He hit $1M ARR with just 2 people. 2 years later, he's worth $1.5B. | Ashwin Sreenivas, Co-Founder of Decagon

Pablo Srugo (00:00:00):
Decagon started 2023, you've raised $230 million already. You've built a $1.5 billion company in two years.

Ashwin Sreenivas (00:00:06):
A lot of first time founders and us included at the time. Spent a lot of time, I think, over-intellectualizing the problem, right? Being able to figure out, oh, you know, what is the exact three year strategy at the start? And I need to know what the exact plan is before I start doing stuff. And I think it's really easy to fool yourself into thinking you're doing, great, important work. And then, you know, you have this grand strategy and you bring it out to the market. And nothing works the way you thought it would. They would tell us, they were like, hey, we really need something and we have looked at, solutions A, B, C, D. And this is all the ways in which they don't work for us at all. Just go to your customers and ask them. There's all these other players in the market. Why haven't you bought one of them? They will tell you your competitive differentiation, why all the other products in the market don't work out for them and what you need to do to be able to differentiate, and be successful. Because all of this ultimately comes out of customer need. They kept telling us the same kinds of problems that they had, right? They were like, this is the problem that I had and which was the same as everybody else. Here are the other solutions I looked at on the market and this is why it doesn't work for me. And that why it doesn't work for me was similar to what everybody else was telling us. And when we showed them our product, they were like, yep, this is great. I'm ready to buy and I'm ready to buy quickly, right? And that's what allowed us to grow revenue so quickly with just two people.

Previous Guests (00:01:20):
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 (00:01:32):
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. Ashwin, welcome to the show, man.

Ashwin Sreenivas (00:01:48):
Yeah, thanks so much for having me. Excited to be here.

Pablo Srugo (00:01:51):
Yeah, man, I'm hyped up for this one. I mean, it's a pretty crazy journey you've been on. Decagon started 2023, so two-ish years ago. You've raised $230 million already, which signals that things must be going pretty insane internally and the crazy thing is, you think about a high level. I mean, it's AI for customer support. Which I would have thought, you know, was kind of this self problem. You've had chatbots for a long time, Drift, this and that. And so many different players over time seemingly trying to solve this problem. And yet clearly you found insane traction. So we'll get into all of that. Maybe as a starting point, tell me just your background. What were you doing the last few years before you decided to start this company?

Ashwin Sreenivas (00:02:30):
Thanks so much for having me, super excited to be here. Yeah, before I started Decagon, I'd actually started another company. Also in the AI space but this was AI for video at the time. Because this was pre-LLMs. So it was a company called Helia, got acquired by Scale in 2020. I was at Palantir for a while before that and then, Stanford before that for undergrad and graduate school.

Pablo Srugo (00:02:52):
Tell me a bit about Helia, what was it exactly? When did you start it? 

Ashwin Sreenivas (00:02:56):
Yeah, so we started Helia with some close friends of mine from undergrad in 2018, early 2019, and at its core. What we did was we kind of processed video data in real time. So at the time, what we did was, we were like, okay, we're looking at these incredible vision models that are out there and it seemed like, one of the only things that they were being used for at the time in production was in self-driving cars. And we're like, okay, what these models are really good at is taking video data, extracting structure from them, right? Oh, hey, this is a person, this is another car, this is drivable road and then doing interesting things with that. And we're like, okay, there's all these workflows within the enterprise that use this as well. Enterprise security was one of those things. You have tons of CCTV data within an enterprise, you have all these enterprises that are worried about security threats, people breaking in and we're like, hey, can we use some of that data and some of those same techniques to help there as well?

Pablo Srugo (00:03:54):
Did you have some sort of security background or was it more the Palantir stuff? The AI piece that put you in this kind of journey?

Ashwin Sreenivas (00:04:00):
No, it wasn't the security background. It was more the AI background of these models are great at taking vast amounts of unstructured data, right? At the time, it's raw images from video feeds and extracting structure from them. And trying to find interesting insights. So, a lot of it was how can we apply these advancements in vision models to enterprise workflows. Which again is very similar to what we did here with Decagon. Which is how can we take these great advancements in underlying LLMs and apply them to enterprise workflows. In this case, enterprise support and customer actions.

Pablo Srugo (00:04:34):
And so you started that in 2018, 2019. You said you sold it to Scale AI, is that right?

Ashwin Sreenivas (00:04:39):
Yeah, at the end of 2020.

Pablo Srugo (00:04:42):
OK, so that was a two year journey?

Ashwin Sreenivas (00:04:44):
It was quick.

Pablo Srugo (00:04:45):
That's quick, man.

Ashwin Sreenivas (00:04:46):
It was pretty crazy times because, you know, start of the pandemic. The whole world was going upside down but, you know, very, very different time than now.

Pablo Srugo (00:04:54):
And what happened like high level? Did you get it into the hands of customers? Did you have traction or was it just acquired more for the team and the tech?

Ashwin Sreenivas (00:05:01):
No, Helia was a much quicker journey for us, right? You know, one of the things that we. I think, really figured out how to do, was figuring out from all this data that we get from customers. It's figuring out how do we understand what data we need to label, to make these models a lot better and, you know, Scale was working on a bunch of similar things at the time. So that was kind of the most interesting part of what we'd built. Of course, you know, it's a super different time than Decogon.

Pablo Srugo (00:05:30):
Did you stay at Scale AI after the acquisition?

Ashwin Sreenivas (00:05:31):
No, I left pretty quickly after. So, both my co-founders were there for quite a long time. So one of them stayed all the way through to the Meta acquisition and went over to Meta. My other co-founder was also there for about four or five years since, and now he's president of Cognition.

Pablo Srugo (00:05:50):
So what did you do between 2020 and starting Decagon in 2023?

