00:13:36 · S1, Brendan Foody
[00:13:36] S1: And repetitive processes. [00:13:37] S0: Exactly. Repetitive processes. And the way I think about our product over a long time horizon is that there's three big things that we need to get right. We know that there's a lot that will change, but for the duration of the business, our customers will always want more candidates available on our platform. They'll always want better matching with those candidates, and they'll always wanna be able to hire more quickly. [00:13:58] S1: How does matching work? [00:14:00] S0: Matching is [00:14:01] S1: That's a tough one. [00:14:02] S0: A very... The most challenging problem because there's sort of two sides to it, both of which are very high dimensional. The first side is is the candidate interested in the job. Right? And so there's all the characteristics of the role, the people they'd be working with, the compensation of that role, the other opportunities they have present. And so we have models that predict the probability that said candidate is interested in the job. The other side of it is from the company perspective of what is the probability that said candidate will perform well on the job once they're hired. And so you need to collect all of the performance data as your eval set to make better predictions around what are the types of interview responses and the characteristics on people's resume and all of that that predict that they're going to do well on a particular job. And it's the blend of those two problems that really go into matching effectively.
00:14:02 · Brendan Foody, S1
[00:14:02] S0: A very... The most challenging problem because there's sort of two sides to it, both of which are very high dimensional. The first side is is the candidate interested in the job. Right? And so there's all the characteristics of the role, the people they'd be working with, the compensation of that role, the other opportunities they have present. And so we have models that predict the probability that said candidate is interested in the job. The other side of it is from the company perspective of what is the probability that said candidate will perform well on the job once they're hired. And so you need to collect all of the performance data as your eval set to make better predictions around what are the types of interview responses and the characteristics on people's resume and all of that that predict that they're going to do well on a particular job. And it's the blend of those two problems that really go into matching effectively. [00:14:57] S1: I love that. Something I've always struggled with is the bias that creeps into the human decision making that is part of the recruiting process. Walk me through how you think about that within your technology.
00:14:57 · S1, Brendan Foody
[00:14:57] S1: I love that. Something I've always struggled with is the bias that creeps into the human decision making that is part of the recruiting process. Walk me through how you think about that within your technology. [00:15:08] S0: For sure. I I mean, I find it so fascinating because so much of the way people hire is very vibes based. Mhmm. Right? Of, like, does this person seem like I get along well with them? And it's not data driven around who's gonna perform well on the job and what are all of the features that actually prove that that person is more likely to succeed. And so I think the first answer is grounding everything in performance data. Obviously, you need to be thoughtful around how you're evaluating performance and thoughtful about that. But so much as you can do that, it's the most effective way to make decisions and optimize for the right thing that you should be hiring for. And I think the second thing is that in automated systems, you can get very thoughtful about the features that you consider. One example of that is that when humans are interviewing, we consider all of the vibes. When our AI interviewer is talking to people and evaluating interviews, it can just look at the transcript, the substance of what people say. Right? Leaving out all of the, like, multimodal aspects of whether that person sounds like us or looks like us and all of that. [00:16:17] S1: That's fascinating. What about nuance? I mean, does that matter?
00:16:17 · S1, Brendan Foody
[00:16:17] S1: That's fascinating. What about nuance? I mean, does that matter? [00:16:21] S0: It does. I mean, a big part of who we hire is like the vibes matter in some ways. Right? Like matter. Yeah. I... And so there's a little bit of a balance there. I I think that our our mission as a platform is very much to avoid that bias and to have as much transparency as possible, around who we think is going to be as successful as possible in the particular role. But at the same time, a lot of the time humans will make the final final call. And there are some things that they should consider that they shouldn't. And there's others that, you know, it's the way it is and the people that that they get along well with. [00:17:03] S1: I think that's really meaningful. You know, you're, you're diversifying the top of the funnel so that when you get to your final few candidates, can apply that level of judgment, taste, perspective that is really valuable, I think when you're making that final decision. [00:17:18] S0: For sure. [00:17:19] S1: So what can you share about your fundraising history?
00:17:19 · S1, Brendan Foody
[00:17:19] S1: So what can you share about your fundraising history? [00:17:23] S0: Let's see. So we, started the business January 2023. We bootstrapped it to a million dollar revenue run rate before we dropped out of college. And then, and our seed round was with General Catalyst. And then we were really focused just on, like, product and not scaling up as much for the subsequent, call it, six months or so. And then we saw this huge transition underway in the human data market, where it used to be this crowdsourcing problem of how do you find a bunch of low and medium skilled people that are writing barely grammatically correct sentences for the early versions of ChatGPT that everyone is familiar with Scale AI solving. But we realized that it was transitioning away from crowdsourcing towards this sourcing and vetting problem of how do you find some of the most exceptional people in the world that can work directly with AI researchers to help push the frontier of model capabilities? And the platform that we've been building for the prior year and a half was perfectly positioned for it. And so we went [00:18:24] S1: place. To Right time. Right products.
00:18:24 · S1, Brendan Foody
[00:18:24] S1: place. To Right time. Right products. [00:18:26] S0: Yeah. There... There's a lot there's a lot of fortune and find... And luck in finding the right market. And so we went to, OpenAI and we shared our vision of how the world was gonna change. And they, along with all of the other top AI labs, correspondingly brought us in. And, I remember Benchmark invested at a $250,000,000 valuation for 10% dilution when the business was at 1,500,000 in revenue. And we were like, no, this is such a discount because we know that the business is about to inflect. They thought we were absolutely crazy at the time. And then our series b was four months later, at $4,000,000,000 valuation, but a lower revenue multiple, at a $20,000,000 revenue run rate, because the business was growing so quickly, and have obviously grown another order of magnitude since then. And so it's been certainly beyond our wildest expectations. [00:19:28] S1: Yeah. I can imagine. So everything has been butterflies and rainbows? [00:19:32] S0: Well, not exactly. Right? Like, there's... [00:19:34] S1: Let's talk about some of the hard stuff.