive AI adoption, rather than focusing solely on overcoming the technical challenges of AI. Uh, in every season two episode of the Lean AI Podcast, we talk with corporate AI leaders just like you, who've uncovered the secrets of driving successful adoption with far less wasted time and investment. Our guests challenge established views and offer disruptive perspectives, providing you with new, actionable insights. All right, Kunal, welcome to the show.

Speaker B: Thanks for having me, Ben. Excited to be here.

Speaker A: So today we're going to discuss your recommendations on how companies can successfully embed AI into their products. And you've written a great article called what CXOs need to do differently to get ROI, uh, from generative AI. We're going to include a link to that in the show notes. Um, and in our discussion we're going to be covering a few of your recommendations from that article, plus covering a few additional topics. But before we get into that, could you tell the audience a little bit about your background? You've had some really interesting experiences at a variety of companies.

Speaker B: Sure. So I started, believe it or not, as a Java programmer and was doing so for as long as uh, two months before I decided to quit it and uh, start doing something totally different that nobody heard of. So almost uh, two decades back, uh, there was not even a word called as data science or AI for that matter. It used to be a really uh, obscure blue uh, collar job like data mining where you go and sniff out some patterns in the data. And uh, that's what I started doing simply because it was challenging, it was fun and uh, I thought it is closer to my actual uh, formal background in electrical engineering. So been doing that uh, uh, since then. Always looking for more interesting problem to solve. Was part of Cognos aospace Labs, uh then went back to school and got my master's in statistics from Harvard and uh, then came back to industry to essentially uh, start running how AI products are built. So um, my first leadership uh, role was at ADP to set up their AI lab on how they can infuse AI into their products. And then I came back to IBM, uh to start its foray into embeddable AI. Uh so currently as a chief data scientist and distinguished engineer my job is to make other companies succeed on their AI mission so that they get a positive ROI and build the products that is going to future proof the companies. So that's what um, I'm currently at uh, and that's A short journey, love it.

Speaker A: And you left out some key experiences at companies like uh, AON and uh, T Mobile as well. Right?

Speaker B: Yeah. So those were uh, some of the early adopters uh, in what we used to call data mining. So T Mobile, uh, when I was there, uh, now we were part of a, in I was working in Germany and some company called Deutsche Telekom which had a largest data warehouse in the Europe. Uh, and uh, just figuring out a pattern of what customer want and uh, a very old school take was a lot of fun uh, before uh big data tools or anything that we had. So managing 200 million roads on every monthly data was a lot of fun. Uh, with different file systems and how to do make some of the algorithms run. Um, back in the day, uh, another fun thing was my grad project like some at that time was to teach machine how to write poems. Um, and uh, long before the word generative AI became so cool and um, anybody saw value in it.

Speaker A: That's great. And I think um, you know, just with, with such a amazing group, uh, of experiences all the way from you know, schooling all the way through to until now makes you really qualified to talk about some of these topics that we're going to cover today. So let's hop right in. You uh, recommend that CXOs, uh, dream big as you put it in your article. Can you expand on what you mean by that? Because I think everybody could maybe nod their head and say yeah we should dream big. But what does that actually mean? And maybe how is it favorable to a more grassroots approach where everybody just you know, tries AI at the company and finds out what they like.

Speaker B: So the idea about AI, uh is that it is supposed to revolutionize the thing that you are doing. Uh, your processes, your workflows or basically running your operations. Now that requires a very clear vision that some, somebody at the top has to set up. Now when we say dream big, the idea is do not try to replicate what is already happening by putting AI into it as a side hustle. So if you want to be successful in AI, you have to be AI native or AI first company which is a different mindset than let me put AI here and there and see some incremental gates. I, I mean the example that I love is that of when first iPhone came. It was a transformational move in the way phones were uh, before that. And look at the first level apps that were there in first two years and you will find that none of those app actually were mobile first or smartphone native. The best example is that of a radio taxi. There was an app on which you can go and actually call the taxi. Now that is the idea where you are uh, replicating an existing workflow. Why do you need to call it when you already have a GPS on that? So if you see the picture that I always show in my lecture is there was a Maps application, Google Maps and there was a phone, a screen. It requires somebody to think why don't I uh, connect the two and build an Uber. So what I see a lot currently is that people are looking for ways to automate their existing workflow and that is not going to give the 10x return that every CX want when they make Genia investment. So you have to dream big if you want to see the return. And uh, sometimes it is an imperative because the cost of building AI product is so high. Unlike you doing web services or website building which is a very low cost. The cost to build, infer and manage the team for AI is extremely high. So you can never see the value out of it unless you actually think big on how I'm going to completely tear down the existing structure and look for not just uh, uh, like efficiency gain here and there, but rather think like uh, how can I simplify as uh, you know Elon like to say the best part is no part can you get rid of something altogether with the help of AI. So that's the idea of dream big and uh, have a long term vision of how I can succeed in what I want to build with.