Ashwin Sreenivas (00:05:54):
Yeah, spent a lot of time thinking about the next thing to build, really. So worked on a bunch of different ideas and then, Decagon was the one that really, really inflected. There's been a lot of time thinking about, how can I now take the next set of AI improvements and apply them to workflows in the enterprise?

Pablo Srugo (00:06:15):
ChatGPT comes out end of '22. Were you tracking closely all the developments happening in GPT before then? Or how big was ChatGPT of a surprise for you who were building in AI for many years?

Ashwin Sreenivas (00:06:28):
It was actually pretty remarkable at the time it came out, right? Because for the longest time, language models were the kind of second cousin of the vision models that worked really well and people were using it for all these cool things. And this idea that, hey, you can just train these models on tons of internet data and have them talk to you. And generate intelligent sentences was actually a pretty remarkable, surprising, unintuitive result, and so it was pretty crazy when it came out. It was pretty crazy how well it worked. It was both the number of different ways people figured out how to use it, right? Because at the start, it was a chatbot and then people were like, oh, actually, if you take context and add context to the model. This idea of, RAG was also a pretty cool idea, right? Now it's become, you know, actually very quickly in two years become, oh, this is a standard. Everybody knows what in-context learning is. But at the time, it was pretty crazy, right? This idea of you can have the model learn from new information at inference time. This was not something that was possible before kind of in-context learning was figured out. So one, it was crazy how quickly kind of new techniques were layered on top of these models and number two, it was pretty interesting how quickly the underlying models themselves improved. For a long period of time and arguably even today, the rapid pace at which the underlying models themselves improved was pretty astonishing.

Pablo Srugo (00:07:52):
Did that lead to the idea of Decagon or were you already working on AI for customer support before ChatGPT comes out?

Ashwin Sreenivas (00:07:59):
No, that was a big part of why we built Decagon out, right? Because the underlying shift was, wow, this is really transformative breakthrough technology that did not exist before that fundamentally allowed us to take this area that was very important to enterprises and have them adopt this new technology, right? Because this idea of automation in customer support and customer experiences isn't something that's new, right? There have been chatbots for like twenty years at this point. But the interesting thing about that was most of them weren't very good, right? I've used chatbots before and the first thing that I would do is go in and say, hey, agent, agent, agent. Because I know, I'm like, oh, this is probably not going to help me and the thing that was interesting was now you have this technology that could adapt to the things that customers were saying it did have, this ability to learn from an enterprise's own data without needing to go in and train custom models. You got all these like really interesting new properties about these models and that's what actually what kind of opened the window for new companies to be born in this space.

Pablo Srugo (00:09:02):
So that's kind of the rationalized, you know, how it happened. But tell me the specifics of how you ended up deciding, yeah, we're going to build AI for customer support. Where does the idea come from?

Ashwin Sreenivas (00:09:11):
Yeah, so both my co-founder and I were very sort of customer driven in how we approached problems to pick this time.

Pablo Srugo (00:09:20):
By the way, was that true the first time around or was that a learning from that first time?

Ashwin Sreenivas (00:09:24):
That was definitely a learning from the first time around. I think, a lot of first time founders and us included at the time. Spent a lot of time, I think, over intellectualizing the problem by being able to figure out, oh, you know, what is the exact three year strategy at the start? And I need to know what the exact plan is before I start doing stuff. And I think it's really easy to fool yourself into thinking you're doing great, important work. And then, you know, you have this grand strategy and you bring it out to the market. And nothing works the way you thought it would. So this time we're like, hey, let's put all that aside. Let's just talk to customers. This is the only thing that's important. Let's talk to customers and figure out what their problems are. And so we were extremely customer driven this time around. And, you know, we realized, hey, you have this really breakthrough piece of technology. And, even pretty early on we were like customer interactions, operations teams were places where there was a lot of overhead within enterprises and enterprises wanted to do better. And we're like, instead of coming up with this grand roadmap on day one, let's just go talk to a lot of these enterprises. So we talked to over a hundred leaders of operations teams, leaders of support teams, leaders of sales teams.

Pablo Srugo (00:10:30):
With a full blank slate or did you have some ideas you were testing out with them?

Ashwin Sreenivas (00:10:34):
We had a couple of ideas we were testing out with them. Because going in a full blank slate is hard, but we were pretty focused in. I think, reasonably early on, on this idea of, hey, using these language models to help augment or automate parts of ops teams', workflows, customer interaction teams' workflows was something that we were interested in.

Pablo Srugo (00:10:52):
So the highest level idea was really LLMs for the enterprise to drive efficiency. Trying to figure out, okay, where's the biggest pull?

Ashwin Sreenivas (00:10:58):
Correct, but the thing that that wasn't obvious to us at the time. Actually, that was somewhat surprising, was how much of a blank space there was in customer support, right? Because from the outside, like you said, it seems like a space that would have been extremely crowded. But the thing that was interesting to us was the part of the market where, you know, you come in and you were like, hey, I want some basic FAQ stuff. I want to put down a credit card for $500 a month. That space was actually very crowded. However, the enterprise part of the market where you're like, hey, I'm a large enterprise. I have hundreds of integrations that I need into legacy systems.

Pablo Srugo (00:11:34):
What's an example? Just to make it tangible, like a company selling what to who? Who would be a good example of this, of this enterprise that you're talking about?

Ashwin Sreenivas (00:11:41):
One would be a company like Hertz, right? They're a hundred year old business, very, very complex, sophisticated enterprise. They're global and also companies like Oura Ring. Also at scale, very tech forward, much younger company, sells a very sophisticated like tech hardware device. You know, I love my Oura Ring.

Pablo Srugo (00:12:01):
So it's not companies selling to enterprises, it's just that it is a large enterprise and with Hertz, there's obviously a lot of complicated pieces of the product.

