Episode Transcript
[00:00:00] Speaker A: Foreign.
So I know you guys kind of have sort of a unique philosophy around how businesses run. I'd love to hear kind of your thoughts on that. What do you guys believe? It's different than what other people believe.
[00:00:18] Speaker B: Sure. So I think one of the main things we're, we're trying to do, we're calling it the autonomous business. It's kind of the way we're, we're structuring our business and it talks a lot about how to build businesses in the age of AI agents and AI products.
So what makes up an autonomous business? It's two main things. There's autonomous product, which kind of honest get into, but it's everything everyone's talking about. It's an AI agent that you can talk to and that's our main side of our product. But the other side is autonomous employees. And what that means is that at Swann, when we're faced with a problem or something, we want to change or fix or enhance inside our organization.
The question we kind of ask ourselves first is not who can we hire to handle this problem? It's a technological question for us. It's what agent can we build? It's what agent that we already have we can improve or what two agents that we already have just need to talk to each other and that will give us a new unlock of capabilities. It's a lot about practicing what you preach and seeing as we have an AI product, we kind of want to make our organization run like our product.
And what that means effectively is that, you know, our marketing, a lot of it, we don't do any ad spend or anything on marketing. It's all organic. From my co founder, Amos. I Recommend following on LinkedIn, but he's putting a lot of content out there and all of this machine that's getting us all our customers, it's not pure luck. We do have like an army of agents that is running and handling this for us. So all the way from content generation, which is something that runs in a loop, goes over maybe previous posts that Amos has done to kind of come up with new ideas and generate new posts based on that, all the way to tracking engagement. So we have agents that go over every, like every comment, every interaction, every new friend request or sorry, connection request that he gets sent on LinkedIn, we have another agent that goes over that tries to kind of figure out who this person is, if we've interacted with them before, if they're an interesting company for us or not, if it's someone we want to reach out to if it is, then maybe generate some content that we can put back. Maybe if they themselves have a post, maybe the recommended action would be to comment back, maybe would be to send a DM if they're connected on LinkedIn and so on. And you're just examples from the marketing side. But we do have these kind of agents all over the business, from support to things that helped me out as CTO with debugging, kind of like tracking logs and stuff like that. But again, that's a big part of how we're building the business. And it means that we are a small team by definition and every person we bring here is supposed to be the best versions of themselves enhanced with AI and agents. That's kind of how we look at things.
[00:03:10] Speaker A: So it seems like if I could sort of sum up your business philosophy in a sentence, it would be that cyborgs like human AI combo can do the job as well or better than humans.
[00:03:19] Speaker B: Definitely. It's all about human enhancement. I think this whole thing and philosophy actually came like from our product journey. Because what we started off as our very first product, which is not what we're doing today, was kind of a product. When the AISDR round a lot of our mission, I think like a lot of people when AI came out okay, this technology is amazing. It's better than humans. Let's replace all humans and just do everything better than them. And we don't need humans anymore, kind of hit a lot of walls. First of all, people didn't really want to use the product. No one wants to use something that's supposed to replace them. Second, kind of a lot of because it's supposed to replace them, it was a product that was running in the background and the people that were using it didn't have a lot of visibility or insight into how it was running and didn't have the ability to affect the product in any way. And kind of understood one that's not really a world we want to live in. And second, it'll be better for our business and for the people using it. If the product was about human and AI collaboration and not about human replacement. Which kind of led us to the version of Swan we have today, which is all about the agent working together. The product is super built around feedback loops, getting as much feedback back from the user as possible at as many touch points as possible, so one can learn and improve over time. And again, work together with the people using it and not replace it.
[00:04:41] Speaker A: And how do you guys determine the space where you're looking at a problem, let's say within your business. So not within the product, but within the business itself. And how do you determine the space where AI should handle things on this side of the line, but humans should handle this other stuff? And I'll give you some context why I'm asking this.
[00:04:57] Speaker B: So.
[00:04:57] Speaker A: So just this morning I was running through a spreadsheet, basically looking at average order values and so on. So I was trying to pull out some interesting insights about pricing, conversion rates.
