Episode Transcript
So this week, I wanted to talk about how we are thinking about buying generative AI products. So roughly speaking, we have two kind of buckets that we put our bets in. We have safe bets, which are like analytics. I would even put cold DM in that category, screenshot, API sheet, best work cloud toybox. All of these were products that were built. There's no kind of AI magic. They had some revenue when we were buying them. So, in Linux was at five or six Cold DM.
Was at 200. So that was a little closer to adventure Bets, which I'll get to in a second screenshot, API was making about 500 a month. Sheet Best was 800 where clout was over 10,000, Toy bucks was a few thousand as well. So. Our safe bets are what you would expect. Maybe when you look at at what we're trying to do, right? We buy profitable SaaS companies that was XOs original charter. And to a certain extent, it still is.
Path two though is. Interesting because we have all of these new bets. And the reason we're even bucketing these into these two categories is really because we don't have in any real sense a fund. XO is not a fund. It's three dudes investing our own cash. That's really it. So we're limited really into the kinds of companies that we can buy. and, And really what I mean by that is the size of companies that we can buy.
These adventure bets? What I'm putting into this bucket is not necessarily just AI stuff. These would also be more experimental things that don't have any customers yet. Support Kai is an AI first product. I'll get to what we think that means in a second. But another one was sentiment investor where actually we thought there was some there was some paying customers and when we bought it there.
Actually weren't any paying customers. And so it became an adventure bet. But really these adventure bets are. The expected outcome is different than the safe bets. The safe bets success is 5% month over month. We're buying 'em for one x a r r or two X a r r. They're doing some kind of revenue. Some could mean again.
A thousand a month, some could mean 10,000 a month, 20,000 a month, somewhere in that range, those are our safe bets. And we'll put them through our playbook. We have skills people, a team in place to execute on that. And again, we're looking for 5% month over month growth compounded till the end of time.
And will have really great outcomes and have had some good outcomes. Executing on this type of playbook. Generally speaking, we'll hold these forever. We have a few of these namely screenshot and sheet best they're a light burden on engineering and a light burden on customer support.
The problem is there just aren't that many of these in our price range, things get a little more interesting. If we could buy stuff that was in the mid seven figures, maybe 2 million, 3 million, and up then it starts to get pretty interesting. The kinds of things we could buy. A lot of our problems with shared services can go away too, because we would actually have enough revenue.
To staff individual people at the company level, which would be a dream. And we hope to get there, but for now we have a shared services group. So our engineers switch between different products. We have one customer support person and she handles all of the portfolio companies. And this is fine for us for now. We can, we can, we can probably five X to 10 X where we are now on the shared services model. That's fine.
But eventually it would be quite nice to buy something that's doing. Let's say we buy something for about 5 million. It's doing one to 2 million in recurring revenue. That's enough for us to hire. Some full-time dedicated people to that company. So we would have an engineer or two or three that are dedicated to just that one product.
We might play around with the idea of a, a CEO for that one company or a general manager for that one company. And that scales out wonderfully, but we're not there yet. So again, if I could deploy all of our money that we have set aside or earmarked for XO into the safe bets, I would. The, The reality is though you go and look on a acquire. There just really aren't that many of these in our price range that we like, that we feel like uh, would be great buys. So that leaves us with.
Either not buying anything for a while and just kind of being passive and waiting for. More of these safe bets to come around or. We go and we make some higher risk bets that might be. AI products with little to no traction. They might be other products with, or without AI that also have little to no traction.
So we just bought one called support guy, little to no Traction. It is an AI first product, and we're kind of running that through our playbook now, but. tb d on, on whether these types of bets work, the expected outcome, though. Relative to the safe bets is, is zero. It should be zero. We should go into these buying them saying actually we're totally fine.
If this goes to zero, it's a complete write off. But if we make 10 of these bets, Or five of these bets. One or two of them will pop. That is the hope. And that's kind of why I call them adventure bets. cuz it's not quite venture, right. We're not really, that's not quite the game, but it's, it's venture in the sense that the expected outcome, the most likely expected outcome is zero. So they're venture they're adventure bets because venture venture funds themselves are subject to power laws. So a relatively small number of investments will pay back the entire fund. And most of them will go to zero and a few might be. know, kind of mediocre base hits. Whatever, but they're not gonna return the foot. And so that's kind of the model that we are playing with on the adventure Bet side In terms of generative AI. Outside of all the hype. Some of our adventure bets are on AI first product. So again, support guy's. the first entry point into this for us, we have two more under LOI that we're looking at. The core driver of these is the large language model that everybody knows and loves from OpenAI. And either these tools will take off because the tech is ready. They deliver value immediately. We've seen evidence of this in the market. It's why we feel like some of these are goodbyes. However, There's nothing like a paying customer to say, what is or isn't true or is, or isn't real or is, or is not of value.
And so the expected outcome of, of most of these, I think will actually be zero. There might be. A year or two. Of kind of feast where everybody's excited to and open to trying these new products while at their. freshened people's mind. And people are really interested in obviously automating tasks that none of us want to do, or, you know, the idea of, of, you know, removing a, some headcount and replacing that with a software tool that cost 20 bucks a month is really alluring. So people will try these tools, but.
Whether they stick around for a year or two or three or four and continue to use them. It's hard to tell. And so we'll, we have this moment now where there's a bunch of interests. We can get users and get initial paying customers, but there is just no way over the long term that they're gonna stick around unless the tech is actually delivering.
So where's the tech delivering right now. Blog posts, content, writing stuff, being a writing aid, all of that with Jasper, there's a hundred of these tools, maybe a thousand of these kind of writing. Writing type tools that, that people find helpful and pay for and use. I think that that's valuable in there today.
But there are also other. Use cases that it's really sort of up in the air as to whether they can. Complete the job to be done today. We have some time over the next couple of years as people kind of catch up or catch on and be patient with the tech while it's, while it's improving. But the reality is us included all of these products that are using these large language models have the same limits. They all have the same roof.
So the trick is for us to make these bets on stuff that we feel like we could actually deliver on the promise to the customer. Right. We could actually complete the job to be done with the tech as it is today. And maybe a couple, a couple bets where we think it might be in a year or two, but I think that that's really, really hard to predict. So that's sort of how we're thinking about buying these generative AI products. Again, if I had the ability to just go buy half a million dollars worth of these safe products I would, but we just can't wait around any longer for stuff to just come on the marketplace for us.
It's just too slow. We're not going to grow as fast as we want to. If I have to sit around and wait for something cool to come on, a choir that we want to go by. And. So that's why we're making some of these adventure bits. I think we'll do roughly four to eight of these in the next in 2023. So we have two under LOI now. And we'll see where those net out, but I, I believe that we could do between four and eight this year.
And just see how that plays out. We'll continue to make safe bets as they come along. But again, that's a little bit reactive. It's hard for us to, I mean, of course we can go out and try and drum up some deal flow and we do, but it's still a slow, laborious process and they really just don't come up that often again when they do we'll we'll strike and we'll strike fast, but we cannot wait for those.
And then of course, just for fun, I was playing with runway ML, which is getting a lot of. Traction right now, and people are just, you know, obsessed with this tool and it looks just amazing and you watch all the demos and it's incredible. Totally incredible Insane. So I trained it on 22 images of myself and asked it to create a picture of me riding a pony. And this is what it came up with. And in case you can't see just how much this looks like me.
Let me do the zoom crop for, you, for you. And I, I think, yeah. If if, if you can't see this, cuz you're listening right now, check out the blog post. And I think that you will agree with me that we have absolutely reached the singularity. Game over.
