He built a $20B public company, left—then raised a $100M Series A. | Dheeraj Pandey, Founder of Nutanix & DevRev

Dheeraj Pandey (00:00:00) :
You have to figure out PMF at every number. Even at a billion, you have to say, do I have a PMF for two billion? And you might not. You might say, look, I'm done. My growth is now two percent, and I just need to reap the renewals, that's it. So I wouldn't think that PMF is ever a destination. Don't sell and run. I think a lot of founders are very good at selling and going to the next account, and selling, and going to the next account. I think from promise to reality is actually journey. It's just like marriage. You date and you think your wedding is basically the last thing you care for. The real stuff happens after the wedding, you know. A use case is a focus, but unless you can make a few billion dollars on your own, you can't acquire your way into becoming a public company over time and those things cannot be retrofitted later. You have to have a platform with a use case that can then extend over time because most of the large enterprises, they don't buy features. They buy into a vision and they want to start with the use case.

Previous Guests (00:00:55) :
That's product market fit. Product market fit. Product market fit. I called it the product market fit question. Product market fit. Product market fit. Product market fit. Product market fit. I mean, the name of the show is product market fit.

Pablo Srugo (00:01:07) :
Do you think the product market fit show, has product market fit? Because if you do, then there's something you just have to do. You have to take out your phone. You have to leave the show five stars. It lets us reach more founders and it lets us get better guests, thank you. Dheeraj, welcome to the show, man.

Dheeraj Pandey (00:01:23) :
Well thank you. Thank you for the opportunity.

Pablo Srugo (00:01:42) :
Dude, I mean, I'm very excited to talk to you. You have quite a ridiculous background. You started Nutanix with co-founder and CEO now, a $20 billion dollar public company. Your a board member at Adobe and now you've started DevRev. Been working on this for about five years, recently raised $100 million Series A. So, quite an accomplished guy and curious to get into your story. And also into the lessons you've learned along the way. Maybe with that said, let's go back to 2008-2009, just the beginning of Nutanix and maybe we can go through that story. Maybe share your background and the context of that time just so we can kind of get into it.

Dheeraj Pandey (00:01:58) :
Yeah, thank you, Pablo. You know, I had been working in the industry for almost ten years, before we decided to start Nutanix and along the way, I was deep into data and large-scale data management. Built clustered file servers during the first dot-com boom, and a couple of years after that as well. Until even after the bubble burst, and learned what not to do in terms of building a company. And I was not a founder there, but.

Pablo Srugo (00:02:23) :
Technical type roles?

Dheeraj Pandey (00:02:25) :
Yeah, so I was a developer and systems developer, software developer, you know, building clustered file servers, Oracle database. At Oracle, I actually went on to become an engine manager and engine director. That's when I met my first investor later, the future investor Bipul Sinha. Who was the CEO at Rubrik, is another company that actually does backup and security. And then after that, for a couple of years, I didn't start a company but I was very early on building distributed databases. Which is data warehouses. This is pre-Snowflake, but an on-prem version of Snowflake. So I built a lot of large-scale data management systems, and the financial crisis happens. While I was a distributed data warehouse company. Basically, there were three of us who knew each other for a decade or more, who were all at this company. I was the VP of engineering there, and we were making a case that we need to actually bring both SQL and NoSQL together. It's very similar to what's happening right now with apps and agents. How do you bring apps and agents together, as opposed to just talk about the future. Which is agents and the past, which is apps. There was a time when SQL was being looked at in the past, and NoSQL was the future. Which obviously happened to be not true. Over time, SQL came back with a bang, you have companies like Snowflake and some of the largest money spinning services at AWS are all SQL services, which is databases and so on. So we were trying to make a case for how systems need to actually embrace both, and it wasn't to be. So we said, look we are going to go out and build our own company. So spent almost six months thinking about how do we bring our strength, which is data to a new product. Now in some sense it was the skills and then there was values, like, hey we're going to build something which is reliable, scalable, available, but we didn't know exactly what product would it be. Then we quit, we started to really spend time on it. Is when we realized, oh, there's a new movement in the data center, which is virtualization. How do you virtualize machines and suck it from physical to virtual? And, you know, we made a case that, hey, this needs a new architecture.

Pablo Srugo (00:04:28) :
This is all kind of born out of, the stuff happening in cloud? AWS, these sort of things. I assume, lead to that?

Dheeraj Pandey (00:04:33) :
And back then nobody even knew how to spell AWS, because there was no AWS. I mean it was probably happening all in the private arena of Amazon. Because they needed to host their half a million to a million resellers, with trying to replicate their catalog, and all this stuff. So they were building all this stuff in private but there was no public AWS. The thing that was bigger than AWS was VMware. 

Pablo Srugo (00:04:56) :
I see.

Dheeraj Pandey (00:04:56) :
And now, VMware was really saying, look, we can bring a lot of machines into the ether. Which is, you can't even touch and feel machines. It's called a virtual machine. It's what we're doing right now with agents is basically a virtual human role. It's like we're trying to virtualize human agency through these AI agents. So very similarly, back in the day, we were taking a lot of physical machines and saying there's a version of that. A digital twin of that in the matrix, and that matrix was called virtualization. It was an exact replica of physical machines. It just so happened that you couldn't touch and feel it. And on a single hardware, a single big physical machine, you can put lots of virtual machines. So it was a little bit like Inception, how does it even happen, you know? Anyway, so it is a new way of doing things. You don't have to buy new hardware to really spin up new projects and all this stuff. And we said, hey, this is the killer app for our data platform.

Pablo Srugo (00:05:52) :
This is the mix? The SQL, NoSQL mixed data base?