I looked at my spreadsheet, it was a whole bunch of data and I kind of had this hypothesis. I was like, okay, I think this is how the data is leaning. Let me pass it to ChatGPT and Claude. And they both come up with the same solution, which was opposite my hypothesis. I was like, that's interesting. I think they're wrong. So I actually made the scatter plot myself and through the trend line and so on, and I was right and they were wrong. So I'm kind of always, on the contrary, inside of the AI thing because I have just found very few use cases in my own workflow where it's like super handy. So I'm curious to see your thoughts, like, how do you guys draw the line between like, okay, this is a task that AI is great at. This is a task that we really need some human guidance.
[00:05:45] Speaker B: So I think a lot of it comes first from experience. You start playing around and stuff like this. I think the next time you will want to do scatter plot or something, if you understand, okay, when I get to this much data or this type of data, I've some experience, I've played around. It just doesn't work.
I think the technology is just at that stage yet, right, where you kind of play around, you see, you feel it, and then you say, okay, it's not good at this stuff, or you learn what didn't work, and then you say, okay, if I just had to give it some more context and explain myself better, then maybe it could understand me. But I think we kind of try to take it case by case. The example that came up to my head when you talked was, for example, our customer support agent, which is an internal support agent we built ourselves.
And there was a, there was a point in time when one of the customers pretty much just blindly asked for like a 50% discount with no reason. And the agent was kind of like, sure, you deserve that. Let me, let me talk to the team and we'll give you that straight away. That was kind of like a point where we had a decision where we had to decide okay, how do we make the agent better? Maybe we don't want it to hallucinate discounts and stuff like that. How can we make the context here better or the prompt or whatever. And we took a completely different direction because we're guided by this north star of human AI collaboration and making us the best version of themselves. We understood that these are just sensitive points in time like billing and pricing and all these kind of anything that's money related is sensitive areas where we just don't want the AI to be involved at all. So kind of the focus became instead of how to making it give a better answer to be which we solve technologically rope to get the customer support engage to just direct all these conversations straight to us as soon as anything AI happens. So I guess I divide my answer into two. There's like a technological side just okay, AI is just not good at this. For example, I don't know counting. There's like a famous example of counting the R's and Strawberry and all these kinds of things where you're just like, okay, technology just doesn't do well because it's a predictive model and it's just not going to be so good at that. And okay, we just can't rely on technology for that. And the other side is more, I guess human based, product based kind of just saying I just don't want even if the AI can do it, I don't want it doing it because I want to be in the loop and I want to do these things myself and I want to be involved.
Does that make sense?
[00:08:00] Speaker A: Yeah, totally. And you mentioned earlier sort of like testing things out and sort of figuring out the boundaries of AI.
Somewhat related to this. On the topic of testing things out, I would love to hear how you guys came up with the idea of Swan. How did Swan come to life? From idea to testing to execution to actually having a real product.
[00:08:18] Speaker B: That's a good question.
I'll try to not just say the long story because me and my founder have been together for like 10 years. Four different companies, five different companies, depending how you count. And everything is kind of like a big evolution of everything. But it mainly started when we sold our last company. It wasn't life changing, but we sold our company. It was in the NFT space, totally unrelated and AI was just becoming like a more of a mainstream thing. ChatGPT just came out. It was obvious this like a groundbreaking technology and we kind of just Doug said we want to build something there. There's obviously Value to create and things to do. And we played around for a while. A lot of ideation, a lot of kind of like these MVPs that were up for like two weeks and then taking them down and stuff. Like we have on YouTube a video of us demoing Lovable. We built Lovable like, I don't know, two and a half years ago. I think it was something like very, very similar, but just the technology wasn't there yet. And so we kind of ditched it. And then we ended up in the world of sales.
I think a lot of people, it was like just very natural because of the s. The SDR work. It's kind of very repetitive. A lot of writing emails and stuff like that, which technology at its base was meant to be very, very good for. So we kind of ended up there. It is a. There's a lot of companies there. Again, because you don't have to have like too much know how in this space to understand. But we did that and that's how we came to that solution, which I previously mentioned about kind of trying to replace people and not empowering them. We hit a lot of walls. People didn't enjoy them. We had some customers, some traction. People didn't really enjoy the product too much and it was kind of hard to get people to adopt it. And then this whole idea and this whole notion, because we were building our internal tools like for human and AI collaboration, but our own product wasn't. Wasn't built for that.
