Building the AI-Native Enterprise: Leadership, Software, and Compounding with Mahesh Rajasekharan – EP.286

(00:00:00) – Intro

(00:00:57) – Mahesh’s background and Cleo’s mission

(00:03:20) – What AI-first actually means

(00:07:21) – How AI transforms software development

(00:11:01) – Enterprise software’s shift to systems of action

(00:13:41) – What executives misunderstand about AI

(00:16:12) – From software features to business outcomes

(00:19:10) – Supply chain orchestration as a category

(00:24:31) – AI economics and smart capital allocation

(00:29:55) – Leadership principles in the AI era

(00:32:44) – Capital allocation across growth initiatives

(00:37:09) – Building for the next decade

(00:41:13) – Final thoughts on CEO leadership in AI

Mahesh Rajasekharan: AI is such a fundamental shift. It’s probably a 100-year shift. And so I, I think having an incredible sense, sense of urgency is the most important thing because you’re gonna learn. At the same time, you need to have a North Star, and at the same time, you should lead from the front as CEOs. I think that’s where sometimes people, I think, make mistakes when they delegate to a chief AI officer.

Mahesh Rajasekharan: Except I’m not saying there’s anything wrong with having a chief AI officer. The CEO should learn what problem they’re solving and see what AI can do for it versus delegation.

Alex Bridgeman: Mahesh, it’s great to see you again. Thanks for coming on Think Like an Owner for a second episode. This podcast is all about how small companies are grown by ambitious owners, and you’ve grown quite a company in Cleo, and I’m excited to talk about kind of how you see software companies growing into the future and through AI tools and, you know, becoming a more AI-first organization.

Alex Bridgeman: It’s certainly relevant for what Rob and I are thinking about too. But for those who don’t know, can you tell us the story of yourself and Cleo and kinda what the business does, a- and then we can get into your, your views on AI for software.

Mahesh Rajasekharan: Sure, Alex. I’ve spent more than twenty-five years in and around supply chain across strategy, planning, execution, and digital transformation.

Mahesh Rajasekharan: And I’ve had the privilege of leading Cleo for a significant part of the journey. Cleo today is a category leader in AI-driven supply chain orchestration. We serve more than five thousand customers globally across industries including manufacturing, logistics, wholesale distribution, and retail. At its core, what we do is actually pretty simple to describe.

Mahesh Rajasekharan: We help companies connect and orchestrate the flow of orders, shipments, invoices, inventory, and payments across their business ecosystems between themselves and their customers, suppliers, and logistics partners. And the reason I find this space so fascinating is that supply chain has evolved from what was historically viewed as an operational function into a major driver of business performance: revenue, margin, working capital, and ultimately customer experience.

Mahesh Rajasekharan: But there is still a fundamental gap in enterprise software, which is planning systems understand what a business wants to happen, but they generally cannot execute. Execution systems, on the other hand, are very good at processing transactions, but they often don’t have the business context to determine the best action when reality deviates f-from the plan.

Mahesh Rajasekharan: So that’s the gap we’re focused at closing at Cleo, bringing together integration, context, intelligence, and coordinated execution. And that’s what we mean by AI-native supply chain orchestration. And with AI, I think the opportunity to transform how these business ecosystems operate is greater today than at any point in my career.

Mahesh Rajasekharan: And that’s a big part of what is gotten me so excited about the next chapter for Cleo.

Alex Bridgeman: That’s a great background. Thank you for sharing. The kind of, I think, crux of the subject you’re excited about and have thought deeply about is how, uh, how do we build better software businesses with AI, and a lot of companies describe being AI first as a key value of how they’re building their businesses.

Alex Bridgeman: But what does that actually mean, and maybe what does it mean to you?

Mahesh Rajasekharan: I think w- we have to be careful, Alex, about the term AI first because today, as you know, almost every software company describes themselves that way. For me, being AI first isn’t about putting a chatbot into your product or adding AI to every feature.

Mahesh Rajasekharan: It’s much, much more fundamental. It changes how we build software, what you choose to build, how we operate the company, and ultimately how we produce outcomes that can create value for customers. So I’m incredibly bullish about AI. I think it’s one of the most consequential technology developments of our careers.

Mahesh Rajasekharan: But I’m not particularly bullish about AI for AI’s sake. The ability to access incredibly powerful intelligence is becoming ubiquitous today. The frontier models are getting better at an extraordinary rate, so simply having access to AI is not gonna be a durable competitive advantage. The question I always keep coming back to is: what exactly are you amplifying?