Dheeraj Pandey (00:05:56) :
Well, it was data. So we said we're going to start with storage and data management for virtual machines. It wasn't SQL, NoSQL, but it was just data. Because, and data has multiple layers. You start with data for virtual machines, data for databases. Data at large is a really hard problem, by the way. Because it has gravity, it has scale, it has reliability issues, availability, security, all sorts of issues. If anything, the whole idea of DevRev, and we'll get to that too at some point. We said, look, AI has to be a data problem. So we made AI into a data information knowledge problem as well. But we started building this thing for the new use case, which is virtualization and we also had to say, okay, so how do we deliver this thing? Because it's otherwise just an abstract platform. So we had to put it into a device. We call it appliances. We said we're going to take these commodity servers that don't get sold in the enterprise market. Because enterprise people were buying enterprise servers from HP and Dell, and IBM, and companies like that. And we said, that's the antithesis of private cloud. The way public clouds are being built. At least consumer clouds are being built, was using commodity servers, commodity hardware. That's the way Google and Facebook, Amazon, they're all using lots and lots, and lots of just servers, and software on top of it.

Pablo Srugo (00:07:09) :
As opposed to? Sorry, what were the big enterprises doing?

Dheeraj Pandey (00:07:11) :
The alternative, they were building big proprietary blade servers and then, big EMC and NetApp, and Cisco, and all this stuff. Complex networks, complex storage servers, extremely bespoke and proprietary blade servers.

Pablo Srugo (00:07:52) :
Very custom.

Dheeraj Pandey (00:07:26) :
And these are big machines that you'd have to pay millions of dollars to buy from Cisco and HP, and Dell, and so on. I think what we said is like, no, no, no, like consumer cloud, like private cloud, why should we be any different. So a very bold and audacious attempt at actually bringing the idea from consumer cloud companies and then they applied that to public cloud as well over time, for the last decade. But then we said we're going to bring the same architecture on-prem to people who still own data centers and so we put this in an appliance. We never knew hardware, so we learned hardware. It's like, you know, we were in love with the iPhone. We said, look, at the end of the day. Apple had to really own it end to end in terms of quality and, you know, fit and finish, and sizing, and all this stuff. Was done well and then after 2010, we also saw how struggling Android was when they didn't own the hardware. When they had to give it to handset manufacturers.

Pablo Srugo (00:08:17) :
So it wasn't just the data layer? You were kind of more of a full stack solution.

Dheeraj Pandey (00:08:21) :
Absolutely, it was compute and storage in one. We said, look, we're going to run your applications. When you look at the data center, it's just a bunch of servers and you have big, and small servers. But you use the software to create a facade of a single machine, even though it's not a single machine. The software is creating a facade of a single machine. So one single application could take up ten racks worth of data, and it will know that it's actually spread over ten racks worth of data. This is the idea of distributed systems. Distributed systems is really about providing a facade as if you have infinite capacity and infinite scale. Even though it might be divvied up amongst hundreds, if not thousands of servers. So we started doing this and said within virtualization, what's the use case? Because virtualization itself is a paradigm, but it's not a use case. So we had to go even more specific. So we said, what about desktops? Because now they're going to go virtual. Because there's Apple on the desk, and there's a tablet on the desk. What happens to the desktop? Windows, especially, and after Snowden. Basically, there was a big push towards virtualizing desktops. Like, okay, no machine leaves the premises. Everything is actually on the inside and it became a great use case for us to say, let's virtualize the Windows desktops in an environment that requires tens of thousands of these things. To be running in a data center, as opposed to running on ten thousand desks, and that became one of our killer use cases to begin with.

Pablo Srugo (00:09:46) :
You're selling this to who? Large enterprise mainly?

Dheeraj Pandey (00:09:49) :
Actually, we had even small ones. We took an entire rack's worth of thing and compressed it down to you. People love the fact that it's so compact and it's everything in one. They used to even say cloud in a box. We used to call ourselves a cloud in a box in the early days as well, and you can start with just a very small thing. And you can pay as you grow, scale it and extend it, as and when you felt like. So the idea of fractional consumption, which was getting to be popular with the public cloud. Was made popular by us in the private cloud. That you can really start with nothing more than $30,000, and you can take the same system and make it $3 million. No questions.

Pablo Srugo (00:10:29) :
Does the emergence of public clouds, of AWS, all that stuff. Which then happens in the background through that timeframe. Is that competitive to you, because it's taking loads away? Or what's the dynamics?

Dheeraj Pandey (00:10:38) :
In Nutanix, the problem statement was where the puck was. The puck was EMC, NetApp, VMware, HP, Dell, IBM, Cisco, all these companies and said, hey, we're going to really go and modernize all this stuff. But the consumption model was still ownership. You were owning data centers, you're building data centers and then by 2014, '15. We started to see how, the idea of not owning but streaming infrastructure. Started to take some, you know, gain some ground. Especially amongst the digital natives. So all these dot-coms or e-commerce companies, were beginning to say, look, we don't need to own data centers. We can start to rent it out. Startups started doing it as well.

Pablo Srugo (00:11:17) :
Yeah, we started. For example, our startup in 2013, 2014, and that was definitely the default back then. Was you're going to build on top of whether it's, I think, Heroku, AWS, whatever it is, yeah.

Dheeraj Pandey (00:11:26) :
Yep, and remember they had this thing about build your data center in five minutes. So it was beginning to kind of rear its head and it was a consumption model competition. Like, hey, do you want to own it or do you want to rent it? Do you want a CapEx or do you want an OpEx? And do you want to start really, really small? Because we were starting small. $30 000 over four or five years. Which is the cycle at which you refresh hardware. But we were looking at the public cloud and said, wow, they could be for hours. You don't even need it for days or months and it could be for hours.

Pablo Srugo (00:11:56) :
Right.