And then kind of that evolved and then we understood that generic sales agents that kind of just spray and pray and do this generic outbound don't really have a future. The market is really saturated. It's hard to get people to open emails and LinkedIn and so on. And so we said we need to figure out two things. How to reach out at the right time and how to reach out in the right channel, which kind of led us to the next version, which was identifying your website visitors said if we have a good, good warm, outbound signal, like someone visiting your website and leaving, then we can use that to engage people and probably catch them at a better time to get responses. And if we reach out on LinkedIn and on email, which is just a bit of a less flooded channel, then we'll again increase our chances of getting those replies. That's how we started out. That's what made our business kind of explode. And on LinkedIn and everything in the autonomous business and all that talk and the current evolution of our product came about from people seeing What Swan can do, which was pretty much taking this website visitor signal, researching those companies that are visiting your website, understanding if they're an ICP fit or not, finding relevant people in the organization to reach out to and then doing automated LinkedIn campaigns to those relevant Personas in the organization to your buying committee pretty much. And then what happened was people just said okay, this is cool, we want to do more things with it. Like why can I only do this on my website visitors? These capabilities of research and outreach on kind of classification of understanding who the company is and finding relevant people in the company. Super powerful and want to do this for more things. And that's how Swan today was born. Just general GTM engineer that can pretty much take any idea you have, any sales related idea you have, you can describe it in plain English and swam executes it for you.
[00:11:55] Speaker A: And what kind of challenges were you guys facing in the early days?
[00:11:58] Speaker B: So I think a lot of it is a technological challenge. You know the just stuff we have today, just less because the models have improved. But when we started out writing an email for example, something that sounds very simple but maybe if it's super tailored to your use case and you have something super figured out, it's easy even then I don't think it's so easy. But getting for example the technology to output a good email generically across multiple organizations, all very different organizations, sometimes different sellers have like different ways they want to write and you kind of have to adapt your product and technology to do the, the writing in a good way for each individual seller in the organization.
That was very hard since mainly the technology was hard to get it to mold it to each person.
In that sense something that's much improved today because the technology has improved technologies like memory. So the agent being able to remember different things about you, your writing style, how you like things to be done. Again today it's not just writing. It's everything Swan does. It can remember and act upon. But it just wasn't really that figured out in the beginning. It was more, you know, it wasn't that different from normal SaaS I guess was an LLN is stuck in different, different places. The big unlock was agentic works for us which is. I don't know if you and I can elaborate on what that means, but that's. That was the big, big unlock for us which kind of made everything click nice.
[00:13:24] Speaker A: And I think kind of looking back to this early time I'd be interested to hear what's the decision you guys made that you're really glad that you made early on or in other language, what's an investment you've made that just like continued to pay off.
[00:13:37] Speaker B: I think one of the main things we understood quickly and saw was that I said it was a hurdle, like the technology wasn't there yet. But on the other side of that was understanding that every two months everything is changing. It's becoming 50% cheaper. All it's like every few months the models are like 50% cheaper, 50% faster, a lot smarter. So we were kind of okay with the understanding that the product might not be perfect. Now like the LLMs part of it, it's not perfect, but that's okay because it's going to get better so much faster. But we should build around and build around the assumption that OpenAI releases an update and suddenly our product just gets that much better. That's not something that happens with products.
One of your vendors just releases an update and selling your product goes from good to amazing. That's where we're at today. And I think that was one of the big realizations that it's okay, doesn't have to be perfect yet because just we're relying on external vendors and the switching cost is zero to go from anthropic to go to GPT to go to Google's models. But someone just has to release a better model and it's going to keep happening and that's going to be a compounding, that's going to have a compounding effect that's definitely paying off today. We built the core around this technology that we knew was just going to get exponentially better constantly.
[00:14:54] Speaker A: And when you have a technology that's sort of so centered around your vendors, how do you guys think about building a moat?
What are you putting in place as far as product or marketing or just team efficiency or whatever to kind of keep you ahead of competition.
[00:15:11] Speaker B: So I think one of the things that it really is a mode today is, is brand. When there's like a ton, you're flooded AI fatigue like you said and everything. I think people kind of want a brand they trust and get to a product that they feel that it's okay and I'm not talk to it and it's going to break after a second and they feel okay putting their data in. That's one thing.