Mahesh Rajasekharan: Because that’s what I believe AI fundamentally is. It’s a tremendous force multiplier. So if you have a great product, if you have proprietary data, if you have deep domain expertise, rich context to trusted customer relationships and mission-critical workflows, then AI can dramatically amplify the value of all those assets And conversely, if the underlying value proposition is weak, simply adding AI isn’t necessarily gonna create a great company.

Mahesh Rajasekharan: And so I do think there will be significant winners and losers in software. If your product essentially automates a relatively simple workflow, then increasingly capable models may be able to reproduce a lot of the value. But you think about a complex enterprise environment like supply chain, like, like I just described.

Mahesh Rajasekharan: So in, in that case, you need to understand the industry. You need appropriately operational data. You need to understand the context surrounding the supply chain transaction. You need complex workflows across the multi-enterprise ecosystem of multiple companies in a supply chain. You need governance, you need security, and ultimately you need the ability to take, take action.

Mahesh Rajasekharan: So that’s a very different problem. So I think something interesting happens as underlying frontier models commoditize. So then context becomes more valuable, not less, and that’s particularly relevant to Cleo. So we, as you know, spent years building deep supplies and expertise and connecting companies with the trading partners, with the flow of orders, shipments, invoices, payments, and all of the interaction around them.

Mahesh Rajasekharan: So in our case, our opportunities in simply to put AI on top of Cleo is to combine powerful intelligence with all of that accumulated data, context, domain expertise, workflow, and execution capability. And so ultimately, there has to be, in my mind, a very simple scoreboard for companies. And the scoreboard is, did we create substantially more economic value for, for the customer?

Mahesh Rajasekharan: And I don’t particularly care how many AI features we launch. I don’t care how many agents we can put on a slide. The question is, did we help a customer protect revenue, improve margins, reduce working capital, serve the customers better, and ultimately grow faster? And that’s the gold standard. When I say AI first, I don’t mean AI everywhere, right?

Mahesh Rajasekharan: I mean u-using AI intelligently throughout the company to build better products, operate better, and most importantly, deliver dramatically better outcomes for customers

Rob Southern: Thanks, Mash. And, and by the way, I just wanted to call out, it’s great to talk to you as a podcast host now. We’re used to seeing each other in the boardroom.

Rob Southern: You were on my board at Send for six and a half years. I’m incredibly grateful for that. And, uh, yeah, good to be here with you again in this context. Related to that question, how has AI fundamentally changed the way Clio builds and deploys software?

Mahesh Rajasekharan: Yeah, I’ll– Rob, I would go further than just saying AI has changed software development or, you know, basically how we build software.

Mahesh Rajasekharan: I think the act of producing software has been turned on its head. Think about the traditional model, right? In the traditional model, humans wrote the software, and all the tools helped the humans building the software, right? Now we’re moving towards a world where humans are increasingly defining the problem, the architecture, the intent, the standards, and the desired outcome.

Mahesh Rajasekharan: And then the fleet of agents, AI agents, can participate in actually producing the software. That was the profound change. We’re already seeing it throughout Cleo. We’re using the highest quality frontier models we can access for coding and testing agents and other parts of the product development life cycle, what we call the, the software development or product development life cycle.

Mahesh Rajasekharan: And the pro-productive improvements we’re seeing is just dramatic. It’s extraordinary. Things that historically required significant engineering time can happen much, much faster. Ideas can move from concept to prototype to testing at a speed that would, would have been unimaginable just a few years ago.

Mahesh Rajasekharan: But also, there’s an important misconception here, which is generating code is not the same thing as building great software. In fact, once code becomes easy to generate, a whole different set of problems becomes more important. If fleet of agents are producing code, how do you know they’re producing the right code?

Mahesh Rajasekharan: How do you validate it? How do you test it? How do you know it fits your architecture? How do you secure it? How do you govern the agents? How do you make sure the resulting product is reliable at enterprise scale? So that’s why at Cleo we built something internally called the Cleo Code Generation Harness, which we’re actively using, and we are effectively building the infrastructure around our coding agents to make sure that as we dramatically accelerate the creation of software, we maintain and, and hopefully improve the, the whole quality of what we’re producing.

Mahesh Rajasekharan: So, so to me, that’s one revolution, which is we are redesigning the software factory itself. But I actually think there’s a second revolution that’s equally interesting and even more interesting, and that’s what you choose to build because the ability to create an AI capability has now become remarkably ubiquitous, right?

Mahesh Rajasekharan: So you can build things today in days or weeks that would’ve taken months or years, or simply would’ve been– wouldn’t have been economically feasible before. And so paradoxically, the scarce capability is now becoming less about can we build this and more about should we build this? Where can AI truly create differentiated customer value?