Dheeraj Pandey (00:11:57) :
So they were going even smaller than we were and for a lot of smaller companies it was actually making a lot of sense. For a lot of burst traffic and new use cases in 2015, '16, people decided to really check it out. But the large enterprises were still primarily on private. Now 2016 happens, we go public as a company. We were kind of fastest to a billion and all this stuff.

Pablo Srugo (00:12:18) :
Wow.

Dheeraj Pandey (00:12:18) :
And we start to get asked these questions like, hey, but what about something that's even more invisible than you guys? Because we used to call ourselves invisible infrastructure. Because everything was software defined and all you could see is a bunch of servers. So we called ourselves invisible infrastructure. But there was something more invisible than us, which was thing that you don't even own, you're going to rent. So by 2017, '18, I think it was pretty evident to me that we have to and along the way we had started to give our software to Dell, and HP, and others. To say, hey, take our software and see if you can build equivalent appliances like the ones that we were building with Supermicro. Which was our appliance partner. In some ways, we had multiple versions of these and they were also competing with each other. But we had become a quasi-software company, because we're giving our software to our OEM partners to also use. But then by 2018, it was pretty evident that, hey, this cloud thing is here to stay and that's when we said we're going to go complete software. We're not even going to recognize the hardware in our books. Then from software to subscription, subscription to one year term. So we really shrunk the contracts from four, five years, to three years, to two years, to one year. So a lot of miniaturization that had to happen along the way to really.

Pablo Srugo (00:13:33) :
Still for private clouds?

Dheeraj Pandey (00:13:35) :
Still for private clouds. The company continues to be focusing on the private cloud, but then February happens, 2020. COVID, and then all data centers are locked closed

Pablo Srugo (00:13:47) :
Right.

Dheeraj Pandey (00:13:47) :
And the only ones that are open are the ones that are called emergency data centers. They're hyperscaler ones. So most every enterprise which was postponing all this stuff. Because they thought they needed to have everything be owned by them, are saying, look, we need to really start to rent things and really burst into the public cloud, and so on. And that's when I took a pause, and I'm like, oh, wow. This thing that for ten years we kind of just looked at it and learned from it. We learned a lot from the public cloud too. But we observed it and we thought it's not going to touch the enterprise. It's now touching the enterprise in a big way.

Pablo Srugo (00:14:24) :
Was there doubt, originally? I'm curious, because you saw that evolution so close about how big public cloud would be. Was it taken as a bit of a joke at first or was it always considered a serious threat?

Dheeraj Pandey (00:14:35) :
I was always a big fan of both Apple and AWS. Always. because there was a developer in me that just loved the way they did it. I had learned design from a lot of iOS and Apple thinking. But also, they gave us the courage to say that our architecture is right. Because both of them had converged everything. They removed gadgets. iOS removed a lot of gadgets and AWS removed a lot of gadgets. They just left shifted all that and said, look, you don't have to worry about going to ten different vendors to spin up an app.

Pablo Srugo (00:15:06) :
This great bundling to value, yeah.

Dheeraj Pandey (00:15:08) :
Yeah, exactly and there was convergence of a kind which really reduced complexity. Including in our personal lives with Apple and Android. I used to look at them as good archetypes for what simplification means. Even what miniaturization means, because you have to miniaturize things to integrate them as well. So I always used to look up to them and say, what do we learn from them? You know, I think there are lots of things that Amazon did. Which was amazingly different, not the least of which is like commerce engine and security engines. They didn't use IT to build them. They just had developers building a lot of these things, you know, and that actually was one of the big things. In consumption, you've got to really respect commerce and the way you meter, and bill, and invoice every month, and all this stuff. I think public cloud was as much about commerce, as it was about open source being commercialized as a service. So there was lots of great lessons that I took away from the ten odd years of watching and being a cousin of public cloud.

Pablo Srugo (00:16:10) :
Now you ended up leaving Nutanix, as I can see at end of 2020 or so. COVID seems to be a pretty important moment in this story. So what happens after COVID for you?

Dheeraj Pandey (00:16:20) :
I think it was evident to me that if you're not going to be a service company. It would probably be a market that's now controlled. Because the real change is happening. The real action is happening in the service economy, in the subscription economy, in the consumption world, which is public cloud. You know, my entire personal wealth was tied to an on-prem stock in some sense, you know. So I think it took me that whole year to figure out, okay, so how do I become an investor in this one? And, go and operate in something else? But the ideas of the something else, the new thing called DevRev. Actually, a lot of inspiration from running Nutanix. We're doing almost $2 billion in annual revenue at eighty plus percent gross margin, twenty thousand customers, thirty thousand sites. So lots and lots of big scale. I mean, we probably are managing about a million machines belonging to twenty thousand customers and to upgrade them, and observe them, and do all sorts of telemetry on them. We've done a really good job of customer support, like amazing job of support. We had a great net promoter score of ninety for the longest time. Yet we had not used Salesforce Service Cloud as a way to really front-end our customers with someone else's support software. We said we're going to build our own, so we built our own portal and our own telemetry, and our own data analytics, and all this stuff. And we had used Slack to the hilt with our customers. In just real-time collaboration and that was amazing. Because we had very, very large customers. $100 million customers, $200 million customers and a lot of that was, because we had done justice to looking at customers as part of our team.

Pablo Srugo (00:17:58) :
Right.

Dheeraj Pandey (00:17:59) :
And real time was key. And everybody learned on the Slack channel from, end user customer to support people, to product managers, to developers and sales engineers. Everybody learned along the way and sales executives. We like, you know, there's something to be said about looking at this as a team, but the customer is part of your team. So the idea of DevRev was really born out of a lot of that experience. Where we said, look, we have actually converged a lot of gadgets in the data center space and the largest opportunities are in converging things. And that's what smartphones became. They converged a lot, and that's what public cloud became. So DevRev was about converging a lot of business software. But then you had to think about, so what's common about business software that you can converge them all? There has to be some thesis behind convergence.