Speaker A: Jnai Mhm. I like that and I think it's interesting because I think that's, that's, that's good advice. Also in your article you recommend that uh, people forget about AI and focus on products. And so those could seem contradictory but they aren't. Could you maybe help the audience understand what you mean when you say you know, forget about AI and focus on products but also you know, dream big when it comes to AI and be AI native?

Speaker B: Yeah, absolutely. As in uh, the idea that everybody's currently running today is I need to do something with AI and it's like yeah, but as a customer do I really care? I'm walking in a uh, let's say you're walking in a retail store buying a shampoo and it says built with AI. Do you really care like whether it is built with AI or not? Just because you put uh, like AI name on the top of everything that doesn't change the customer's life. So you have to be obsessed um, with solving the customer's problem. And not really obsessed with using AI. I uh, always say oh, let's use chain AI. Now um, sometimes the problem can be solved in traditional algorithm which will run at uh, 1% of the cost. So forget even if you need genai or AI, are you obsessing in solving something that everybody struggles with? And uh, that is the challenge that I think most companies have. They change their name from dot com to AI overnight. But that doesn't mean that company's entire uh, thought process is going to change. And you uh, need to think is like how can I build a product that changed somebody's lives uh, without actually using AI? In fact I always say the companies that actually make largest use of AI don't even have it in their name. Like do you go to Amazon because it uses AI? No, because it gives a great experience or Apple for that matter. So you don't have to advertise it as long as you build the brace product. And uh, that may seem very obvious when you think about it, but that's not how most the decisions are made. When the pitches are currently circulating in CXOs, they always want to hear am I doing something with the AI or am I missing the bus? The goal should be am I giving some kind of a 50% improvement in the current workflow? Am I actually simplifying the time to do achieve something for my customer? Am I reducing the cost by 20, 30% to what it is now? Those are the KPIs customer and in enterprises care about B2B, B2C or uh, whatever organization you are in and not if you have actually adopted AI as a like organization.

Speaker A: Yeah, I love that perspective. And you know we see that at the lean startup company all the time. Uh, people that just want to do AI for AI's sake and don't just take a basic product mindset, starting with customer problems, you know, and going out and experimenting on anything that's not aligned to a compelling vision, for example. So um, those are, those are solid points. I guess the other thing is, you know, embedded in your uh, in your writings is the idea of solving customer problems. And that customer doesn't need to be an external customer, could also be an internal customer. So same thing for your employees or your staff. Um, they don't care if it has AI either. They just want it to work better. Than what, than, than the current workflow.

Speaker B: Yeah, absolutely. As in you think of uh, like people trying to do AI for everybody else but not using it themselves. So yeah, I think uh, and um, that reminds me m of somebody I dated a long time back when I was young. She was a physiotherapist and should always give advice on exercises to other but she would never do it herself. She's very happy. So like how is the life of your own engineers is being changed by adoption of AI is a uh big factor. I'll give one simple example is like uh, everybody who works with a client has some CRM system and uh, you go into the meeting by going through data from the CRM system like you find out how big is the deal, what happened, everything and that is like 20 screens I need to navigate now why not use Genai to come up with the summary of all the history that is with the customer. Or uh, as an engineer you need to provision something and believe it or not AI makes a big use of gpu but GPU provisioning is still very um, like old fashioned. Now nobody has fully created a uh code gen workflow for GPU which is what I tried in my team or how you build the pipeline. So the best ideas on how to uh, replicate uh, how to avoid uh replication and get the boost comes from your own team on what makes their day difficult. So yeah my leader here used to say tell us one thing that frustrated you this week and that's the start of all the good ideas that you can solve with the gen AI and uh, if you are not using it you are actually not also seeing the challenges with the AI. Ah most people think that you build a solution and uh, it's going to uh, uh work the way you are uh expecting it to work and to be used by end user. The whole idea of a gen AI is that ask me anything. So we always found that customers use it in a way that you never even imagined simply because you made just ask me anything. So they will try to break it. And uh, that is the big challenge for enterprises because unlike ChatGPT which is uh actually very uh trained on an omnipresent data and works well in that context in enterprises it is normally not that uh omni uh channel experience that you, it is designed for. And the best way to test your products is then again with your own employees they will help you break it. So tell them just break my m solution.