Mahesh Rajasekharan: And that’s a product management and a domain expertise question. And so in our case, I really don’t want the Cleo team, uh, forcing AI into every feature just so that it can call everything AI power. That’s really not our objective We want our product leaders and engineers, industry experts, and our customers, in many cases we work with customers directly, to identify the handful of places where AI-driven intelligence can fundamentally change an economic outcome.

Mahesh Rajasekharan: And then I want us applying extraordinary technology against those opportunities. So I actually think that means judgment becomes more valuable in the AI era, not less. When everything becomes easy to build, knowing what to build is, and what is worth building, becomes incredibly important. And that’s one of the things that excites me most.

Mahesh Rajasekharan: The best software companies aren’t simply gonna build the same products faster. They’re gonna build better products that weren’t previously possible

Alex Bridgeman: And one thing we’ve heard, Mahesh, you say often is enterprise software is entering, entering a new era. What do you mean by that?

Mahesh Rajasekharan: You know, I think we’re really moving from systems of record to systems of action.

Mahesh Rajasekharan: And, you know, for decades, for decades, enterprise software primarily recorded what happened, right? Very much about after the fact systems of record. Then we added analytics. We added dashboards. We added applications on top. We really added increasingly sophisticated systems of intelligence that helped humans understand what happened and make decisions.

Mahesh Rajasekharan: The key is humans understood and made decisions, right? But the operating model remained fundamentally human driven. A person logged into the software. They looked at information. They interpreted the, the, the dashboard or the, the information. They made a decision. They clicked something. They initiated action.

Mahesh Rajasekharan: Okay? Now, if you think about in the AI world, AI fundamentally changes that model, right? Increasingly, agents will participate directly in those decision-making loops. They will discover information, understand the context, evaluate alternatives, make recommendations, and with appropriate controls, and this is important, you know, control and governance and hardness becomes important.

Mahesh Rajasekharan: With the right appropriate controls, the agents can increasingly take action. So I think we’re really going through a paradigm shift. We’re moving from what I would call human-led technology-assisted work towards human supervised technology-driven operations. And that doesn’t mean humans disappear. It’s very important.

Mahesh Rajasekharan: Humans move up the stack. They focus on judgment, strategy, innovation, relationships, and ultimately accountability. And that changes how enterprise software itself needs to be redesigned, right? Historically, we optimized software for a human sitting in front of a user interface. That’s how UX is designed, right?

Mahesh Rajasekharan: Increasingly, and we’re doing it at Cleo every day, which is we are looking at software that needs to be agent friendly. It has to expose trusted data, business context, permissions, governance, and secure ways of taking action. And that’s also why I don’t believe the simplistic argument that AI makes enterprise software irrelevant.

Mahesh Rajasekharan: Actually, AI provides extraordinary intelligence, but that intelligence alone isn’t enough. You still need context. You need trusted data. You need deep understanding of workflows. You need controls. You need security. You need domain expertise. You, you need the ability to maintain and the ability to extend, and ultimately, you need the ability to execute.

Mahesh Rajasekharan: And that’s where I think this next generation of enterprise software really gets interesting in the AI age

Rob Southern: M-what do most executives misunderstand about AI today?

Mahesh Rajasekharan: I think there are several things executives misunderstand. You know, the first is really confusing the ability to generate code with the ability to build great software.

Mahesh Rajasekharan: Touched upon that. AI is making code and coding dramatically cheaper to produce. That’s an extraordinary development, no question. But enterprises and customers don’t buy code, right? They buy outcomes. They want reliability, they want security, they want governance, they want accountability, and most importantly, they, they want trust.

Mahesh Rajasekharan: So those responsibilities just don’t disappear because an agent wrote the code. That’s, that’s most important. If anything, they actually become more important. The second misunderstanding is believing that because you can put AI into everything, you should. I think that’s gonna create a lot of wasted investment.

Mahesh Rajasekharan: So the question is– the question is not about where can you use AI, because the answer for that is almost everywhere you’re gonna use AI. The better question is, where can AI produce a materially better outcome than what you’re doing today? And that’s a much harder question, right? And there’s a third misunderstanding, which is the AI conversation is often framed almost entirely, and, and I would, I would say this based on all the discussions I’m having, the AI conversation almost always is framed around replacing human beings, replacing people.

Mahesh Rajasekharan: Of course, jobs and roles are gonna change, but every major tech-technology transition has done that, right? This is not new if you look back at the history of technology. I think the opportunity that I’m most excited about is something really different, which is now AI is gonna replace friction on people.

Mahesh Rajasekharan: Think about how much time really talented people spend searching for information, reconciling two systems, investigating exception between ERP system and the CRM system, for example, following up with someone else, manually moving information between applications. That isn’t the highest use of the extraordinary human capability.

