E79: Now at $2B Valuation: Mercor CEO On How They Started

0:02Welcome back to Turpentine VC, the podcast where we discuss the art and science of building successful venture firms. VC to VC. This week, we're airing something a bit different in light of a significant$2 billion valuation announcement from Mercore, backed by Felicis, GC, and Benchmark. Six months ago, Turpentine's show, 1 to 100 host Usman Hanif interviewed Brendan Foodie, Mercore's CEO. At that point, the company had grown from zero to mid seven-figure revenue in just one year. This interview is a glimpse into the foundations on which the company is scaling and the massive market opportunity Mercore is going after.

0:37Please enjoy and check out 1 to 100, which also features similar interviews with the CEOs of Braintrust, Foundry, Clay, and Modal.

0:48Brendan, thank you so much for doing this. Thank you so much for having me, Isman. I'm excited about the conversation and have listened to a few of your episodes, so thrilled to have a conversation. Awesome. Well, for starters, you guys have gone from zero to mid-seven figures in revenue in a year. Clearly, you have something that people want. For listeners who may be unfamiliar with what that is, I'd love for you to explain what is it that you guys have built that people are dying for? Yeah, so a basic way of thinking about the business is that there's this huge industry around staffing where people will manually review resumes, manually conduct interviews, and manually match people with jobs.

1:29And Mercor builds and trains models that automate all of those processes. reviewing resumes, just like how humans can go about doing that, joining a Zoom room and conducting an interview that's far more in-depth than a human is even able to do. And our most important technology is in facilitating the matching process, where we train very large models that understand all of the characteristics that predict why someone is likely to do well at a given job or task. And an interesting application of this technology is that many of our clients are hiring high volumes of expert contractors to build out human data teams, leveraging all of this infrastructure to better find the best experts in the world.

2:14I see. And so concretely, if someone was to log on to Mercore's website, what does the experience look like from the candidate's perspective and then from the perspective of the employers? Yeah, so they can go on to work.mercore.com as a candidate, where they upload their resume. They'll join an interview where they can talk to an AI that will ask them questions about their projects, about the nuances of their work experience. They'll get feedback on those and then get potentially matched with a job. On the client side, they can go to team.mercore.com and they can query for any role that they want to hire for, where we have an AI-powered search and evaluation system that will recommend the candidates that have applied to our telepool and are the best fit.

3:01So we have over 300 ,000 candidates that have gone through that process on the applicant side and can surface all of the best profiles for employers and companies that are looking to hire. Awesome. And so what do you think it is about your guys' experience that makes it PenX better than, say, other competition or what companies are used to doing in the status quo? Is it speed, quality, user experience? You guys have a good amount of traction and some large companies using you. Why are they picking you? Yeah, so the dimensions that customers tend to care about the most are speed, quality of the candidate that you're surfacing, and then also the price.

3:44And all of these dimensions are usually constrained by manual processes. When you're hiring a software engineer remotely, you might need to conduct a dozen in interviews to find a candidate that's good. And even then, it might have not been a good pool to start with. And the candidate might not have been exactly what you were looking for. But when all of a sudden you're able to leverage technology to assess and interview hundreds of thousands of people at the cost of software, you're able to find candidates that are far more relevant. You're able to have candidates that will start immediately tomorrow and achieve that all of a cost structure that is dramatically lower than what people ordinarily find with human processes.

4:28And so when we're talking human processes, you're referring to actually hiring recruiters who sit and, you know, browse through different talent networks, say LinkedIn and reach out, do initial phone screens and interviews. Is that what you're referring to? Yeah, that's part of it, but also even internal human processes, right? Like most founders in building a company and hiring people have to go through a huge volume of manually reviewing resumes and manually conducting interviews. And certain elements of that are awesome and need to be maintained, such as the human connection of making sure that you really want to work with someone.

5:10But the unfortunate consequence is that lots of people that are an extremely good fit for a given role might not be considered. And there's not the right sort of wide net that's cast on evaluating candidates, which becomes possible when you're able to lower the cost of assessing each new candidate to the cost of software. Yeah. So I completely see it from the perspective of if you're doing this at a human level, the net definitely is smaller for what you can manage. But what about in terms of other pre-existing talent marketplaces with pre-vetted professionals? I'm thinking like TopTal, Turing, both, at least I've seen to be pretty popular for contracts, software engineers, at least.