Speaker A: I love it. I think it's also interesting implicit in what you just said is you're not asking customers, you know in this case it was internal or even external. You're not asking customers hey what should we build with AI? Because there's the Steve Jobs quote, right? Like it's not the customer's job to know what they want. Um, it's, it's our job as entrepreneurs to understand the problems the customers have, including the internal ones. You know, what frustrated you this week. And then, uh, the entrepreneur, the product team. It's our job to say, how might we solve that, um, given the technology that we currently have available. And that's especially true in the space of AI where hardly anyone, including myself, knows what's possible. I mean it changes on a weekly basis. Right. And so, um, I think it's really important for, you know, leaders out there to not focus on asking customers what they want, but to go out, find problems that are really juicy, that are worth solving and then say, how might we solve that?

Speaker B: Yeah. And just to add to that, there are two famous quotes on that as well. One is all the way from Henry Ford. He said, if I would have asked people what they want, they would have said faster horse, uh, carriages. So don't ask them. But there is also bigger uh, perspective for CXS to keep in mind is what Jeff Bezos has said is like focus on the problem, which would still be relevant years from now. So he said like no one is going to wake up and said, I want to spend uh, two hours shopping. I want, everybody wants lower and lower time on shopping and everybody wants to have lower and lower prices when they shop. Those are the fundamental problem if you think about the retail as a business. And those are the problem if you are obsessed about, you'll be in the business for decades. Uh, so it's the problems that matter. And you're absolutely right. People normally ask what can we do with aas? Forget it. Just tell us your most, uh, uh, frustrating problems and then figure out what you can do about it.

Speaker C: Hi, this is Jonathan Burfield, senior director at the Lee Startup Company. If you're a corporate executive looking to drive broad adoption of the AI centric products you're developing with far less wasted time and investment. We invite you to join us for a free 45 minute one on one consultation where we'll help you understand key tactics of validating use cases early in the development journey to identify the optimal sequence for rapidly driving to scale and to navigate the potholes that have tripped up other levers in similar roles. Head over to leanstarter Co. Contact AI to reserve your spot. You'll find this link in the show notes, but don't wait. Spaces fill up fast and we don't want you to miss out again. That's LeanStartup Co contact AI. Let's make successful incubation and scaling of AI centric products a reality for you, your team today.

Speaker A: Yeah, yeah. I mean the best innovation doesn't happen when you ask people what they want and then you build that. Um, and to, to, to use the uh, the Henry Ford example and there's always debate about whether Henry Ford actually said that or not. I don't think it matters because the quote is, what's interesting is um, you know, why are they asking for that? So don't take it at face value but why are they asking for that? Well because it's really a pain to get to work or to get around and takes forever and your horse gets sick and it's unreliable and there's, you know, you need to feed the horse and there's manure and there's all these things. How could we make it simpler? Right? And so um, it's something that we're always encouraging our clients at Lean Startup Company to, to look at is if we're really, if we really have a tight problem statement defined that's a problem even if our company doesn't exist. Right. Even if the, even in, in the absence of any solution, that problem still exists for, for years to come. So um, I love, I love the perspective that you shared on that. Let's jump into funding because I think that's something that you've really got some strongly held opinions on uh, that are highly beneficial. So uh, in your article you wrote about Burning Star syndrome for Capex and I love that you named it that because I remember it now because it's a very unique way of, of phrasing it. Can you help the audience understand a little bit about what you mean by Burning Star Synd Capex?