Mahesh Rajasekharan: So I think if AI can remove enormous amount of that friction, then people can operate at a much higher level using the judgment, creativity, strategy, innovation, and customer relationships. So I find that future incredibly exciting. But getting that requires real leadership, right? You just cannot delegate an AI transformation to an AI team and sit in the corner.

Mahesh Rajasekharan: The CEO leadership team have to be deeply engaged because you’re redesigning how the entire company actually works going forward

Alex Bridgeman: You’ve talked about software features to business outcomes and moving in that direction towards outcome. Can you explain how you mean and what do you mean by that?

Mahesh Rajasekharan: Yeah, this is something I believe for a long time, and I think really AI makes it even more important, right?

Mahesh Rajasekharan: Which is the ability to focus on business outcomes. Customers don’t wake up wanting software features. I, I go and, you know, really sit down with a lot of the, the C-level executives in supply chain and, and information technology all the time, and almost all our con-conversations are not about software features.

Mahesh Rajasekharan: They have businesses to run. These executives want to grow revenue. They want to improve margins. They want to reduce working capital tied up in inventory. They want to serve their customers better. They want to enter new markets faster. They want to operate more reliably. So our core philosophy at Cleo is start with the economic outcome and engineer backwards.

Mahesh Rajasekharan: And AI makes that discipline critical because today it’s remarkably easy to build something that looks impressive, right? You can create a fantastic AI demo in an afternoon. You’ve got young product managers who can just wipe code and, and show amazing things. It’s helpful because you know what it looks like.

Mahesh Rajasekharan: But there’s an enormous difference between an impressive demo and a demo, a wipe code of software, and a production system that creates true economic value every single day. So let’s take a simple example. Let’s take a late shipment, right? There’s an order from a retailer, and you ship late, or the, the shipment is going late.

Mahesh Rajasekharan: Telling somebody the, the shipment is late is visibility. It’s useful, but it’s limited, right? You’re just, you know, you’re just sharing information. But now suppose the software understands that shipment in context. Which order is it connected to? For which customer? You know, is it a large retailer or a small retailer?

Mahesh Rajasekharan: What commitment did we make? How much revenue is potentially at risk? And what inventory alternatives exist? Can I ship it from a, a different warehouse? Can I redirect something which is due later to a customer who needs it right now? What are the downstream consequences? So now we’re actually moving from data and information to context.

Mahesh Rajasekharan: Right? At this point, intelligence can now determine potential responses. And finally, the system can coordinate the appropriate action across the companies and systems involved. So I think the right progression is data leading to context, which powers intelligence, powers action, and powers outcome. So you’re going from those progressions, and when you go from that progression from data to context, intelligence to action, to ultimate business outcome, the outcome is ultimately what matters.

Mahesh Rajasekharan: That should become the scoreboard for enterprise software companies, not how many transactions did we process. It’s how many– And not even how many alerts we generate. What business results did we change, and did we make the customer incredibly successful?

Rob Southern: That all sounds like orchestration. W-why do you believe supply chain orchestration represents the next major enterprise software category?

Mahesh Rajasekharan: Rob, I’ve spent more than 25 years around supply chain, and there is a real structural problem that has fascinated a long time why people never solve, which is you have these multi-billion dollar planning systems that do forecasting, supply planning, inventory planning, fulfillment planning, et cetera. And people have spent billions, and they keep on trying to upgrade to future versions of those planning systems.

Mahesh Rajasekharan: Those systems understand what the business wants to happen, right? They understand the intent, understand the context. And also companies have spent billions on execution systems, ERP systems, transportation systems, warehouse management systems, and so on. And those execution systems are very good at processing transactions.

Mahesh Rajasekharan: But, but there is an enormous gap between the two, right? Which is planning understand the intent, but generally cannot execute. And execution systems can act, but oftentimes they don’t have the business context to determine the next best set of actions. And so we have a, a gap. We have a coordination and synchronization gap.

Mahesh Rajasekharan: And, and also the real world ex- never executes perfectly according to plan, right? You have something. There’s a tariff related, you know, hold back, uh, you know, at the, at customs or the Strait of Hormuz got closed, right? You cannot get through the, through the Red Sea or there is a, there is a strike. So the real world never perfectly acts according to plan, and by the time you get back to planning, it takes days.

Mahesh Rajasekharan: Let’s look at an example. Supplier is late. Okay, so inven-in-inventory isn’t where you thought it was. A shipment misses a connection. A customer changes an order, which happens all the time. Customers change quantities, they delete a line item, they want a different date. So something happens every day. And so we define this as supply chain orchestration.