Ashwin Sreenivas (00:12:07):
Yes, these are not companies selling to enterprises. These are large enterprises themselves that are very consumer facing.

Pablo Srugo (00:12:13):
Got it.

Ashwin Sreenivas (00:12:14):
Because both of these companies are at scale. They have millions of customers, very complex policies, lots of systems internally that they need to integrate with. So having AI support systems that can kind of handle that complexity That was kind of a wide open space and we realized this, actually by talking to them, right? And they would tell us, they were like, hey, we really need something and we have looked at solutions A, B, C, D. And this is all the ways in which they don't work for us at all. And the interesting thing about that is, this helps that intellectualizing part of that kind of early stage founder. But instead of doing it yourself, just go to your customers and ask them, hey, you know, there's all these other players in the market. Why haven't you bought one of them? They will tell you your competitive differentiation, they will tell you why all the other products in the market don't work out for them and what you need to do to be able to differentiate, and be successful. Because all this ultimately comes out of customer need. They won't necessarily tell you how to solve the problem, that is your job as a founder. But they will tell you what their problems are and what they need fixed. And why existing solutions on the market don't fix it today.

Pablo Srugo (00:13:18):
And the other thing is when you come at it without too strong of an opinion, you get a sense of what's truly a top priority problem. Because a lot of times you come in too specific, you know, with this idea of trying to validate if this is a problem. You might get, yeah, this is a problem but you don't find out what the bigger problem is, that you've never asked about and that's the thing that's really getting budget. I got a bunch of questions about these conversations, because I think this is. This will really set a tone for what you decide to build and how, and all this. First one is you had a hundred of these conversations. What was your way in? What did you even say to these people for them to give you fifteen, thirty minutes of their time?

Ashwin Sreenivas (00:13:48):
Yeah, a lot of this is just early sales hustle in a way, right? It's using whatever connections you have into a company. You know, one of our earliest customers, the way that we got in was my co-founder went to school with a PM there and he asked her for an introduction to their head of operations.

Pablo Srugo (00:14:09):
Under the guise of, I just want advice or I just want to learn more about your job, or I want to sell you something? What was the kind of context you set up?

Ashwin Sreenivas (00:14:15):
Well, I actually, at the time it was legitimately a, hey, we want to learn about, the problems that you're facing and how we can help solve them. So, we were pretty open about that. We were pretty good at using our early investors for introductions into a lot of these companies and also spent a lot of time cold emailing, cold LinkedIning folks just to make sure that. This was a problem that resonated with people, right? If you can cold email someone about something and say, hey, this is a thing that we're working on. Is this a problem you have? We'd love to chat. That's also some amount of validation that this is a real problem.

Pablo Srugo (00:14:47):
The other piece I want to dive into a little bit is, you went to like operations, to sales, into customer support. I assume that there were problems to be solved everywhere but you kind of, you felt something different in customer support. Can you maybe just dive deeper on the comparisons? What you saw there in customer support that you didn't see as much of. Let's say in sales or ops, just to get a sense of what true market need kind of feels like?

Ashwin Sreenivas (00:15:10):
Yeah, it was willingness to pay immediately, right? We with a couple weeks of work, we were at the point where people were like, yes, if you can deploy this thing, I will sign a $150,000 check immediately, right? And this happened repeatedly. This wasn't a one-off thing. So this ability to go from here's what we're building and you know this V1 is going to be ready in two weeks, and it's going to cost you $100,000 a year, and we need that commitment from you upfront. That yes, we'll make sure that A, B, C, D things work but the moment those work you're ready and willing to sign a check for this. That happened repeatedly with the support use case just because the kind of customer pain was so deep and they were ready to buy.

Pablo Srugo (00:15:56):
Did you try that with sales and offset? Did you get to that stage where you're like, okay, what if we did this? Would you pay? And you just kind of felt like, okay, the pull it, you know, these guys will pay. These guys are kind of like, ah, yeah, maybe we'll see.

Ashwin Sreenivas (00:16:06):
Yeah, so we had a number of different ideas that we tried. Where either it was love the ideas is super interesting and then when we went to that, oh, how much would you pay for this conversation. Became, oh, you know, this quarter budgets are tight, maybe.

Pablo Srugo (00:16:19):
Yeah, come back next quarter. Yeah, classic.

Ashwin Sreenivas (00:16:21):
Or it was, oh, yes, this is useful but, maybe I would pay a thousand dollars a month and we really need to pay month to month to see if this works. Whereas for support it was, yes, this is awesome. If you can make it work up $150,000 a year. That's yep, we can do that. Not an issue, right? And so that that kind of stark contrast in willingness to pay. It's just direct signal of how much business pain is there truly.

Pablo Srugo (00:16:44):
And then just so I have the timelines right. When do you like incorporate? When do you raise money? And then when are you having these conversations?

Ashwin Sreenivas (00:16:51):
Yeah, so, both my co-founder and I were second time founders. So we incorporated pretty much right away and actually ended up raising capital pretty much right away.

Pablo Srugo (00:17:01):
This is like beginning of '23?

Ashwin Sreenivas (00:17:03):
It's probably middle of '23. So, we incorporated basically right away and raised capital pretty much right away. And then, started having all these conversations right away.

Pablo Srugo (00:17:11):
And your pitch to the investors was, we're going to find a way to deploy AI in the enterprise. We'll figure it out sort of thing?

Ashwin Sreenivas (00:17:16):
We were actually preempted. We actually, nicely enough, we'd actually never run a full fundraise for Decagon. Because luckily we ended up being preempted at every round. Our first round was led by a16z and a couple of other folks, including Astar. Because we just had very long relationships with all these folks for several years across both our prior companies.