Pablo Srugo (00:18:46) :
And by the way, are you doing this? You decide to leave and you figure it out, or all this is happening kind of simultaneously as you're thinking about the next thing?

Dheeraj Pandey (00:18:53) :
I mean, it was at the back of my mind for like three, four years. Because you don't know, I was always big into systems. Including business software at Nutanix, like, the best implementers of NetSuite and one of the best implementers of Workday. The way we did a lot of these things. I didn't look at them as costs and I'm like, look, these things have to be done really well. The best support portals, not using Salesforce like the way most other people use Salesforce. To using Slack and all this stuff. So I always was passionate about process and systems. So I'm like, it was at the back of my mind but I would say the last half of 2020. These things became more concrete and yet it was nebulous. Because you just didn't know what converging business software would mean.

Pablo Srugo (00:19:39) :
Yeah, that's right.

Dheeraj Pandey (00:19:41) :
Look, most SaaS tools are work management tools. We just happen to call them by different names. The work for sales is called opportunity, work for marketing is called campaigns, work for support is called tickets, work for ops is called incidents, and work for developers is called issues.

Pablo Srugo (00:19:58) :
Yeah it's a system of records in different departments across the organisations. Something like that.

Dheeraj Pandey (00:20:03) :
And they all have states and stages. Some constraints on how these states and stages work with each other or not. You have to be able to customize them. So that customers can customize all this stuff and there's automation stuff. So that's work and then all work is tied to some identity. Which is who's working on work? And on which customer's behalf are you working on this work? And then what are you working on? Which is parts, we call it parts. Parts is the stuff you build or the stuff you serve. So products or services. So work, identity and parts were the canonical representation of a knowledge graph of a company. And the best part, it's reskinnable. You can call it opportunity, you can call it tickets, you can call it incidents, you can call it issues. So that was the aha moment for us. That look, a lot of these SaaS companies are built in silos. They just change physics, which is location from on-prem to SaaS. That's all they did, in those twenty years. But nobody did the chemistry. It happens that after GPT, everybody's talking about that. That, hey, agents don't need to worry about all these apps and so on. Because everything is about a knowledge graph. But we were thinking about this even from apps pre-GPT as well. We said all apps need to have a common trunk, which is a common knowledge graph and then on top of that, a common middleware, search, analytics, and workflows. These engines are very important and SaaS never got these engines right. I mean, we struggled to search for things on Salesforce.

Pablo Srugo (00:21:21) :
I've been at VC since like 2018, and I remember. I mean, this thesis of, you know, the SaaSification of all these departments and all this fragmentation, and trying to stitch things together. And, you know, Dropbox put something out at one point. I mean, I don't think anybody's figured it out sort of thing but.

Dheeraj Pandey (00:21:34) :
Yeah, yeah and I think these engines were important. The search engine, the SQL engine, because typically you had to pour out all the data from SaaS every night and put it in Snowflake to even get to integrate a lot of things from across different departments. So the only integration in a company was a warehouse. Like, hey, you got a warehouse and you can or a data lake or a lake house, as people would call it.

Pablo Srugo (00:21:56) :
And then for like business dashboards and these sort of things. Like, just put things together.

Dheeraj Pandey (00:21:59) :
Yeah, but there's so much more to integration, than that. You know, the workflow engine itself, like people use Workato or Zapier or ServiceNow to build a compute layer that could actually go across events coming from different departments. But then search is as important. So search, analytics, workflows became the three sort of pillars of this trifecta. So initially we were building apps pre-GPT. We said, hey, we're going to have a support software with a chatbot. So we were conversational from day one, but it was not really intelligent chatbot. It was probably a chatbot like the previous decade and then we had a build software for software developers, and it was a cousin of support. And then we had a grow, which is a sales CRM. So we had three apps and a chatbot. Support, build, and grow. And then Pulse GPT, I think something struck us that, hey, I know we've been replacing apps. Which people want to, but is there something even more accretive and maybe a little bit less of a zero-sum game that, hey, if you want my app, you got to get rid of the other app or something. Because that's how we were really doing it.

Pablo Srugo (00:23:05) :
I see.

Dheeraj Pandey (00:23:06) :
We were replacing Zendesk and Service Cloud and some ServiceNow.

Pablo Srugo (00:23:09) :
You were trying to build functionality all in one? Like different apps but all in one?

Dheeraj Pandey (00:23:14) :
No, there's still three apps, you know. There were three apps, so for three personas.

Pablo Srugo (00:23:17) :
I see.

Dheeraj Pandey (00:23:18) :
But they had a common infrastructure. Common search, common analytics, common workflows. In fact, post-GPT, we have three agents now. But they all share the exact same trunk. So the trunk of the tree and the roots of the tree are all common. So the roots is the way we replicate data from other systems back and forth. We call it AirSync, and the trunk is the knowledge graph. It's search analytics and workflows. And there's three apps, and there's three agents. And they share the exact same infrastructure. So the leverage is immense. We have a customer experience agent, which is a cousin of support. We have a developer experience agent, which is a cousin of our build app. Which is for software development and then we have sales experience. Which is a cousin of the sales CRM and the grow app we have, you know. And along the way, the agent builder. Because we really had to build all this stuff on top of a studio, and that's an agent builder, those skills builder. These things have become so good for us that we're offering to our customers saying, hey, you can also build agents on top of DeBro. But then you get the strength of data and the knowledge graph, and the search engine, the SQL engine, the MCP server. You get all of this stuff that otherwise will take you two to three years to build, and just give it to you. Because we've done it so well.