Mahesh Rajasekharan: Th- you need to orchestrate between planning and execution. You need to connect the execution data with the business context and increasing the intelligence. So now let’s look at alternative reality where you imagine a system that, uh, understands the complete context surrounding an order, right? It understands the customer who place the order, the supplier, the inventory positions, the shipment, the logistics provider, the invoice, and all the economic surrounding it.

Mahesh Rajasekharan: Is this gonna be a late fee? Is, do I have to deliver on time? And if I don’t deliver on time, what is the penalty the retailer is gonna charge me? At that point, when reality deviates from the plan, that system can detect the problem. It can understand its impact, determine possible responses. Now you can coordinate across the entire ecosystem, take action, and learn from the outcome.

Mahesh Rajasekharan: So to me, that’s much bigger than automation, right? People use the word automation too loosely. For me, automation makes individual tasks faster. Orchestration makes the overall outcome better And there’s another reason I really find supply chains particularly interesting, because no single company owns the entire process, right?

Mahesh Rajasekharan: Now everything is outsourced. There are multiple participants in the supply chain. Your customer is involved, your supplier is involved, your logistics provider is involved. Sometimes the banks are involved. But ultimately, you’re not just trying to create an intelligent enterprise, you’re trying to create an intelligent business ecosystem because you’re getting across multiple supply chain participants.

Mahesh Rajasekharan: And so this is where AI combined with orchestration makes it increasingly possible, and that’s why I believe this will become a very, very important enterprise software category around supply chain orchestration.

Alex Bridgeman: I would imagine as a follow-up to that, that AI would make pulling information across all of those different groups along the supply chain and w-summarizing it and organizing it into a cl- a single clear picture a lot faster and easier than a version of that process, you know, five or 10 years ago too.

Alex Bridgeman: Yeah.

Mahesh Rajasekharan: Absolutely. Because if you think about any supply chain, just imagine there are probably thousands of signals coming at you, right? From, from ships, from ports, from, you know, custom solutions, from ERPs, customers sending orders, cancellations. The first order of decision-making is which is true signal, which is noise, and be able to filter out the noise and focus on signals.

Mahesh Rajasekharan: And then even within the signals, humans can only process so much, right? So you need to then decide what the payoff is. If I go and make this, this shipment on time for Walmart, and I earn their trust, I’m gonna get more, more volume, and I can go from the meat section in Walmart to the dairy section of Walmart.

Mahesh Rajasekharan: It’s a much bigger payoff than a long tail retailer for tech, right? It really depends on the decisions you wanna make. But codifying those payoffs is what’s gonna allow the, the human decision-makers to operate at a higher level. So in other words, we are trying to really make the shift from sort of very reactive firefighting mode to proactive decision-making mode in the time you have to make those decisions.

Alex Bridgeman: How should other CEOs think about building with AI and balancing those investments with smart capital allocation, especially as there’s more and more models available to use across from open way, open source to, you know, the latest and greatest, most expensive models and a whole lot in between?

Mahesh Rajasekharan: Yeah.

Mahesh Rajasekharan: They, they’re not slowing down. If you look at the, the frontier model, they’re getting increasingly better and better. But I think CEOs have to hold two ideas simultaneously. The first one is AI demands tremendous urgency. You need to act, right? You cannot think, you have to act. But urgency cannot become an excuse for abandoning economic discipline.

Mahesh Rajasekharan: So one thing we’re learning with Cleo, and, and this is the last couple of years of being deep in AI, is there’s a very different economics depending upon where AI is being consumed. So let me give you a very concrete example When we’re building software, and our software processes billions of transactions in hundreds of billions of GMV flows through our platform.

Mahesh Rajasekharan: So in our software, we want to access the, the highest quality frontier models. The way we build, we’re using sophisticated models with coding agents, testing agents, increasingly large fleet of agents participating in the software development process. And in that environment, we’re not focused on and not particularly interested in minimizing token usage.

Mahesh Rajasekharan: We’ll just spend the money. Now, suppose I spend significantly more, more on model consumption, but I, I’m enabling an exceptional engineer to produce much better software or dramatically increase development velocity. That’s an extraordinary economic, uh, trade-off. So in other words, we’re perfectly happy burning tokens and expensive models at human scale, keys human scale to build great software But now you move to transaction time, right?

Mahesh Rajasekharan: That’s completely different, meaning what we put in our product for our customers to use. Then the question becomes, you know, imagine in our case, AI operating inside high volume, high velocity supply chain processes across enormous number of orders, shipping invoices, and other transactions. Now the economics matter tremendously.