Pablo Srugo (00:17:38):
And that was a $five million dollar round? 

Ashwin Sreenivas (00:17:39):
Yeah, $4.8 I believe, yeah. 

Pablo Srugo (00:17:42):
Tell me just a little bit more about like, take Hertz or any of these other early customers. What it was that whether it's Drift or Ada, or the million other chatbots was just unable to solve for them. Why they felt there was still such a big gap.

Ashwin Sreenivas (00:17:55):
Yeah, so a lot of it was around this ability to handle complex multi-step workflows. I'll give you an example of what that means. Let's use a simple kind of generic example of an e-commerce retailer, right? One way if someone comes in and says, hey, I need a refund, right? It's one thing to give them the FAQ version of how to do that. It's like, oh, go to your account and then click on your order, and then click I want to return my order, and then follow the steps. That's one thing, anybody can do that. However, if you want something more complicated, more personalized to that end user. What it might look like is being able to say, OK, if someone comes in and says, I want to return an order. First, you need to go check this fraud database to make sure this user isn't flagged for fraud. If they're not flagged for fraud, then you need to check the CRM system to see if they are a gold tier member. If they're a gold tier member, you always want to let them execute their return. Otherwise, if they're a silver tier member. Check NetSuite to see if the order was placed within the last fourteen days and if it was. Then offer them a full refund and then hit the UPS API to print them a shipment label. So this idea of I need to work across lots of different systems, execute business logic but also do this in a way that's very conversational with the end user. Doing that was what was hard, right? And as you can imagine, if you can do that well. That really feels like, oh, this feels like there is a human on the other side. This feels very personal, it feels very smooth, that's what was important and that's what none of these other companies could do well.

Pablo Srugo (00:19:25):
Does that mean that you need to set up these kind of rigid, if this then that workflows? I know, especially in e-commerce, there are some players that let you do that and it can get out of hand pretty fast. It's like a PM is trying to figure out, somebody internally or on the support team is trying to figure out like, okay, this case. Okay, you got to do, you know what I mean? How do you?

Ashwin Sreenivas (00:19:43):
No, it's a great question. Effectively, what you need to do is you need to build a product that can do both well, right? So in cases where you want a lot of flexibility. You want to give the model a lot of flexibility to say, hey, just work with the user and find a solution. In cases where you do need a lot of rigidity. You need to have a system where you can constrain it down and say, you know what. For refunds, I don't want you to accidentally give a refund out to someone that's not eligible to it. So in certain cases, you want a lot of that rigid, if this, then that logic. But in, you know, let's say eighty percent of cases. You want the model to be very flexible, very personable with that end customer.

Pablo Srugo (00:20:18):
What was the first version? You mentioned at some point you went out to, you know, even to sales and ops with different ideas. But to customer support with something that they would pay for. What was that first kind of MVP like?

Ashwin Sreenivas (00:20:30):
Oh, actually, for this we ended up needing to build most of this out for the MVP. But, both my co-founder and I were technical. So this was something, you know, of course, it wasn't the full robust platform that we have today. The first version was probably something my co-founder and I built in three weeks for the first set of paying customers, right? And, you know, we were very scrappy in the early days. By which I mean because both of us were technical. I think we probably built up to around, the first million in revenue with just the two of us and that's probably in about six months or so.

Pablo Srugo (00:21:05):
No other employees.

Ashwin Sreenivas (00:21:06):
Yeah, yeah, nobody else. I think we got our first employee, Amy, who's great. Probably around, I think $950k in ARR or something like that.

Pablo Srugo (00:21:14):
What was the reasoning behind that, especially since you kind of figured like, you knew what to build? Why not get at least five engineers or ten engineers in?

Ashwin Sreenivas (00:21:22):
Well, it's kind of because in retrospect. It was the right thing to do and actually, once we settled on that initial idea. This is probably a month in, we did not pivot at all but this is something that was obvious in retrospect, right? At the time the goal was, hey, we want to be very nimble. Up until the time where we're like, hey, we definitely have, we feel like we very likely have product market fit and at the time. It was just the two of us, we were moving fast. It's like call customers during the day, code at night kind of a thing and honestly, a big part of it was also we're spending so much of our time talking to customers and coding that we didn't have as much time to go do recruiting calls.

Pablo Srugo (00:22:04):
Was this also learning from first start? Because that's another difference that I find a lot of times, repeat and first time. You don't realize just how much being nimble and flexible matters. Until you face the time when you're not nimble and you need to change fast, and you're like, oh my god.

Ashwin Sreenivas (00:22:19):
Yeah, yeah, no, exactly. This was something that we kind of learned from the first companies. Luckily for this one, it was a much shorter period of time needing to be agile and nimble in terms of, you know, shifting, completely pivoting the company. But yeah, it was something we were pretty cautious about and it was what led us very rapidly iterate to this idea in the early days.

Pablo Srugo (00:22:42):
So tell me about after three weeks. Just tell me a bit more about what that product did and again, the use case for that product.

Ashwin Sreenivas (00:22:48):
Honestly, it did a lot of what I said before. It had the ability to do this, capture this enterprise complexity and do workflows end to end. And at the time, we didn't have the full self-serve capability. So it was a lot of custom builds that we built out for these enterprise customers and so from there on. Essentially what we were trying to validate was not, can we build this amazing platform like product. Where anybody can do anything in it? That wasn't the goal, right? The goal was to say, if we custom built everything perfectly for one person. Can we give that person a great experience, right? And then once we got our second and third customer, if we were to build something perfectly custom for this person. Can we give them a great experience where they're willing to pay for? And then once you build the first three, you take a step back and say, OK, we can't do this for customers four through ten, right? So what is common amongst these one, two, three customers? And then how do we build that into a great platform?