Pablo Srugo (00:24:31) :
We have tens of thousands of people who have followed the show. Are you one of those people? You want to be a part of the group. You want to be a part of those tens of thousands of followers. So hit the follow button. By the way, you mentioned search a few times. We had Arvind Jain also from Rubrik. You mentioned Rubrik earlier, on the show. Where does Glean, for example, which is also trying to unify it. But just strictly on the search, at least that's where they were born. Where would they fit into your view of the world?

Dheeraj Pandey (00:24:55) :
I think they started well with unstructured documents, but there's so much to search that it gets to answers, and I differentiate answers from search. Because search takes you to a document, and then you click on that link, and it takes you to a paragraph. Then you cite the paragraph but answers is about integrating a lot of stuff on the fly and if you don't have a really good knowledge graph. Which understands, all sorts of structured data coming from different departments. You know, your account data, your opportunity data, your ticket data, your user data, your incident data, your product engine data, all this stuff. Plus, of course, the unstructured stuff. So for us, we own the data because we make a secondary copy of it. We believe that the knowledge graph has to be native and if you don't have a native version of this. Which is also syncing real time with your existing systems and so on. You can't deliver the best experience for not just search, but answers. Not just answers, but reasoning as well. You've got to reason on things, which at some point will be about fine-tuning models and not just using prompt engineering on frontier models, and so on. So we look at this as a journey of search, answer, reasoning and you need a lot more than unstructured document search. Including a lot of structure, and you need SQL. SQL is really important. Without SQL, you can't say you have a great knowledge graph. In fact, we want to go and tell the market that the last thirty years, you've looked at data warehouse and SQL as your panacea for integration. Don't forget that, SQL is still really important and then the last one is custom code. Which is workflows, because LLMs will sometimes not know how to generate code with SQL or to do dynamic searches using RAG, which is search. Sometimes you need to take control and humans need to write code, and work through their API. Which is called MCP now. That MCP is ours, we don't just depend on Salesforce's MCP or ServiceNow's MCP. Which, by the way, they won't be very good at. I can tell you that, because they don't want to be a data-only company. These large incumbents, they don't want to be a data-only company. They would rather have the eyeballs too, and this happened in the 90s with mainframes. Where they became invisible over time, because they sucked out and hollowed out all the data from them. So the incumbents will resist a lot of the MCP movement. So owning the knowledge graph and really taking great care of structured data, and really linking it all together. And because we have app roots, we have to do all that.

Pablo Srugo (00:27:24) :
Yeah, this is what I wanted to get into is that storyline. There's going to be a pre and post, the stuff that happens with AI. What products do you build and where were they deployed?  Is this like large enterprise? Do they get rid of their existing apps in that world and then they adopt DevRev? How are things working?

Dheeraj Pandey (00:27:39) :
So the large enterprise, we do agents and there's the agent for the generalist worker. Which is like a Glean, except that the knowledge graph is native to us and we do structured data really well. We do permissions natively, we do workflows natively but then for the specialist. It sits next to your Salesforce support software or Salesforce sales software, or Atlassian JIRA software. So we go to multiple departments and show them the value of agents. No different than the way public cloud said, hey, come and build some new apps on top of us. And then, you can always lift and shift the old stuff to us.

Pablo Srugo (00:28:14) :
So at first, it's not a replace. It's keep using your stuff, but just use us on the side. We'll integrate with it. Something like that?

Dheeraj Pandey (00:28:19) :
Exactly, now in the mid market. They do want to replace, because they do want an AI native software and this is where we're replacing a lot of Zendesk, a lot of Service Cloud, and so on. So we're looking at this as kind of a two-phase journey. There's the mid-market, the SMB, and then there is the large enterprise. In the mid-market and SMB, we replace a lot of Intracom, a lot of Zendesk, you know, some Jira, and then we bring them agents as well. So apps and agents in the mid-market to smaller. And agents only in the large enterprise

Pablo Srugo (00:28:51) :
And I'm curious to talk a little bit about, because you've seen so many cycles now. You mentioned earlier, where the puck was, Just that line that you said when we were talking about Nutanix. Where do you see things going? What's the end state with AI? What happens to work? What happens to SaaS? What is the future of software sort of thing? I'm curious if you have big picture ideas on that.

Dheeraj Pandey (00:29:10) :
Yeah, I mean as I mentioned before, I think the biggest revolutions are integrative. They integrate stuff, like the iOS integrated stuff, the AWS integrated stuff. So AI's biggest value creation is integration and in ease of use, Which is can you miniaturize the complex app? Which was built for specialists and the partner ecosystem. To go and customize the heck out of it over one, two years of implementation. I think we basically just made it so complex that no one could end up using it anymore. Except for a few people who are stuck onto it for eight hours a day. I think we have to democratize business software and think about the casual user. You know the rubber necker, the window shopper, you say, that interface has to be simpler and probably miniaturized. And that's what conversational brings to the table.

Pablo Srugo (00:30:03) :
Yeah, is it just natural language? Is that the new kind of UI?

Dheeraj Pandey (00:30:07) :
For a lot of generalists, true. For specialists, they don't want to keep typing and saying the same thing over, and over again. So there is a design pattern for those people who are going to spend eight hours a day on apps and one, you could argue that why are they spending eight hours a day after all? Because agents should do everything. But let's just say that, and here is where I bring the analogy of e-commerce. In '95, I think there were companies that argued that everything should be driven by e-commerce. But by 2019, this is 25 years of Amazon.com. Only nine percent of US retail was e-commerce.