Mahesh Rajasekharan: You don’t necessarily want to spend, uh, or send every problem to the, the most expensive frontier model. You want the right intelligence for the right problem at the right economics. So, so sometimes it’s a, it’s a frontier model because the workload is sophisticated. Sometimes it’s more a purpose-built model.

Mahesh Rajasekharan: Sometimes it can be an open weight model, such as the models Llama family, where we can combine the model with Clio’s domain expertise and proprietary context, and we have our, uh, RAGs, or retrieval-augmented generation, to provide the right governed Clio or customer context at inference time. Right? So at the inference time, we can provide the context, and where appropriate, we can specialize or fine-tune, fine-tune the models for a particular class of problems.

Mahesh Rajasekharan: So then you add intelligent model routing, right? So the question to ask is: What is the complexity of the problem? What quality is required? What latency is acceptable, and what’s at risk? And that’s critical because then you get towards economics, and then you accord-accordingly route it. It’s a, it’s a class of work we do called model routing optimization.

Mahesh Rajasekharan: So the principle we use at Clio is at design time, which is when you’re bu-building software, you’re optimizing for intelligence. At transaction time, when people are accessing the, the platform, we optimize the intelligence per dollar. And I think that’s an incredibly important distinction for CEOs because the answer isn’t to use the cheapest model everywhere or wide-using frontier models.

Mahesh Rajasekharan: And it isn’t to use the most expensive model everywhere. What’s important is to understand the economics of the outcome you’re producing, and there is an interesting strategic implication. For example, the raw model intelligence is really becoming more ubi-ubiquitous. But as that is happening, the proprietary context you have as a company actually becomes even more valuable.

Mahesh Rajasekharan: Because more context and more indexing you can provide, you don’t need the most sophisticated model, right? So everyone has, has access to the same good intelligence. What is your differentiation, right? The, the question for software companies is your differentiation now moves towards who has the best context, who has the deepest domain expertise, who owns the workflow, who has the trusted data, who has actually first-party data or normalized second to third-party data, who can safely take action, and who can learn from the outcome.

Mahesh Rajasekharan: So that is a very interesting world of, world for Clio where We have spent years building those capabilities. And so now, as we think like an owner, you know, I simply don’t ask, “How much are we spending on AI?” I wanna know what economic value are we creating for every dollar of intelligence you consume.

Mahesh Rajasekharan: And that to me, ah, lets us capital allocation

Rob Southern: So i- in those examples, you’re, you’re talking about trade-offs and the decisions you make around those trade-offs, whether it’s different kinds of resources, people, capital, time, and then the principle that you bring to bear to actually make that decision. What other leadership principles become even more important in building an AI-native company?

Mahesh Rajasekharan: Yeah, Robert, I, I, actually I think about this a lot, and I think AI makes subtle, traditional, age-old leadership principles even more important. You know, to me, the first is clarity of purpose. You know, when technology changes this rapidly, there are hundreds of interesting things you could pursue, right? A leadership team needs a very clear true north, or the organization will chase everything.

Mahesh Rajasekharan: So that’s first, clarity of purpose. So you know what’s in north star. For us, it’s, it’s orchestrating supply chains, so we can create tremendous economic value by closing the gap between planning and execution. That’s a north star, and that we believe is a one point six trillion market problem, and the unpack individual problems like chargeback prevention, which are all in the fifty, sixty billion per problem we solve.

Mahesh Rajasekharan: So north star is important. The second leadership principle is really people. I, I’ve always believed that great companies are ultimately built by exceptional people, and AI actually makes exceptional people more valuable, not less valuable. I think it makes them more leveraged, right? A great engineer equipped with AI can accomplish things that previously required a much larger team.

Mahesh Rajasekharan: A great product leader can explore and test ideas incredibly quickly. A great salesperson, this is actually really important. Great salespeople are now able to get information across various sources, whether it’s Salesforce, whether it’s Gong, whether it’s, you know, different customer interactions, emails, and really leverage AI for account planning and prepare for customers at a completely different level.

Mahesh Rajasekharan: But there is a flip side, right? Which is when execution becomes easier, actual judgment becomes even more important. So getting the right people in the right roles remains fundamental. And the third is something we talk a lot within Clio called divine discontent, which is, on one hand, be proud of what we accomplished, but never become satisfied with it.

Mahesh Rajasekharan: Just push the boundaries. And Clio’s, uh, reinvented itself multiple times over the years, and AI now requires us to renew ourselves again. As CEO, I cannot stand on the sidelines and tell everybody to become AI native. I have to understand these technologies. I actually sit down with our developers, and they show me every week what progress they’ve made, what interesting things they’re working on.