Pablo Srugo (00:23:44):
We have tens of thousands of people who have followed the show. Are you one of those people? You want to be part of the group. You want to be a part of those tens of thousands of followers. So hit the follow button. One of the things that struck me is, as you said, the amount of complexity that you're handling is the standard way to do it would be to verticalize. To be like, OK, we're just going to do, say, e-commerce and everybody in e-commerce needs returns, and they need coupon codes, and they need. You know, here are things that happen and so we'll just get super deep. But it doesn't sound like you did that.

Ashwin Sreenivas (00:24:15):
No, we didn't and to be clear. There is an, absolutely a world in which that's the right path to take but what we noticed was the problems that someone like Aura and someone like Eventbrite and someone like HearthSpace were very similar, actually. And this was somewhat counterintuitive at the time but we were like, hey, let's just be a completely horizontal platform, right? Within the enterprise, of course, but across enterprise industries. Because the problems that all these large enterprise companies faced were very similar.

Pablo Srugo (00:24:48):
But the way to do that, which is not. I would say, that common is to not worry about scalability, not worry about platform, just go in super custom, deliver the full value. Because this is the problem, if you try to get everything. You try and go horizontally. Try to build platform, you'll have a hard time delivering true value, full value, and then you don't get going. You don't get the customer love that you really need. 

Ashwin Sreenivas (00:25:09):
That's absolutely right, but the way that we kind of got around this was by going very deep with every customer. Not worrying about scalability in the early days and by the early days, I mean months. You know, kind of one through four. We didn't worry about it and months four through twelve, we're like, OK, great. How can we with a more platform like approach give all of these people the exact same experience, right? So at no point did we ask any of our customers to compromise on that fully tailored feeling that they got. The hard challenge for us there was how do we build a platform where you can configure it. To be something that kind of fits like a glove for every single one of our customers and that, of course, is the interesting product challenge.

Pablo Srugo (00:25:52):
And then for those first few customers, how did you structure the sale? Was it a pilot? Because also there's this worry about hallucinations and stuff a lot more back then. How did you kind of make them be willing to take the risk? Because it's pretty critical. At the end of the day, you're serving customers. I mean, if you do a really bad job, there's big impact.

Ashwin Sreenivas (00:26:09):
Yeah, so again, we built out. You know, even in the very early days. We built out a number of controls to do things like detect and catch hallucinations, and prevent them from going out to the end customer. So we built out a lot of that and in terms of the sales process. We structured it like a regular sale, right? So we said, OK, here is the product that we have today. Here are the changes that are coming in the next two to four weeks. We're going to do a pilot and it's going to be for not a very large fee for let's say four weeks. And after that we're going to move into a full annual contract, and you know it's going to cost. Let's say at least a hundred thousand dollars a year or something like that, right? So reasonably meaningful amount of money.

Pablo Srugo (00:26:46):
And it was all pre-signed up front? Because a lot of these things, I mean, they matter in terms of speed. A lot of times startups will get, especially selling enterprise. Which move at a different speed than a startup does. You'll get stuck in like pilot land, for example. Yeah, they love the pilot, but now we're trying to work to commercial. Did you do anything there to streamline that conversion?

Ashwin Sreenivas (00:27:03):
So in a way, this was a part of our kind of pain validation, right? If you get stuck in pilot land forever, in a way, there's a signal that the enterprise doesn't care that much about the problem that you're solving for them. Because if they really care, they want your product in production as quickly as possible. So for us, we actually didn't end up with too many of those problems. Because this was a true, deep, visceral pain that these companies were facing. Because they were like, look, we have a lot of customers reaching out to us. We can't get back to them in time, right? And being able to give our customers a better experience is such a high priority. That they're like, great, if it's people getting stuck in commercials land. Let's kind of bulldoze through that. Let's get to a place where we can deploy this in production, because it works for our customers.

Pablo Srugo (00:27:46):
What was the KPI that you center around? When you think about ROI, a lot of these things are like, oh, cost reduction. Because you need less customer support people, or maybe it's like NPS. But that takes a long time to measure. What did you center in on as the thing that you delivered value on?

Ashwin Sreenivas (00:28:00):
Yeah, it was two things. It was what percentage of conversations can we handle ourselves? And what is the NPS for those conversations, right? Because you can handle, quote on quote, handle a conversation by just frustrating the user, right? And we want to make sure that our customers feel good about the conversation we handle. So we're like, great, we'll show you that we handle eighty percent of conversations ourselves and those percent of conversations have a high NPS score. So that both of us feel good about the fact that we are actually solving the problem for these end customers.

Pablo Srugo (00:28:32):
So you would ask at the end of the chat, rate this conversation one to five sort of thing?

Ashwin Sreenivas (00:28:36):
Exactly.

Pablo Srugo (00:28:37):
And from the perspective of the buyer, is their ROI ultimately mainly around this higher NPS? Is it around having less customer staff? What was the biggest thing for them when they thought about spending this, $100k a year on Decagon?

Ashwin Sreenivas (00:28:50):
Yeah, so it was the way that our customers thought about ROI differed from customer to customer, right? For some people, their business was growing so quickly that they were like, wow, to just keep up with the scale of customers. Which meant an equivalent scale of customer support increase. I'm going to need to double or triple the size of my human support team and they're like, I physically cannot hire enough people to do that. And therefore, if Decagon comes in. I can, you know, pause hiring, instead of doubling my team. I increase it by 10 percent, right? So that was kind of number one. Then number two, which interestingly enough for a lot of customers was also, they would expand the amount of support that they offered, right? So instead of saying, oh, you know, we offer support for only our premium members from 9 a.m. to 5 p.m. Monday through Friday, it became everybody gets support all the time, right? Because there is this significant kind of latent demand for support in a way, like the tiny paper cuts where it's just annoying enough where you're not like, I don't want to dial a number and call someone and have them fix it. And most of these brands were like, hey, let's just make support available to everybody, let's make it easy, let's make it available all the time. Because at the end of the day, that's what leads to happier customers that activate more often and are more engaged, and just get their problems solved immediately.