Pablo Srugo (00:30:41) :
It's funny, there was another by the way. Post COVID was the same, just on commerce. It was like this acceleration and now we're going to be on this new inflection point. And it kind of just came back down to the old kind of steady ten percent, twelve percent yeah.

Dheeraj Pandey (00:30:54) :
It's actually nineteen percent now, but realize it's almost thirty years since the internet and we had twenty percent, nineteen percent. And by the way, we're not replacing intelligence. There's a brick and mortar. There was a lot of physics that was being replaced with digital. I feel like this is a twenty, thirty, forty year journey and there'll be new jobs for humans. Simply because we evolve faster and we need to create history for the LLMs to copy from. Because they are basically caricatures and of course, they do have emergent capabilities. But more often than not, we just ask them to live in history in the past. Live in the past, I'll give you a million examples. Even for fine tuning and we give them evals, and benchmarks. So I would say that human jobs will evolve. A lot of chores will actually get done by machines now. But to say that all jobs are gone is also foolhardy. What else will be true? I think we'll probably have fewer things that will be extremely departmental siloed, like the way SaaS was. I think we're going through the chemistry of SaaS. In the first twenty years, we just wanted apps to be on the internet and people look so happy, like, wow, I have a browser access to a business software. Which is great. Because before that, it used to not be on a browser, you know.

Pablo Srugo (00:32:05) :
That's right. You can access it on your phone, you can access it at home, that's the big kind of thing. It's anywhere.

Dheeraj Pandey (00:32:09) :
And even the phone one didn't happen well. I mean, business software is still stuck in point click and scroll. In a sort of the human computer modality of 1984. Because the mouse was commercialized in 1984, and I don't think business software moved to mobile after.

Pablo Srugo (00:32:26) :
No, the old ones tried to mobilize and a lot. Then the new ones like a Slack or a Dropbox that was built in world did pretty well.

Dheeraj Pandey (00:32:33) :
Yeah, and they did well as a system of engagement. There was no intelligence in Slack. It was a system of engagement.

Pablo Srugo (00:32:39) :
Correct.

Dheeraj Pandey (00:32:39) :
So now the question is, can we make the record invisible and still make beautiful interfaces that are engaging and intelligent? And this is where we use the word team intelligence a lot. I think there's a great opportunity to bring the agent to be a teammate. And how do you get rid of Slack channels? Because at the end of the day, my only teammates in Slack were people and I had to keep asking questions to people, and wait for people to answer. I think in the new world, collaboration will be with agents who are not over eager to chime in and butt in for every question. They know everything, but they also have extreme EQ to know when to answer and what to answer, and so on. So we feel like it's headed in that direction where work will get collaborative. The idea of channels will probably wither away. Because channels were also prescriptive, as opposed to dynamic. You could cluster things on topics dynamically as well. So what does the role of AI look like in this world of collaboration? It's actually a big jumble. I don't think anybody's gotten it. We do know that we have to reduce Slack channels.

Pablo Srugo (00:33:48) :
Well, channels is funny. It's like a way to organize something that's very hard to organize and it works until it doesn't work at all. And then you realize you've made a bigger mess than it could have been.

Dheeraj Pandey (00:34:00) :
Absolutely, absolutely. But yet at the same time, it was real time and people registered their intent in topics. So there was something to be said about, hey, we hate the system record. It's not collaborative. So we need a system of engagement that's collaborative, but it lacked the record. So, you know, there was either you get extreme workflows and constraints on business, and business process, and automation. And some clunky UI, or you get engagement but not both. And I think it's a big opportunity to really have a view that says the record matters. It needs to be awesomely engaging with conversations and it needs to be very intelligent because of agents.

Pablo Srugo (00:34:41) :
And then maybe just to go from the future to the present. What is the state of DevRev product today? Maybe give me an example of one of your customers, by name or not? And how they're using DevRev today.

Dheeraj Pandey (00:34:51) :
Yeah, so we are taking customer experience, the CX route. I'm the first big agent to go and charm our prospects, and customers with. So we have a very large payment software company. Probably one of the largest in the country, and they are a Salesforce shop. They wanted to reduce the cost of customer support. They're spending upwards of $30, $40 million on just support. Every ticket was $18 of cost and they struggled with the agent force. And remember, L1 support is just awesome search, because you're just deflecting based on search. And even that for Salesforce is a really hard problem. You know, L1 alone. Forget about L2, which is workflows. You know, you get into my account, my bills, my subscription.

Pablo Srugo (00:35:34) :
This is common queries? The very top of funnel, easy stuff?

Dheeraj Pandey (00:35:38) :
And that alone, they were struggling at like thirteen percent deflection rates. Out of which ten percent were abandoned sessions. People are like, man, this thing is a moron.

Pablo Srugo (00:35:47) :
Right.

Dheeraj Pandey (00:35:47) :
Never going to answer anything from me and remember, this is just awesome search. And this is why I take you back twenty years ago, twenty-five years ago. There were twenty search engines, and then was Google. There was Google, right? I mean, even Microsoft couldn't touch them. Yahoo couldn't touch them. So search done well is still a really hard engineering problem, you know? And then you add, all sorts of things about namespacing and brands, and regions, and locations, and languages, and all this stuff. You know, it's a really hard. So even L1 support is a very hard problem. We went in there and we did eighty-six resolution for them.

Pablo Srugo (00:36:21) :
Wow, what are you doing different? Especially in this case, from ten, fifteen percent to eighty-five percent. That is leading to that kind of a delta in terms of quality?