Mahesh Rajasekharan: I have to use them. I have to build my own agents. I have to change the way I work. I have to ask better questions. And- And, and, and fundamentally, you lead the transformation as a CEO from the front. And finally, I think great leadership in this era requires what would sound almost contradictory, which is have a long-term direction, but you need short-term intensity, right?

Mahesh Rajasekharan: You have a North Star measured in years, but when the world is changing this quickly, you have to operate today with enormous sense of urgency. And I don’t see those ideas as being in conflict. I think the best companies do both

Alex Bridgeman: So with that rate of change, how do you allocate your own time and capital across acquisitions, new talent, innovation, markets?

Alex Bridgeman: It’s a lot going on. How do you think about those decisions?

Mahesh Rajasekharan: Yeah, look, I, I mean, for us it’s, you know, it’s a lot of growth at Cleo. And so one of the biggest transitions you make as a, as a CEO, as a company gets larger, not just AI, generally getting larger, and of course, AI has been a profound shift, is really realizing that almost every signi-significant decision is ultimately a capital allocation decision, right?

Mahesh Rajasekharan: Hiring another hundred people is capital allocation. Building a new product is capital allocation. Entering a new geography is capital allocation, right? Acquiring a company and, and doing, you know, sort of inorganic growth is capital allocation. So it’s, it– Also, choosing not to pursue something, especially in AI, it’s when there is a lot of pressure on people to, to do something, choosing not to pursue something is also capital allocation.

Mahesh Rajasekharan: So my orientation has been always towards long-term compounding, and we, we present a clear compounding blueprint every year. We, we bring it up in our, uh, you know, vision summit, and we talk about this is a long-term compounding. And I’m not trying to optimize Cleo for one quarter or even one year. The question we always ask is: how do we build a company that becomes structurally stronger every year for a very long period of time?

Mahesh Rajasekharan: And when you look at companies, many of those most important assets compound. Customer trust compounds. Domain knowledge compounds. D-data and context for us compounds a lot in supply chain. Product capability compounds because in our case, we have a platform and all of the context intelligence is building on the platform.

Mahesh Rajasekharan: So more signals and more transactions flowing through the platform allows us to build better products, especially in the AI era. Talent absolutely compounds. We have a, an internal form system, and we grow people. Of course, we hire great people outside, but our real goal is to just grow people from internal, internally upwards.

Mahesh Rajasekharan: Go to market capability compounds and culture compounds. So what’s really interesting about a major technology transition like AI is that occasionally something comes along that allows you to unlock those accumulated capabilities in ways that weren’t previously possible, right? That’s one of the reasons I’m so excited about where Cleo is today.

Mahesh Rajasekharan: We have spent years building many of these capabilities, and now AI gives us an opportunity to absolutely amplify them. So when we’re evaluating investment, the one question I find useful is- Does this investment simply produce a result, or does it allow us to build a capability that can help us durably compound, continue producing results?

Mahesh Rajasekharan: And those are two very different things. The second one can compound, right? And that’s alone how, how I think about acquisitions. So we’re absolutely open to acquisitions when they accelerate something strategically important, like technology, talent, data, domain expertise, go-to-market capabilities, market access, better than we can accomplish organically.

Mahesh Rajasekharan: But we’re not just doing acquisitions simply to, you know, get bigger, right? So we really believe in building businesses that can compound over time. And the same, same is true for new markets also. We always look at another market you can enter, whether it’s a new industry, getting to channels, getting to Europe, for example.

Mahesh Rajasekharan: But growth adds enormous complexity, so if you’re not smart, it can distract the organization and take capital away from a much higher return opportunity, which isn’t necessarily good for growth. So you also need the ability to say no, and this is where an owner mindset becomes incredibly useful, right?

Mahesh Rajasekharan: If it’s your money, would you spend it on, on, on any initiative? So the whole question then reduces to, you know, “This is entirely my own capital, and I had a 10-year horizon. Would I put the next dollar here?” And I think that question for us at Cleo removes a lot of noise because ultimately, capital has an opportunity cost.

Mahesh Rajasekharan: Every dollar you invest in one opportunity is a dollar you cannot invest somewhere else. So our job as leaders is to continuously move towards and move resources and dollars towards opportunities where we have the highest and the greatest combination of strategic advantage, customer value, and long-term compounding potential.

Mahesh Rajasekharan: And that’s really how I think about capital allocation.

Rob Southern: You mentioned compounding, you know, long-term time horizons. So looking ahead 10 years, what will enterprise software and leadership look like? And what advice would you give founders and CEOs like Alex and I building for that future?

Mahesh Rajasekharan: Yeah, Rob, I- I’m always a little cautious about 10-year predictions per se, because the world has a wonderful way of humbling, you know, great people, great CEOs who make those predictions.