Pablo Srugo (00:30:07):
By the way, had most of these enterprises tried out and paid for other solutions, and then churned out? Or most of the cases they saw the demos and said, this is just not going to good enough. It's not going to work.

Ashwin Sreenivas (00:30:16):
Now, a lot of them had tried things before. Which is also another signal that we look for when validating ideas. Which was like, hey, this is a pain. I'm going to go try a bunch of solutions on the market. Because this is a real pain that I'm trying to solve and so most of the customers that we talked to back then and even today have either seriously looked at or tried a number of solutions on the market. That's why we were also able to hone in on what is valuable to build. Because our customers were very informed and educated at the time. They would say, hey, we're looking for A, B, C, D and, these solutions provide A and B but the C thing is what's really critical to us. And we haven't been able to find that on the market yet.

Pablo Srugo (00:30:55):
That is a great indicator. Obviously when a company is already spent on trying to solve this problem, clearly it's a problem. The only flip side to that is sometimes you almost have like burnout. Which is the last ten companies have promised me that they could do all this stuff and then when I put them in, they can't do it. So, why are you any different? Did you have some, was there any of that burnout in the market? 

Ashwin Sreenivas (00:31:12):
Not yet, but again. The nice thing about it being a really painful problem is again. The customer really wants to solve it, right? And you're right, there were some customers that are like, hey, I've been burned by this before. Because these other vendors would come and tell me they can do everything. And why are you any different? But then the burden's on us to kind of say, great, you're worried about, these two things. Let us de-risk them for you right away, right? If it's inability to handle really complex logic. We're like, great, let's start in the pilot just doing the complex logic workflows, right? And we'll show you right away that we can handle those well. Because if we can handle the complex ones, well, obviously, we're going to be able to handle the simple ones as well.

Pablo Srugo (00:31:54):
Let's talk a little bit just about competitive dynamics in AI. This is something that I've noticed and you know it's pretty common so it's not like I've noticed something unique but it's really changed pre-Gen AI post-gen AI. I mean competition has always been a thing but I almost feel like pre-Gen AI was a bit of a secondary concern. I mean you're worried about incumbents and systems of record but new entrants coming in was like, if I have enough of an edge, it's just going to be hard. I mean, they could copy me, but I'm always going to kind of be a little bit ahead. With AI, it's just things are changing so fast that you have to worry about foundation models, you've got to worry about incumbents, you've got to worry about new entrants and in this particular thing that you're doing. You know, one of the things would be, well, AI for customer support. I mean, there's going to be so many people trying to do this now that the LLMs can do all the language piece. How do you think about competition? How do you think about that whole piece of it?

Ashwin Sreenivas (00:32:43):
I think in terms of competition, I think the dynamics really differ based on how much of the value you provide comes directly from the foundation models themselves, right? So I think in some spaces where ninety percent of the value provided just from the picking the right model and the model doing the right thing. I think their competitive dynamics are much, much higher. Because again, the only thing that comes down to is, is your model better than the other person's model? I think in the space that we're in, right? Which is customer experiences specifically targeted at the enterprise. Yes, having the best models is really important, right? We spend a lot of our time on taking the best models, fine-tuning our own models in-house. We have a research team in-house as well. But there's also a lot of other things that you need to do, to be able to deploy this well within the enterprise, right? You need really good versioning and rollouts, right? You can't just hit save and have this deployed to a million people. You need really good tests and automation. You need really good insights and analytics, especially if you're having a million conversations a month. No one's going to go read all that, right? What can you learn from what your customers are trying to tell you. So basically there's a lot of software you need to build and this is traditional non AI almost, you know, regular SAS software that you need to build around something like this. To make it work within the enterprise. This idea of, you know, not all the value just comes from having the best model. That's where the actual competitive dynamics change quite a lot.

Pablo Srugo (00:34:07):
Well, I assume in your case, all the natural language processing. If you want to call it that, is from the model, like the chatting piece. But then the actions behind it, the workflows, the if this then that, the business logic. That's the part that has nothing to do, frankly, with the models, I would think.

Ashwin Sreenivas (00:34:21):
Of course, we invest very heavily in our kind of agent orchestration system. In my view, I think we have some of the most sophisticated orchestration systems out there on the market today. But in addition to this, it's all of the other things that we built around it to make it enterprise ready, right? You know, deep integrations with their existing systems, the ability to version rollout, experiment, analyze conversations at scale. All this other software that you need to operationalize it. I think that's what makes it particularly sticky within the enterprise.

Pablo Srugo (00:34:48):
And then what about existing startups that already have distribution? So Ada support is one I keep mentioning just because it's in Toronto and it happened. Actually interviewed Mike on this podcast. He's one of the first PMF show that we did like four years ago. But it doesn't have to be them. I mean, any of these companies that already have tens of millions, if not hundreds of millions in ARR. They already have distribution. They have customers, so they would have a lot of the things around it that you're talking about when it comes to delivering products to the enterprise. But obviously they started pre-Gen AI. How easy is it for them to flick a switch and all of a sudden be like, oh, we're actually Gen AI too. We can do all this conversational stuff.

Ashwin Sreenivas (00:35:20):
It's actually quite difficult because again, like I said, the easy orchestration piece is also quite hard, right? Being able to do simple FAQs, not hard at all. This is just directly you give the model a bunch of stuff and say, hey, what's the right answer? This is like building a chatbot over your data is like the hello world version of AI these days, right? But the hard part is being able to say, OK, when I'm an enterprise scale, I want to follow these complex workflows. Well, like doing that well is actually quite difficult. and I think that's where the initial wedge for Decagon to get built out was created. Because doing that well was hard and I think our kind of agent operating procedure approach seems to be what has won the market.

