Speaker unverified: evaluation

Transcript context

…rd. I think what Eoghan and then Darragh came up with was just like, “Hey, let's just charge per literal resolution, like every time we do work, we charge them. When we don't, we don't.” It was a bet. Yes, it was definitely a bet. We made the decision, and I think the market responded really well, because I think it was very clear that the single statement of, “We only get paid when we do work; we don't get paid when we don't do work,” it's from the same vein as the guarantee—which is just like, that's how you know that we believe in our product and that's how our product works. Yeah, how much are the humans costing you? Yeah, exactly. And how much is their office costing you and all the other stuff. But people like certainty. How do you get them okay with the variable component? Obviously you can contract out whatever you want, but what we offered people is, hey, most of the time people have at least one or two years look back on, with customers who spike for tax season or customers who spike for Christmas or whatever, and we can basically say, “Hey, let's contract your base rate and let's talk about overages for the months you need it.” And that totally works. What we're basically saying is, “Yes, you don't have predictability in the sense of it not being fixed, but you can model it based on what's happened in your history,” and it's only really brand new startups that don't have a clue what's going to happen, but they're not usually worried about this. So you're using a relatively new product, Stripe usage-based billing, for this. How has that—you migrated from Zuora for that. How has that process been? Yeah, I mean, I would say just to go back to your earlier, we afforded ourselves too much complexity. We kind of codified that complexity in Zuora. I guess the best way to describe it is we just twisted ourselves in knots, and it got to a place where we actually, we ended up like Ciaran Lee, who was our CTO here, he ended up actually returning to the company with one mission, which was like, “I am going to, one, Fix billing. Yeah, exactly. To fix this, right? And it worked, right, but it was a substantial amount of work to unwind so much and then to deconstruct so many of these à la carte Microsoft Word style deals into something that was a go forward, acceptable whatever, and then obviously moving towards a clean, transparent seat-based pricing, and then just layering on a usage base on top was actually pretty simple in the greater scheme of things. And all the stuff that we needed, you guys were ahead of us on discounts for volume, et cetera, all the sort of obvious stuff people would push for. ally pretty simple in the greater scheme of things. And all the stuff that we needed, you guys were ahead of us on discounts for volume, et cetera, all the sort of obvious stuff people would push for. Yeah. Is this just where pricing in this new world goes? Because obviously no one buys labor on an unlimited basis, and at least for the moment, the inputs of AI do actually scale with usage for a significant basis. And so it feels like you have to have some usage-based pricing. This is certainly the bet we are making where again, the reason that billing, the top thing they're thinking about is making billing work well in the usage-based world is it just feels like many products are becoming much more expensive to serve and therefore have to have a usage-based component. But is this permanent or I dunno, does the AI get cheap enough that maybe we go back to unlimited plans or I dunno, I dunno if unlimited plans will ever, well, I dunno. Here's how I think about it. I think ultimately all AI has two vectors, there’s how much work you are doing and how well are you doing it? Yes. And the volume of work you're doing, it's almost, well actually both of 'em are going to be proportional to how many tokens you're burning or whatever. So you're going to want to factor that in, especially if you're a consumer app as well, we're going to go nuts. So I think you have to have some, I'm not a fan of cost plus pricing, but it does place a kind of lower bound on what you can do here, which is just like, hey, unlike SaaS, you are actually sending money out the back door as well. So I think you have to have something that's proportionate to how much work you're doing. And then I think aside from that, you have to charge consistent with how much work are you displacing. I think that's where you can say, hey, for us anyway, if you take an average person who sits in a seat to do customer service, if they do, let's just say they do 20 conversations a day, that's what 400 conversations a month, when we were thinking about how we charge, we're like, hey, well if that person does 400 a month, Fin does 65% of that seat, we're still up. We're only charging, whatever, $90 for the seat. So from our point of view, it was an obvious, an easy swap. I think for a lot of businesses it might not be if your AI doesn't work or it's spurious or its value can't be articulated. Isn't it cool that you can now dynamically summarize a GitHub issue or something like that? You're like, cool, I don't know how much people will pay for that. They don't know either. Or, hey, you can now generate random graphics in your newsletter tool. You're like, okay… It's like vitamins versus painkillers, AI pricing. how much people will pay for that. They don't know either. Or, hey, you can now generate random graphics in your newsletter tool. You're like, okay… It's like vitamins versus painkillers, AI pricing. And it's specifically in this case, the painkillers are very strict, if we don't do it, a human's going to do it and we know exactly what they cost, and the vitamins don't have anything approximating that. So not only is it nice to have, it's like I don't even know what it's worth. I saw a while ago someone said when Studio Ghibli came out and everyone was using that, someone said, hey, the fiverr.com equivalent of all these things would've been trillions of dollars. You're like, right, but no one was ever going to spend that. So there's no sane way to actually talk about what actually happens here. I think it was Byrne Hobart who said that when you're tied to business outcome, that business outcome is usually done by humans. I think it's going to be really, really easy to make a business case for saying swap this over to AI. It's better, faster, cheaper. I think when your AI is not tied to business impact or is debatable in quality or whatever, I think you end up with these people who are just like, oh, let's just stick a tenner on the seat and see what happens. So it's like you can have a normal seat or an AI seat and then you're kind of like, I hope no one uses the AI too much. You're permitting yourself to build weak AI stuff if you do that because you're not pushing yourselves to say, hey, we need to articulate the value of each incremental usage here. Well, when you talk about this AI pricing dynamic, one thing that really strikes me is just how fast AI companies grow from a revenue perspective. So I just saw Mati from ElevenLabs. We actually had a great session at our customer event in London, but he tweeted that they've just passed 200 million in ARR and that's 2 years after founding it, maybe 3 years after founding. But in my day, businesses didn't do that. And it's really striking for me how somehow they seem to climb the revenue ramps much quicker. I know, I mean… You guys would Fin is another example… Yeah, for sure. We forecast like Fin will be 100 million probably early next year or whatever and back at… Yeah, from when? Starting from… I dunno, probably about two years, something like that. Yeah. So two years to 100million in ARR… When we started and probably when you guys started it was like that was the threshold to go public. Exactly. It used to take a long time to get to 100 million in ARR. It was like seven years. That's the simplest AI investing framework I've heard. I'll tell you why it's simple because you're going to make me write no checks. So I guess I'd say most of the AI companies I’ve invested in probably three or four. Three of the four I'd say.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.