Mahesh Rajasekharan: But I think there are a few things I have pretty high conviction about. First, I think enterprise software becomes increasingly agentic. Agents will participate in virtually every important business process. Agents will gather information, they’ll understand context, they’ll evaluate alternatives, they’ll make recommendations.

Mahesh Rajasekharan: And I believe increasingly, with appropriate governance and human accountability, they will execute. So software itself changes. We in Cleo call it as autonomous supply chain operations, right? The ultimate holy grail is a ship rerouting itself. But even with- without a ship rerouting itself, so many decisions can be made agentically, right?

Mahesh Rajasekharan: So I believe the winning platforms won’t simply have the best user interface. They’ll provide trusted data, rich context, domain intelligence, governance, security, and really the ability for humans and agents to work together and take action. And human beings will increasingly move towards the things we are uniquely good at, which is strategy, judgment, innovation, relationships, leadership, accountability.

Mahesh Rajasekharan: But if I were talking to a founder of a software company today I actually wouldn’t tell them to start with what’s an AI strategy. I would start with a much more fundamental question, which is what important and enduring problem are you willing to spend the next ten or 20 years of your life solving?

Mahesh Rajasekharan: Because the technology will change. The models, the frontier models are so excited about today won’t potentially become obsolete. The architectures will change. Today’s breakthrough will become tomorrow’s commodity, but important customer problems endure, right? So I would, I would, I would recommend to, to founders and CEOs to pick an important problem, get extraordinarily close to the customer, understand the problem, develop deep domain expertise, and then start in this AI age.

Mahesh Rajasekharan: Build appropriately data and context that’s your own, right? Create capabilities that become stronger with scale. So with scale, you need capabilities that just compound and, and become enduring and, and actually create an amplifying effect. Surround yourself with exceptional people. And when, when you do that, whatever the b-the best technology is available, use it to deliver extraordinary outcomes.

Mahesh Rajasekharan: And that’s really how, you know, I think about Cleo. And we’ve been building this company for a long time, and we have gone through multiple technology generations, right? And this happened over time, and we’ve had multiple chapters of reinvention. And I think AI is probably the biggest opportunity we ever encount-encountered.

Mahesh Rajasekharan: And one, one of the things I’ve learned is that the job of the CEO isn’t to perfectly predict what the world will look like ten years from now. The job is to build an organization that’s capable of renewing itself for the next ten years. So I would say have a long-term North Star. Surround yourself with extraordinary people.

Mahesh Rajasekharan: Stay incredibly close to your customers. They will teach you, or they’ll tell you if you’re doing something incorrectly and reward you if you’re doing it right, and show you what to do next. And be willing to challenge your own assumptions. Be humble. Move with tremendous urgency because the world is changing, and especially in the world of AI, you know, a mistake becomes a, an advantage and a benefit because you want to learn from that quickly.

Mahesh Rajasekharan: The systems learn. And so the key is really maintaining the discipline when everybody else gets cau-caught up in the hype. And never confuse technology with the outcome you’re trying to create, because ultimately, great companies are– aren’t built around a technology cycle. They’re built to compound through technology cycles, and that’s really what we’re trying to build at Cleo.

Alex Bridgeman: Mahesh, this has been awesome. Is there any question we should have asked you that we haven’t so far?

Mahesh Rajasekharan: I, I think you, you, you, you touched on pretty much, you know, everything I would think, uh, at this day. Just to reinforce, I think that the key really is AI is such a fundamental shift. It’s probably a 100-year shift.

Mahesh Rajasekharan: And so I, I think having an incredible sense, sense of urgency is the most important thing because you’re gonna learn. At the same time, you need to have a North Star, and at the same time, you should lead from the front as CEOs. I think that’s where sometimes people, I think, make mistakes when they delegate to a chief AI officer.

Mahesh Rajasekharan: So I’m not saying there’s anything wrong with having a chief AI officer. The CEO should learn what problem they’re solving and see what AI can do for it versus delegation. So I think that’s, uh, the one thing I’ll leave with, which is get, get actionable, get in the thick of things, learn, make mistakes, and, and recover faster than a competition does.

Alex Bridgeman: Amazing. Mahesh, thank you. Always love getting to chat and especially on the podcast, so thank you for sharing more of your time.

Mahesh Rajasekharan: Awesome. This was fantastic, Alex. Rock.

Alex Bridgeman: Thank you for listening. I hope you enjoyed today’s episode of Think Like an Owner. If you enjoyed the show, please consider leaving us a review and telling a friend to help more folks find Think Like an Owner. For full episode transcripts and our weekly newsletter, please visit our website at tlaopodcast.com.