Pablo Srugo (00:36:03):
So let's talk a little bit about go to market. I think, go to market tactics is something that frankly, every early stage founder really cares about. So, you've got your first three customers, you hire Amy kind of first employee, you start to build out more of a platform. So on the product side, you probably feeling like you're onto something. How do you go out and close the next ten, twenty customers? What's the approach?

Ashwin Sreenivas (00:36:24):
Honestly, a lot of early go to market is the same early sales hustle as finding the people to talk to, to validate your idea, right? Which is find any connection that you have into the companies you want to get to, right? Is it an old classmate that is a co-worker there? Ask them for an intro and leverage your investor network to get introductions to companies, cold email people, ask your existing customers. Actually, existing customers are a great source of this because they are friends with their peers at other companies, right? They've met at conferences and meetups and, you know, they've interviewed for the same position and things like that. So they know each other. So going to them and saying, hey, you know, you're a happy customer of mine. Which of your friends do you think would find something like this interesting and useful? So a lot of it is just doing the, again, unscalable thing and at a certain point. Of course, you have a sales team and a marketing team to generate inbound and, SDRs, and things like that. But up until that point, it is just doing the same things that you did to get those first ten, fifty, a hundred conversations.

Pablo Srugo (00:37:22):
I'm curious if you agree or disagree with this, but one of the things I've found doing these interviews is that. Because I'm talking mainly to founders like yourself who are on a kind of crazy growth trajectory, and I find a lot of founders' perception of these companies is they must have figured something out on the go to market side. They must have figured something out on the distribution side that I haven't cracked yet. My feeling or my learning, let's say, talking to people like you is. For the most part, the stuff they're doing on go to market is pretty vanilla. The thing they've unlocked is insane value delivery. And frankly, insane value just sells itself. I'm curious what you think about that statement.

Ashwin Sreenivas (00:37:59):
Yeah, I think for most companies, that grow really fast is just because they've identified the right pain, right? And they're solving the right set of problems that you could have an amazing perfect go to market motion. And, you know, you've done all of the tricks and all the standard stuff. And you're just doing all that perfectly. But you're just not solving the right problem, where you're solving a problem nobody really cares that much about and you're not going to grow very fast. And conversely, you could just solve a problem that some subset of people care a lot about. And they're willing to pay a lot for, and you can just have the absolute worst go to market motion, and you will inflect very quickly. All of this just comes down to picking the right problem.

Pablo Srugo (00:38:42):
How much did word of mouth referrals play a role, if at all. Through your growth trajectory in the last two years?

Ashwin Sreenivas (00:38:48):
It was, it was important. But we're very enterprise product, right? So unlike a consumer business, it's not that, you know, it's just all word of mouth and, a hundred people sign up the next day kind of a thing. Word of mouth was very important though. Because within the enterprise, trust is really important when making a buying decision. Especially when it's hundreds of thousands or millions of dollars. So being able to point to other happy customers was extremely important for us. Even today, for a lot of our larger deals, they'll say, hey, can I talk to some of your other existing customers?

Pablo Srugo (00:39:19):
What about time to value? That's another area that I've noticed matters a lot more than people realize, especially in enterprise where you know it's going to take a long time. But finding a way to deliver some value as fast as possible tends to have a big impact in terms of conversion rates and speed and all these sort of things. Is that something you think a lot about and have you done anything interesting in terms of being able to deliver some value really fast?

Ashwin Sreenivas (00:39:42):
Yeah so the nice thing about a product like this is it's very obvious when it works, right? And it's very obvious that it can create a lot of value. So we were pretty rigorous even in early sales calls. We'd show up and we'd say hey you know don't worry we already scraped your help center . Anything we could find publicly and we made some example AOPs. Agent operating procedures for what we think some of your common workflows might be. And here, let's go do this together. And oh, you know, we didn't guess this process right. Great, let's change it right now and show you what it feels like, right? So for a customer, it's very easy for them to say, oh, you know what. This actually seems to work and if it works, it's pretty obvious that it can create a lot of value for them.

Pablo Srugo (00:40:28):
I like that, actually. What's an example of something you might have done in a demo? Because obviously just answering an FAQ was not it and yet you're not plugged into the database, you're not plugged into their stuff, so you can't really do a refund. So what would be an example of something you might have put into that kind of a demo?

Ashwin Sreenivas (00:40:40):
Right, you can mock an API call or a database call, or something like that, right? So for instance, when we went and spoke with one of our airline customers, right? Instead of saying, oh, here's an FAQ thing. We're like, hey, let's build out what your reimagined check in experience could look like, right? And so we built a if someone comes in and says, hey, I want to check in and, you know, we'll show them, oh, this is the seat that you currently have on your airplane. And oh, by the way, because of the status. I see you checked in a bag but don't worry, that's not going to cost you anything. And oh, this is because of this status, we can offer you an upgrade here and it's going to cost this many miles. Would you like to do that? We can kind of simulate all these things without actually having to do it. So you can paint a very vivid picture of what it would actually look like in practice instead of showing a pre-cam demo from an industry that has nothing to do with them.

Pablo Srugo (00:41:26):
So you would tailor it but it would be obviously like within a box. But it would still be real AI. It's not like you're just clicking through an interface. So you could still, you know, they could say, ask it this, and then it'll figure out what to do.

Ashwin Sreenivas (00:41:37):
Exactly and we'd tell them, we're like, hey, we don't have access to your databases, obviously and so this is fake data that we made. But the workflow is our best guess at what we think your actual internal workflow is and if we need to change anything on the fly. Let's just change it right now. We'll show you how easy it is to do that.