5:57How do you guys stand out from, say, some of those players? Yeah, so I'll start with Topdell and then we can get into Turing. Topdell started really as an arbitrage on Eastern European developers where they would hire people there, they would manually interview them, manually match people with jobs. And it was a much better experience than Upwork, which they were competing against at the time because they could assure a certain level of quality. The flip side of that quality is that they would charge an insane fee structure, oftentimes averaging 40 % and not having the same wide net that you can get with automated processes.

6:37And so our thought process was, what if you can take this high service quality of Topdoll and finding really high caliber candidates, but instead of having huge margins and huge fees, because all of those processes are manual, doing that with software, assessing not just like thousands of people, but hundreds of thousands of people finding better candidates at a lower cost structure and delivering them to candidates much faster. So that's the comparison to TopTel and automating those processes. Turing started with the premise of doing a lot of automation in the late 2010s. But the challenge at the time was that LLMs were not capable enough to achieve what they wanted to or models in general.

7:26And it wasn't until, you know, ChatGPT and GPT-4 and all those models being possible to fine tune that it actually became possible to assess talent at the cost of software. And so Turing built up their entire business as a manual business that was manually doing all of these processes with a lot of AI product marketing and really, really doesn't have complete automation. So multiple executives that have left Turing are now advising us, are excited about the approach of completely automating these processes and really building the best models to predict human ability is the most impactful thing on just ensuring that clients are having good experiences.

8:11Yeah. And so I guess touching on the why now a bit, Would you point that to technology, which is sort of the advancements in AI recently allow you to do some of these operations a lot more efficiently and cost effectively? Hugely. Yeah. I mean, language models are getting extremely intelligent, especially when fine tuned on the right data sets. and that was just previously impossible. Like the idea of an agent conducting an interview of a human two years ago was completely impossible, right? And now agents are able to not only conduct human level interviews, but conduct interviews that are way better.

8:51It's as if a human had prepared for 24 hours before their interview, right? Reviewing all of their GitHub repositories, thinking about all of the details of projects that they have on their resume in asking this in-depth, thoughtful interview about someone's background, giving each person the chance to be considered. And that wasn't even close to possible before we had super powerful foundations. Yeah. It sounds like you've sort of offloaded a lot of the vetting process from humans to technology, specifically AI. We have a pretty technical audience. I think it'd be really fun to dive into some of that.

9:29On the side of the candidate, they're often, as you mentioned, talking or always talking to the AI interviewer when they first sign up. There's also a resume analyzer tool I believe you guys use, plus a system that you mentioned earlier that can even predict candidate performance. I'd love to just start diving into all of that. What is the technology behind that? How did you start building this? How do you even tackle this problem of teaching AI how to vet a candidate? Yeah, yeah. I mean, it's a really challenging problem because humans struggle with it significantly, right? So when you're creating evaluations for model performance, the baseline in question that you're trying to optimize for is very challenging.

10:13I would say, like, the first thing that's really interesting about the product is the AI interviewer, where being able to, you know, scalably do this volume of traffic requires a lot of pre-processing to, you know, achieve the results that we want to, as well as agentic behavior during the interview to be thoughtful about what the next questions are going to be. And then the other thing that's really fascinating are all of the models around predicting which candidates will do well when we're surfacing candidates. The rough framework that we use for this is that there's two dimensions to how good a fit a candidate is for a role.

10:52There's the relevancy of that individual, and then there's the excellence of the individual, how competent they are. But these dimensions are deeply intertwined. And so we do a lot of work with not only leveraging language models, fine-tuning language models on huge amounts of proprietary data, but also on training, like learn-to-rank models from the ground up and doing all sorts of fine-tuned embeddings to surface the best candidates for customers. Yeah, in terms of like quantifying or even trying to qualify excellence, how do you sort of put that? I mean, you're mentioning how you're fine tuning models.