Speaker B: Let us start with something that everybody knows which is that AI is very expensive. So like look at the numbers for 2025 projected by all the big tech. It has hundred billion for I think Amazon 75 to 80 for Google and Microsoft. I think another hundred billion for Apple. I just counted on max 7 and it's almost 600 billion in capex for them. Now then that is just that and forget all the other players like Salesforce and IBMs and oracles of the world and uh, all this money like let's say 1 trillion in capex spending is going to be absorbed by somebody who is building applications on the top of AI. Now that uh, cost that somebody as a startup is absorbing, they need to recoup that somewhere. So I always tried this uh, uh, analogy to tell to startups is ChatGPT or Amazon for that matter is not in the business of AI, they are more in the business of compute. They are selling you the compute but your customer is actually not going to care about that. He's only care about the value you are adding to their life by AI. So you have to navigate carefully in how much your capex cycle is going and how fast you are ah burning it out. Because if you are trying to build a really big model and burn out a lot of cash too quickly then you will not be able to see the ROI from genie. Ann, everybody who is in enterprises know that it requires couple of years before you start seeing the money flowing down in the positive ROI sense. Uh if you want to survive till that point you need to reject how much money you can spend on training and inference. And I um, saw so many startups which had great ideas but they never reached a point where they had a stable model that they can commercialize. Uh, so the challenge that startups face is exactly the startup challenge that big companies also face which is that everybody is uh, spending so much in building AI. But uh, then where is the ROI coming out of it? I think the great example is uh, last week's result from Salesforce where they said that agent force isn't going to add to revenue anytime in 2025. Now that should be an eye opener that it takes longer time for your AI investment to first see the light of the day for product being actually going into production and getting into customers hand and then being adopted and then generating roi. So I call this Burning star syndrome because uh, if you think about uh, like what kind of a star harbors life then it is not the biggest and the brightest because they burn out too soon and either become a black hole or um, turn and burst out in a supernova. And you will find many such companies somewhere in AI as well. At the same time you can't be a uh, brown dwarf which is actually not burning any fuel and doesn't actually attract any kind of a critical mass for it. So it is somewhere in between like our sand which is runs dimly for long period of time and good enough to harbor life and sustain the ecosystem that requires to be successful. And that's the same thought process you need to have is how I want to build AI. I want to consume a lot of gen AI but how do I plan it out for next one to two years and uh, there is such amount of what in executives, uh, a FOMO of genai or I normally call it fud. Like uh, fear, uncertainty and doubt about uh, I don't want to miss the buzz that they are jumping into the use cases uh, whose uh, actual value conversion doesn't match the investment being made there.

Speaker A: Yeah, so, so there is such a thing as investing too much ah in AI too quickly uh, because then it's going to implode under its own weight. But if you don't invest enough then you're never going to, you're never going to harbor life to use your, to use your metaphor. And so it's somewhere in the middle. How do you uh. And then another thing that you, you'd mentioned in your article is um, uh, basically having that compelling vision, that grand vision, but also uh, getting really specific on executing uh, maybe on a weekly basis. Can you talk about that a little bit please?

Speaker B: Yeah, absolutely. So uh, any AI which has actually been successful has been built with a clear vision that lasted years and something like OpenAI had a decade long vision of how they want to get there. So you are not going to build uh, a roam in a day, but at the same time you have to make sure that you are actually going towards that path path. And uh, that has to be seen on weekly uh, execution of how you are seeing the continuous improvement in your system. So we all say that AI is here to improve life. But uh, if you are not seeing any improvement on let's say week on week basis or month on month with the product that you are shipping onto the underlying workflow, then how are you ensure that one year down the line this is to going, going to work. So uh, you have to measure whatever is your KPI. Let's say you are building an agent which is supposed to simplify our business workflow that does say everything in uh, CRM or everything in financial force. But uh, if you don't measure it by shipping it regularly to your users, even if it is an internal user, then you will find that you built a product that is either not viable or not valuable. And uh, there is another key aspect to it which I normally uh, think most CxOs underestimate because when they work on the AI product they treat it like other products. So they come up with an idea, they go and start building it. The big challenge with AI, unlike others, is that your idea may actually just not be possible with the current technology. I come up with so many uh, sessions. Where is it doable? It may be doable if you are an OpenAI and want to invest in next AGI ad, but can you actually bring a model off the shelf that is currently available and put it in production without doing an R and D to execute that idea. So is it doable? And that is a very important part to build a successful product is you need to pivot, uh, always regularly to the correct direction in building that product by mapping is it valuable, is it viable and is it doable? So whatever is your grand vision, the weekly execution results gives you that signal of what is possible and aim actually building something that will be relevant.