We call it more of a softer mode. And I think real, more data oriented mode is this concept of memory.
So today Swan, you talk to Swan and everything you say, Swan remembers, you know, your definition for your organization and how you sell and how you can. What makes up your competitors and what makes up your ICP and what makes up your writing style. All this kind of stuff, which is the easy thing is getting something like out of the box that happens in your onboarding. You sign in, we understand your domain, we build this kind of profile around you. What's not easy is the nuances and the things that change over time. And actually, that's not really our Persona. And actually, when we reach out to this person in this kind of organization, the emails should be like this. And when we sell in Europe and the emails are in German. Well, German has to be really formal. So we want to write out our emails like this and kind of build up this huge memory, a contextual memory about an organization. And that's a real note. When someone tomorrow switches to one of our competitors or not, not going to have all that data available to the products and models that are using, they're interacting with.
[00:16:33] Speaker A: Yeah, totally true. Proprietary data is.
That's. I think it'd be very powerful for the future. And I really see a lot of businesses kind of leaning on that as they're remote right now in an AI world, which makes sense. Like, if you have these machines that are just like, getting better and better and better at everything, which I'm skeptical of, by the way. But if that story is true, you have things that are getting better and better and better. Like one day they'll be better than humans at pretty much everything.
[00:16:57] Speaker B: That's what you're skeptical about, or you're skeptical about something else?
[00:17:00] Speaker A: I'm skeptical at the pace that they're saying it's happening and that it's significantly better than most humans at the moment.
[00:17:07] Speaker B: No, I agree with that. No, with you.
[00:17:09] Speaker A: So I'm skeptical of the pace and I'm skeptical of current capabilities.
[00:17:12] Speaker B: I get that.
[00:17:14] Speaker A: But yeah, I mean, in that world, if that is true, then data really is quite powerful. So I'm kind of curious what data you guys are prioritizing most as far as sort of building this mode or differentiating you guys in the future.
[00:17:26] Speaker B: Yeah. So I kind of touched about it. It is the main one. It's this contextual memory that we're building. I think that definitely is the biggest thing we're focusing on.
Mode is an afterthought, to be honest. We're not building it because we said, okay, if we have this, we're going to have a mode. It just makes the product a lot better. So if I can remember everything. If the product remembers everything you tell it and knows how to retrieve it in the relevant context. That just makes the product much better to use. You kind of want this memory data all architecture to feel seamless for the user. User just wants to say, when I write emails, I like them short. He can say it in some random context that's not really relevant to anything, but the product should be able to take that, save it and then understand later on where that's relevant and where he should pull that out and when he should use that and when it's not relevant. Which I think is kind of part of our core technology because a lot of parts today have memory but sometimes, you know, he remembers something that's just not relevant and that messes up some other flow.
So it's kind of hard to get that, get that. Right.
[00:18:30] Speaker A: Totally true. And I think switching gears for a minute, I would love to hear sort of more of your thoughts around business philosophy. So how are you guys creating an environment around allowing people to do their best work and however you want to define that.
[00:18:43] Speaker B: So I think one is subscriptions to everything, you know, that we don't say no to anything. You want to get Claude, Perplexity, chatgpt, whatever, go ahead, play with it.
We have, I think we're using all of them like the Zapier agent builder. And it kind of depends the role, right. Because like a scale, so n 8 n for example is super technical.
Kind of. Some people will use that if they know how to code a little bit maybe and build their agents there.
Others will use Zapier, which is kind of a more user friendly, less technical way of building agents. And I guess it's just kind of pushing people to, we're driving that as the founders of the company. But I think it's just so baked into our mentality that whatever problem comes up, we really do ask the question how can we solve this with any agent? And obviously it's not, but it doesn't, it doesn't solve everything. But that's the first thing we're trying to ask with for every problem that comes up.
[00:19:37] Speaker A: And you mentioned earlier that you're sort of asking this question instead of asking how do we hire a human who can do this? So what are your guys philosophies and thoughts around building a team in general?
Ideal team size.
What is your current team size and why have you kept it that way?