Pablo Srugo (00:41:51):
How many employees are you today?

Ashwin Sreenivas (00:41:53):
We are about two hundred and twenty-five people today.

Pablo Srugo (00:41:56):
It's insane even just to imagine internally as an organization market pull aside just what it's like to have that many people join a team, you know, that quickly.

Ashwin Sreenivas (00:42:06):
Yeah, yeah and we had to be very thoughtful about maintaining culture as we grew that fast.

Pablo Srugo (00:42:12):
Are you in person?

Ashwin Sreenivas (00:42:13):
We are in person. So we have offices in San Francisco. We have an office in New York. We actually just announced our London office a few weeks ago now, I think. So, yeah, we're an in-person company.

Pablo Srugo (00:42:24):
And then you mentioned you hit a million, about a million ARR six months in. How is that ramped to like $10 million? Because a lot of times, how fast you hit a million matters but how fast you get to ten, you know, tends to matter a lot more.

Ashwin Sreenivas (00:42:35):
Yeah, so we don't share that revenue growth rate publicly but it was quite fast after. So we were able to maintain, actually, like substantially increase the rate at which our revenue grew from one to ten.

Pablo Srugo (00:42:48):
Perfect, well listen let me stop it there and I'll ask the kind of last three questions that we always end on. The first one is when did you feel like you'd found true product market fit.

Ashwin Sreenivas (00:42:58):
I think it was probably around the fifth or sixth customer that we had. Because of two very specific things. One, they kept telling us the same kinds of problems that they had, right? They're like, this is a problem that I had and which was the same as everybody else. Here are the other solutions I looked at on the market and this is why it doesn't work for me. And that why it doesn't work for me was similar to what everybody else was telling us. And when we showed them our product, they were like, yep, this is great. I'm ready to buy and I'm ready to buy quickly, right? And that's what allowed us to grow revenue so quickly with just two people. So, hearing that same thing repeatedly from the fifth, sixth customer in a row in a pretty short period of time, and everybody saying, yep, I'm ready, I'm willing to buy. Let me sign a $100k, $200k check right away. That gave me confidence.

Pablo Srugo (00:43:50):
I'll just go, just a bit of a tangent here but we talked about competition earlier. Has, at that time you were selling a bit into a vacuum in the sense that nobody was able to offer what you were offering. Is that still true today or now you finding yourself like most deals are competitive deals where it's a bit of a bake off and there's other people kind of at the table?

Ashwin Sreenivas (00:44:07):
Yeah, no, I mean, even in the early days. You had other people at the table, right? But a lot of your value is showing, hey, here are the three things that you care about and this is why we're the only people that can do it, right? And what those things are might change over time as the market matures. But for instance, even today, like one of the really valuable things we have. Is this idea of the agent operating procedures, being able to allow non-technical people at your company to, in natural language, build these very complex workflows. So, you know, that has tended to be constant throughout and, over time we kind of add new things that different people care about. So, for instance, several of our customers in more regulated spaces. Care a lot about our testing and simulations feature so that, every day we can simulate hundreds of conversations that they care about and show that, yes, we're still handling these as your internal team requires to handle them. So the kinds of things that customers care about changes over time as the market matures. But a lot of the core things still remain the same.

Pablo Srugo (00:45:04):
And then was there? On the flip side, was there ever a time where you actually felt like things might not work and things would completely fail?

Ashwin Sreenivas (00:45:12):
You know, luckily for this company. We've been growing so quickly and there's just been clear market demand that didn't happen.

Pablo Srugo (00:45:22):
Well, you've built a $1.5 billion company in two years. So yeah, I can't imagine there were too many near-death moments. Hopefully never, but certainly so far. 

Ashwin Sreenivas (00:45:30):
Yeah.

Pablo Srugo (00:45:31):
People are talking a lot about this. At least talking about one person, billion dollar startups, right? You've raised $230 million so far. As you said, you've never actually gone out and done a roadshow. So people are just, you know, effectively preempting you, but you're deciding to take that in. What is your philosophy for deciding to take that money? Are you using it to grow and hire or you just. A lot of it is just sitting there and it's just like, just in case. How do you think about that?

Ashwin Sreenivas (00:45:55):
Yeah, we've actually been very capital efficient in that we've spent very, very, very little of the money that we raise. A lot of the reason to raise capital has been we're building a big team here because, again, really taking advantage of this opportunity requires that we have the best research engineers that can build us the best set of models, the best infra engineer. So we can build something that's scalable to the entire enterprise and secondly, it's about bringing the right partners on board, right? Because we're trying to build a company here that will ideally outlast all of us. And having the right partners on board to help us get through that scale is something that we're looking for as we raise capital.

Pablo Srugo (00:46:38):
Last question, what would be your top piece of advice for an early stage founder that's looking for product market fit?

Ashwin Sreenivas (00:46:45):
The only thing that matters is what customers care about, right? So don't spend any time ideating by yourself in your head. Just go pick some types of buyers, right? Maybe it is I'm going to go talk to every VP of marketing that I can find at a five hundred to a thousand person company and just go ask them what their problems are, right? And then try and figure out what is it that they care about that you're hearing repeatedly that you can build a great product for. Because if these are, you know, reasonably informed buyers. They'll know what's out there in the market and if they still say it's a problem for them. That means they have some unmet need. So instead of guessing ideas yourself, just go talk to real buyers and find out what their problems are. That's going to be the quickest path to part market fit.

Pablo Srugo (00:47:31):
Ashwin, thanks for jumping on the show, man. It's been great having you on.

Ashwin Sreenivas (00:47:34):
Yeah, thanks so much for having me.

Pablo Srugo (00:47:36):
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.