Speaker A: I love that. I think, you know, you're talking about in uh, maybe some phrase slightly differently, you know, desirability. Uh, is this solving a real problem and is it uh, creating value by solving that problem? Uh, two different components of desirability and then feasibility can it be built? And viability, should it be built? And typically, uh, with new product development or innovation or tech commercialization, companies tend to way over index on feasibility upfront. So they'll spend just so much time on feasibility and then, you know, two years later they'll actually test a first version of a product, even if it's just a marketing mvp, you know, a uh, representation of brochure in effect, or a landing page, and find out they wasted all their time testing the feasibility of something that nobody cared about in the first place. However, notable exception in the lean AI space, we actually do recommend more feasibility upfront, not too much because again, it's maybe comfortable and you don't have to go out and find out that your baby's ugly by running experiments. Um, but certainly a little bit more upfront. Can it be built in the first place? Because just to your point, we've seen that so many times where companies will get way down the line and then figure out they can't even build the thing in the first place. So I think that's a key element of a lean AI approach. Um, let's, let's talk about the back to the funding and the uh, the kind of, the continuous delivery. One of the things we found, whether it's the AI space or corporate innovation or whatever, is, hey, we're working on something over here as the, you know, as the AI team and keep giving us a lot of money. Maybe we're that star that's too bright where we're burning really hot, right? And then just trust us that five years from now we're going to come up with something that's great. And in the modern world of focus on short term, you know, quarterly results, return on invested capital, even if you've got a senior executive that says, yep, we Understand it's going to take five years. You know, we're going to, we're going to protect you. It rarely happens that way because the leader moves on or there's a stock price event or something happens and then you know there's a knock on the door and someone says hey, we've been giving you a lot of uh, a lot of money. Where's the return? We need it now. And so um, I think establishing a path of quick wins to medium term wins to long term wins is implicit in what you're talking about. Rather than just saying hey trust us, everything's going to be great eventually but we're not going to ship anything, we're not going to demonstrate any proof points between now and five years from now.

Speaker B: That gets absolutely critical. In fact I normally used to call myself an in house entrepreneur because I think that's the hat that anybody working in large MNCs has to put on themselves if they want to be successful in AI. What do you do as a startup? You build something, you show it to investor and you get more funding that hey, I'm delivering some value. Now that is the same attitude big enterprises has to adopt when they are building AI. Uh, uh, because if you are not showing the value then it becomes like a rabbit hole. In fact many executives always, uh, before JNAI were shy of putting money in AI because they were like oh it's a rabbit hole, I don't know if I have when I will see something out of it. And that to a large degree I would say is the failure of uh, AI leader or the data scientist themselves. In fact the joke is that a uh, data scientist is a person who is very good at solving wrong problems. So it all boils down to did you pick up a right problem? The problem as I said, which is uh, like uh, feasible, uh desirable and viable. I mean is the problem still relevant two years down the line? I mean if you are building a wrapper on ChatGPT, those are all gone. And that is if that is where you're building last year. All that code is now in the drain because with the agent all that thing is gone. So you need to have a perspective. Clearly did I find the problem that my end customer will pay top dollars for? Uh, I always call this cost cutting versus growth. If I build this product is that is going to drive my growth or I'm just trying to cut the cost somewhere here and there because it is very intuitive to do with the AI. Oh I can like you know uh, replace uh, that HR from Four to two people. Okay, that's great but is there somebody directly paying like a top dollars for it? So think in that way is it going to get me new customer? Is it going to increase the engagement of current customer their satisfaction store something that directly affects your top line growth, growth and if that doesn't happen, if you can't relate to that then and show that happening consistently over a quarter on quarter basis then it will never fit. Like two years down the line you bring something and it may not actually um, gel at all with what customers were looking for.

