How we think about investing in generative AI

How we think about investing in generative AI
generated with midjourney - adult female pixar character, struggling with a bicep curl with a hilariously large dumbell



Spotify Amazon Apple

We are in the middle of our 9th acquisition (to be announced later). It's not another AI tool although we have 2 of those under LOI right now.

So what's going on here? Does XO buy profitable boring saas companies or pre-revenue experiments with no traction?

Both, for now.

A barbell approach

Safe Bets vs Adventure Bets

A safe bet is a product we buy for 1x or 2x ARR doing some revenue. Some could mean $1k or $10k or more. We consider these safe because we have some history now of buying these and growing them quickly. We have skills, a process, a team, etc in place to execute on these. Success here is 5% MoM growth. We'll hold these forever if they're light on engineering and customer support and continue to grow. We've made a few of these bets. They're lovely. We'll do more of them. The biggest problem is that there just aren't that many of these in our price range.

Adventure bets are venture-like bets in a distribution of expected outcome sense. Venture funds are subject to power laws (a small % of the fund outperforms and generates all the gains while most of the portfolio kind of sucks). The expected outcome of the adventure bets is that most go to zero, but a small number of them are breakout hits. I never wanted to be in the hit-making business but there's no escaping it for the kinds of companies we can buy at the moment. There are lots of these available for < $25k. We can make 5x more bets here than the relatively larger buys that are a little farther along. Again if I could go deploy ~$50ok we have earmarked for 2023 tomorrow into the safe bets, I would. But there are just so few of them!

Our Adventure bets are not all AI bets. Some are just cool little products with no traction yet but perhaps serve some internal function across the portfolio (like acquisition #9) where we feel a lot of other companies could use it and do use a similar product already. Again, we can make 5x more bets in this area. The big question is really about the outcome. Can 5x the amount of adventure bets = 1x a safe bet (assuming availability)? Check back here in a year :)

Generative AI

Hype notwithstanding, some of our "adventure" bets are on AI - first products. Support Guy for example is an AI first product. The core driver is an LLM that ingests documents and creates a custom chat bot for founders to reduce the burden of customer support. I think tools like this will either take off or be abandoned because the tech is either there now or it's not. In certain cases it certainly feels like it's ready for prime time, but nothing like a paying customer to tell you the truth. We will be making a series of bets like this with a wildly different expected outcome than our safe bets. The expected outcome of all of these (or at least the mindset we have to have going into them) is that they will all go to zero. That's what we're prepared for.

It could be good for a year or two and then flatline. Or it could just fail to get off the ground. But either way, if XO is going to accelerate our growth we can't wait around for more "safe" bets to just come along, we have to go out and take some calculated risk.

Out of the $500k ish of dry powder, I'd say we're comfortable putting $100k into these bets. So roughly 4-8 products where there is little to no traction but products where we see a practical use case of the tech.

Runway ML

Here's a little perspective. Runway ML has raised $50M at $500M valuation. $50M (!!?!) dollars have gone into this tool or will in the next year or two...

So for all the talk about how robots are coming for people's jobs, I thought I'd offer a little perspective:

Runway offers the ability to train a portrait model based on your photos. I carefully selected 22 fine ass pictures of myself, trained the model, then asked it to create a picture of me on a pony... this is what it came up with:

Runwayml - why does the horse,the lake, even the grass look so good but my face looks like I was hit with a shovel?

And in case you're unsure of the resemblance let me crop in for you.

AI generated Andrew is not hot.

ladies and gentlemen ... we have reached the singularity. Game Over.