Dheeraj Pandey (00:36:30) :
Yeah, so let's start with the basics. Like, crawling and indexing. Recently we had a customer that had almost a hundred terabytes of SharePoint and Dropbox to actually get in. That's enterprise-grade stuff. They had documents that were worth gigs, like single documents. If you try to do this in the old service-oriented architecture way. You'll be dead and then you have made assumptions about limits, and file sizes, and folder sizes, and data sizes. The way we have done crawling and indexing is through serverless. We said that it's going to be completely through Lambdas. A lot of the work that we do on behalf of indexing and crawling is like swarms. We do swarms. When a new customer is getting onboarded, we can get done 100x faster than any other company out there. Simply because the way we have done the cloud native architecture. I'm not even talking AI native yet, just cloud native thinking. We're probably one of the only AI companies that uses Lambdas the way we use it.

Pablo Srugo (00:37:30) :
And that's just the structuring of data? The organizing and understanding of the data?

Dheeraj Pandey (00:37:35) :
This is just compute. How do you swarm compute? Because you have to parse every document at scale and do it. I mean, it would not take months to actually take 100 terabytes and ingest it in, right? That's just one. Now you get to indexing. Now, of course, we had to build a multi-tenant architecture. I'm going to say this twice, just for effect. There is no AI company that's multi-tenant right now. There is no AI company that's multi-tenant. In fact, we make it micro-tenant. So even a three people startup can use our stuff. We don't even ask them for paying us. Simply because these indexes basically are multi-tenant. Now we do have a single-tenant version for somebody who really wants to make it single-tenant. But the point is that if you have to embed data from everywhere into something that's multi-tenant. You need to really spend time thinking about how you'll actually push down filters and authorization. I mean, authorization, for example, you don't want to bring up a hundred documents only to know in your code that ninety-nine of them are not even authorized to be seen.

Pablo Srugo (00:38:40) :
Right.

Dheeraj Pandey (00:38:40) :
You have to push authorization down into the engine and this engine is actually open source. So now you're using AWS for a lot of these things. But how do you really convert all your business logic, all your authorization policy engine into queries that actually map to AWS OpenSearch at scale. You know, is a really, really hard problem. Not the least of which I just said multi-tenancy and all that stuff to come in as well. Now, there are things that you need to do that are syntactic, which is keyword search, and the things that you need to do that are semantic. Now, on the semantic side, you need to know what embedding models are you going to use. Because obviously a big part of AI is taking language and converting it into the math, which is what we call vectors. You know, the idea of vector databases and so on are very, very important. But vectors is garbage in, garbage out. So if your embeddings, which is the vector space that you pick. Are not done well, then there's nothing that you'll get out properly as well out of that. So even to think about the small language model that you're using to actually convert language into vectors is something you need to keep fine-tuning every day as well, and on a per-tenant basis.

Pablo Srugo (00:39:50) :
Does that impacts ultimately the quality of the search once you go out and search?

Dheeraj Pandey (00:39:54) :
Absolutely and then, trade-off is between accuracy and latency. How fast can you get it out? Because the user is not twiddling. Waiting for this thing to come out and so on. So there's so much of this trade-off in accuracy and latency. You got to think about benchmarking for different kinds of data sets. Language is a big issue. You can't think in English and respond in Japanese. The embeddings actually take Japanese documents, they need to convert into vectors and bring back Japanese. Because translation is not an afterthought. It needs to be pre-built into your vector thinking and so on. So lots and lots of examples like this. I mean, I can go to the depths of search alone, and this is the reason why we geteighty-six percent. Not because we just did some hack for this one customer or something.

Pablo Srugo (00:40:39) :
Is this strategy, now that. Let's say you've got this customer support agent that works better than others. Because you have this much bigger, broader, I should say, vision than just customer support. So do you then go on that vision or you just sell this customer support agent to, ten , a hundred customers and then build the other? You know what I mean? How do you think about the two paths?

Dheeraj Pandey (00:40:57) :
See, I mean, even at Nutanix. There was all this in the first four or five years, I would say. Maybe even ninety-nine percent of the employees are like, why are we doing anything beyond virtual desktops? But we knew that desktop was a use case for the platform. It was not the company. That's why only we 

Pablo Srugo (00:41:15) :
It's very different, right? Because the standard startup 101 stuff is focus. Find your UI, your use case, just do that.

Dheeraj Pandey (00:41:21) :
A use case is a focus, but when you look at Gong, for example. It's in tatters. Like, hey, we're just going to do a $10,000, $15,000 thing, and we just created a $200,000, $300,000 business. But unless you can make a few billion dollars on your own. You can't acquire your way into becoming a public company over time, or even a large enough standalone company over time and those things cannot be retrofitted later. You have to have a platform with a use case that can then extend over time. Because most of the large enterprises, they don't buy features. They buy into a vision and they want to start with a use case. Because it's too much for them to have so many vendors. A large enterprise buys operating systems. at scale. Otherwise you can just be one of those companies that wants to exit at $50 million like if someone else is biased. Why do you need to worry about anything more than $50 million? Now focus is, I would say, the other end of the barbell. That use case better be awesome.

Pablo Srugo (00:42:18) :
Right.

Dheeraj Pandey (00:42:19) :
But then if you're not thinking about what the next use case will be, it will basically be just that and I think for us. We knew that agents and agency, and knowledge graph, and search, and analytics, and workflows, these things are not just for CX. But CX is a very important use case. I need a phone that has music playing to it. Yeah, there'll be a camera, I'll get to the camera later on. There'll be a flashlight, there'll be a GPS. All these things will be there. Where do we start? And AWS is like, man, I don't know what a public cloud is. Can we start with S3? Just S3, like object storage only. Once I like object storage, then I move to EC2 and then I'll move to EBS, and EFS, and RDS, and all this other stuff. So at some level, every platform has to start with a use case.

Pablo Srugo (00:43:09) :
So right now you have this one use case? Do you stick with a subset of customers and build out use cases, and then you go to market with all of them? Or do you do everything at the same time? You go to market with this one use case, as you develop the other ones on the product side?