Speaker A: Yeah, so I love this and I mean uh, the way to express that type of funding uh, in leap in lean startup parlance is metered funding. Right. That's why startups don't get all the money up front. They get a small, a small amount, very small amount, precede and then seed and then series A, series B. And it grows as they demonstrate that there's a, ah, there, there. And so just to tack that onto the Burning Star analogy. So you've got to get your overall funding right but then the way that that capital is deployed is really important because you don't want to go all in on a few things and really hope you get it right. You want to take multiple smaller bets and use the data to determine what problems are worth solving and where are we really adding value. And that way you've got evidence as opposed to just conviction that you're investing in the right things in your portfolio. So we're almost, we're getting close to being out of time here and there's one more topic that I wanted to make sure that we covered and that is uh, you mentioned that companies should avoid the it ification of AI. And this is really important in terms of the talent, the teams that you have working on your AI products. Maybe expand on that a little bit if you could.

Speaker B: Yeah, I think there are two uh, aspects to it. Uh, so when I say that AI is not it then uh, it is like uh, you come up with a problem and that uh, you go and put some team around it and you expect like the problem will be solved. Now there is a big uh, unknown in AI which is that do you really have the deeply skilled people who can actually solve this? Um, there has been this idea that you can um, do prompt engineering. So anybody who knows how to write English is now an AI engineer. Now I willing to stick my hat out and say that for any enterprise who wants to build real AI product, you still need uh, expertise, ah, you still need people who knows how the data science work. And for that you cannot just give them on two weeks training and expect that they deliver the value and build great products for you. So make sure that um, this is the part where you invest and get right team in place firsthand. And uh, this is one challenge that people realize often when it is too late. Um, so AI is a game of quality. It is not the game where I built a website and you have a website and we are both okay. Uh, it is like you uh, want something which is really good in terms of accuracy. So even if your team can take it from 0% accuracy where you are probably today to 50% it is not good enough uh, for to generate any value for you because 50% accuracy is still crap as far as end customer is concerned. And do uh, you need to get to 80% often 90% and as you go closer and closer it's like the speed of the light problem. You need you, you can accelerate something to 10% to the speed of light in your small lab, but to go to 99% you need large heteron collider. And that is exactly the analogy that works in a complex AI uh project. So if you want 80% accurate result, make sure that you have that level of talent who knows how to do it or who has done it somewhere in the past. At least I Normally call it 1/3 of your team, just 1/3 if not 100% who are deep experts and not starting, not just uh, like um, switch over from it to AI yesterday because now it is the cool and thing to do that pays a lot.

Speaker A: So Kunal, we're right at the end of our time together today. Uh, as you know this, this uh podcast is built for corporate AI executives. Uh, what three pieces of insight would you share with them even if we've already talked uh, about them. Uh, that would really help them to be better at their job.

Speaker B: So uh, here is what I would say if anybody wants to successful in Genai. First uh, always dream big but deliver in weeks and uh, dream something that you are actually simplifying the life of your end user. So simplify, simplify, simplify, that cuts the entropy from your end user than just doing air for air sake. Um, the second part is manage your funding really really carefully. Um, make sure you are building something uh, that uh, has a clear vision of how it will be adopted and uh, you have capex figured out till then. The third and the most important part which is often overlooked, uh, is what you need to build AI other than GPUs and the server is a great team. So make sure that you build an culture in AI for your AI success. A culture and a mindset which is about problem solving, which is about pushing the envelope and where you can quickly adopt new, uh, techniques without having to over index and rerunning your pipeline. So having that culture built ground up with the right set of people is what I consider will be a secret sauce that differentiate highly successful companies from the companies who are trying to do AI.

Speaker A: Well, that's great advice and really appreciate you spending some time with us today. And again, we'll provide, uh, a link to your article that we've been referring to in the show notes. So thank you so much for your time and hope to speak with you soon.

Speaker B: Thanks a lot Ben. This was a lot of fun and uh, thanks for having me the Lean

Speaker A: AI Podcast is brought to you by the Lean Startup Company. To find out more about our Lean AI approach to generating more AI wins with far less wasted time and investment, including our training workshops, pilot programs and full implementation offerings, Visit us at leanstartup.co Search the Lean AI podcast in Apple Podcasts, Spotify, or wherever you get at your podcasts so you don't miss a future episode. On behalf of our entire team here at the Lean Startup Company, thank you for listening and sharing with your colleagues and friends. If you found this episode insightful.