[00:19:53] Speaker B: So to be honest, we're just the three founders at the moment and we're managing this, you know, nearly 200 business, business paying customers and we're managing to do this ourselves, we're at an inflection point. I think it's more of a feeling. I don't have like a hard metric or anything to give you, but, you know, kind of feels, you kind of feel when, okay, I'm supposed to be doing mainly building the product because I'm the CTO and I'm writing all the code. But if we reach a point where, okay, we're doing support to a level, we're ANSWERING Currently around 80%, I think maybe around 85% of our tickets are being answered automatically.
But, okay, if the business is growing so much that that 15% left over is becoming too much for even us to split between us, like the requests that are going or, you know, like I said, for example, pricing stuff, runner, all the kind of stuff that the agent can handle themselves, technical stuff that comes up, then, okay, we'll, we'll say who's the next person that we're bringing here? But the people that we are looking to bring, and we kind of started looking around for some people, maybe someone in engineering, someone to help out with marketing, really looking for generalists, someone who in their mentality are adaptable and are able to say, okay, I'm all around, I can do a lot of things.
I'm coming in for one area of the product. I like this philosophy of building out a ton of automated agentic workflows to help me out, do everything.
And that's kind of the people we're looking to bring here. Just because I think that's the business that's worth building today. It gives you an edge in today's world.
[00:21:25] Speaker A: And I think you touched on this a little bit. But when you guys will be looking at hiring in the future, I'm going to claim that into being for you. But when you guys are looking at hiring in the future, when you guys grow, what are some of the roles that you think are probably first on your hiring list?
[00:21:42] Speaker B: So I think one of the main things, because again, I think today engineering seeing this as a cto, but I think go to market, for example, is much more important today and then engine to start off with. So I think one of the main things we're going to want to expand is kind of our community work and super user work.
It's all together, but kind of building a community of users who like our product and building a community of super users who are fans of the product and kind of get the word about Swan out there and maybe bring some in to take that end to end. So talk with, talk with Users and find those fans and kind of build up programs and things for them to scale that and get the word about Swan out. And the second one, it will probably be engineering. There's a limit to how far I can take the product by myself. I think we're getting there.
We'll get there soon. I think so. Definitely someone like that. But again it's not going to be someone like the profile of the person that we're going to be looking for is not going to be someone who has just been.
Their last role was a senior engineer, for example. Ideal profile. Probably someone who was an executive recently or wants to get their hands dirty recently. Wants to get their hands dirty now, I mean and kind of write code, but can see the big picture, take things from end to end, but again has to be do a lot of hands on. We don't have the privilege because we're looking to stay small. We don't have the privilege of saying like okay, this guy is newly there, but okay, we're hiring like 30 people so it'll be okay. Like we'll hire 15 more and everything will be fine. Kind of have to get those unicorns to take the ride with us.
[00:23:20] Speaker A: And as far as taking the ride, what do you guys think is next versus one? Where do you want to take it from here?
[00:23:26] Speaker B: Good question. So I think we're building out this GTM engineer that can do. You know, you tell people and you see people because it's an agent. So it's amazing the feedback you get from the product itself because people are talking to it and just asking it to do a ton of things.
And I guess we kind of want to improve on the capabilities and the things that Swarm can do. It's kind of built around these main sales use cases today in terms of the systems that it's connected to.
So many like CRM, Slack, email outreach, LinkedIn, but there's a ton of other systems that we want to connect to and build and kind of make Swan more, even more of a, of a generalist.
So I guess that's the, that's the main direction we're heading.
[00:24:10] Speaker A: Amazing, man. Well, any questions that I should have asked you but I didn't before we sign off?
[00:24:16] Speaker B: Good question. No, I don't think so. Nothing that comes to mind.
[00:24:19] Speaker A: Well, great. Thank you so much for your time. People want to learn more about you or Swan. Where should they go?
[00:24:25] Speaker B: So definitely LinkedIn. Check us out.
Follow my co founder.
Maybe I'll start posting soon, but follow my co founder Amos on LinkedIn. Definitely he's putting great content out there on gets1.com.
[00:24:36] Speaker A: Awesome. Thanks so much.
[00:24:37] Speaker B: Thank you, Brady. Have a good time.