Dheeraj Pandey (00:43:23) :
So we have been very clever architecturally to know what is platform and what is custom. Really, really, so the idea of workflows. We're like, no, we're not just going to build it for support departments only. A workflow engine is across all departments, so it's platform. I'm not going to hard-code the fact that there is a ticket to be in the search itself, as opposed to others will actually make it so. In fact, I have to make it work for custom objects. Things that I don't even know what they mean. Like an HR record, which I'm not sure what it means for me. Because it's not a native object. But if somebody sucks in custom objects, I should be able to search it, I should be able to SQL it, I should be able to workflow on top of it.

Pablo Srugo (00:44:03) :
So a lot more thought up front as a result, but more rewards later.

Dheeraj Pandey (00:44:08) :
Absolutely and you know that there's no instant gratification in this. Because I can just hack it up right now and I mean other examples like, okay guys, we're going to build a Kanban board. But not just for build. The same Kanban should work for opportunities. The same Kanban should work for tickets. Why not? It's a platform. It's not part of the app, right? So even the UI features, like sprints. Why should sprints only be for issues? It should be for tickets, too. I mean, there are professional services people who use sprints for tickets. So the way you think about what is shared and what is specialized is a very important piece of the puzzle. It slows you down a little bit.

Pablo Srugo (00:44:45) :
Yes.

Dheeraj Pandey (00:44:45) :
Because you have to think about a general, sort of generalized way to do this. But you know, it slows you down twenty percent. You know you can actually run really fast later on as well.

Pablo Srugo (00:44:54) :
In the classical conceptualization of these startups. Especially for, and this is where you have to think about an experienced founder versus a first-time founder. Because for a first-time founder, it doesn't just slow you down. It adds risk in the sense that if you're building the wrong thing and you're building it very thoughtfully, and all this stuff. But it's still the wrong thing that doesn't get adopted, you know, that's all waste, right? Now, if you're sure that, or you have high confidence in where you're going. Then it's all value towards that, realizing that vision. But it's a subtlety that for the founder who's not sophisticated enough, it can be ultimately deadly. Because it's like, oh, I built all this stuff and nobody wants it.

Dheeraj Pandey (00:45:29) :
And by the way, all this stuff is actually about optionality. It's not you're building everything right now. Like we build, build for which is to build our own software. We're like, look, we're not going to use Jira. We're going to build our own software and build. But we didn't sell it. The first thing we sold was support and we didn't use that much support early on. Because we didn't have early customers. This is two, two and a half years ago. So there's a dichotomy in this. You need to know what you're using and what you're selling as well. But the sophistication piece is basically what you get paid for over time. What is common, is a very hard design problem and you get paid for actually thinking about what's common. Most people in the world, they're like, no, no, but it's different. But you don't realize that the hardest thing to say is eighty percent is common. Find commonality before you find differences.

Pablo Srugo (00:46:14) :
Perfect. Well, Dheeraj, let's stop it there and I'll ask the last three questions that we always end on. The first one, you can answer this for Nutanix or for DevRev, up to you. But when did you personally feel like you had found true product market fit?

Dheeraj Pandey (00:46:25) :
You know, it's actually at every milestone, there's a million, $10 million, $50, $100, $250, half a billion, billion. You have to figure out PMF at every number. Even at a billion, you have to say, do I have a PMF for two billion? And you might not. You might say, look, I'm done. My growth is now two percent, and I just need to reap the renewals. That's it, you know? So I wouldn't think that PMF is ever a destination. The only thing I can say is, depending on probably some sort of a rule of forty in there as well. Is like, depending on how much you're growing. How much you're adding with new customers versus how much you're reaping from existing customers is what you need to really worry about, and as a younger company. If you're growing fifty, sixty, a hundred year over year. At least thirty to forty to fifty percent of your current quarter business. Should come from existing customers and a proxy for that is net dollar-based expansion. So if your net dollar-based expansion is upwards of $125, $130, you are in good company.

Pablo Srugo (00:47:22) :
Second question is, and again, for either of the companies. But was there a time where you thought things might not work out? Things actually could just fail?

Dheeraj Pandey (00:47:31) :
I think at Nutanix we really took upon things that we never understood, like hardware. When we were software guys, pure software guys, and two times at least in the first five years. We felt like we had a near-death experience and even VMware didn't work out for us in the first two, three years. I mean, we threw upon it a load. A workload that it had never seen before and man, it felt like we can, you know, basically shut shop in some sense. But the fact that we kept going, I think it's about being tenacious and luck favors you. And I think luck favored us a couple of these times.

Pablo Srugo (00:48:03) :
And then last question, what's a top piece of advice that you would give an early stage founder that's trying to find product market fit?

Dheeraj Pandey (00:48:10) :
Don't sell and run. I think a lot of founders are very good at selling and going to the next account, and selling, and going to the next account. I think from promise to reality is actually journey. It's just like marriage, you know, you date and you think your wedding is basically the last thing you care for. The real stuff happens after the wedding. I mean, just to finish off the thought, it's like, sell and stay. Get your existing customer to pay more with you, for you and that's the only way to really, really get product market fit.

Pablo Srugo (00:48:37) :
Love it. Well, Dheeraj, thanks so much for taking the time and it's been great having you.

Dheeraj Pandey (00:48:41) :
Likewise, thank you again, Pablo.

Pablo Srugo (00:48:43) :
Wow, what an episode. You're probably in awe. You're in absolute shock. You're like, that helped me so much. So guess what? Now it's your turn to help someone else. Share the episode in the WhatsApp group you have with founders. Share it on that Slack channel. Send it to your founder friends and help them out. Trust me, they will love you for it.