11:28What data are you looking at? How are you training these models? Anything that you can speak on there would be great. Yeah, there's two kinds of evaluations or KPIs that we go on. The first are KPIs around human preference, where we have internal members of the team that will create evaluations for which candidates we think are best for roles, for what reasons, and then we'll map to allowing models to predict that in the same way that humans can. considering all sorts, leveraging language models and various other models to predict based on all of the attributes of an individual ranging from features of their GitHub contributions to their facial expression during an interview, who is likely passionate and likely perform well in that given role.

12:16The other kind of evaluation KPI that we go on, and the more important one in particular as we scale, is around customer satisfaction, where we collect the reviews of all of the candidates and the experiences that customers are having on them, and then train models that can predict for a given candidate, what is the likely experience that they're going to have. So imagine like a transformer model that can intake all of the performance reviews of candidates for given roles and then can predict what is the most likely next performance review that this candidate is going to have based on all of the data about their background, the role that they're being considered for, and all of the data that we've fine-tuned on.

12:59I see. Wow. I mean, I think it's really cool. You mentioned you guys are even taking into account facial expression during these AI interviews. I suppose a human does the same thing, right? And so, you know, touching on that a little bit more, since it sounds pretty interesting, is this a matter of sort of collecting all this data, whether it be facial expression, plus what they said, and then feeding it to an LLM and letting it reason? Or how are you guys approaching this? Yeah, so we started out with just textual data, like just the transcript of the interview. And then we expanded to also do visual stuff.

13:31And I will say that we're just starting to dip our toes into a lot of the more complicated visual stuff, just because models are only now beginning to have really precise, you know, ability to extract features from these interviews. But I find it to be such like an interesting technical problem, because it's something that humans struggle immensely with and models are actually super human. Like MIT released a paper that they were able to predict the heart rate of an individual in an interview with a reasonably high degree of confidence based on the recording of that interview. And I believe that there's all of these subtle nuances and features of the heart rate of the individual, the facial expressions they have, the tone, all of these little nuances that actually go a long way in predicting that person's attitude, their passion, their competence, and all of these different characteristics and how they'll perform for a role.

14:29And so our focus now is in digging in beyond just what they're saying to all the other characteristics of that individual and how all of these subtle things that are often overlooked can be priced in reasonably to predict their performance in a given task or job. And have you seen any promise with that, like being able to take their facial expression or their voice inflection, whatever it may be, is that actually a pretty reasonable predictor? So we found that audio is certainly something that's really, really important and has way more signal than people's transcript. However, However, the visual evaluation is just starting to, we found some signals in visual evaluation.

15:15There's interesting things like the blurriness of someone's camera actually contains some signal. Wow. Yeah, yeah, yeah. And there's like little features like that. We're really just dipping our toes into the facial expression stuff. But I believe that as we get more data about this, that there's going to be a lot of signals that emerge. One thing we talk about a lot internally is that like this fascinating problem of predicting human ability is extremely high dimensional, right? There are so many things that go into what make you good at a given task or job, everything from, you know, all of the different tangentially relevant experiences to your motivation to the manager in that job, whether they're like incentivizing you and you want to do a good job.

16:02And because the problem is so high dimensional, similar to predicting the next token and a transformer model, it requires an immense amount of data to actually help you meaningfully extract the right features. And so one of the reasons that we're so excited about the business and all of the progress that we're making is that it allows us to operate at a much larger scale with a lot more data where we're actually able to make sense of these small features that we previously weren't able to. And so as the performance data expands, it makes all of that possible. Yeah, absolutely. And just to probe a bit more here, just because I find this technology to be so exciting, when you're collecting this information, now are you taking into account more, say, multimodal models, vision language models?

16:51Is this where you guys are headed now, if you're expanding beyond just the textual space? Completely, yeah. So that's why the frontier Your models that can do good multimodal evaluation are extremely compelling to us. And we've been using vision models for a while and are starting to, you know, explore a lot of what's possible with audio models, as well as more sophisticated features with vision models. There also are other kinds of custom models we've run experiments with, like the facial expression stuff that we were previously discussing as an example of that beyond just language models. But there's all sorts of models that go into extracting good features.

17:32And then it's about mapping those features to the right performance metrics that we want to optimize for. Yeah. And I'm curious in the early days, it makes complete sense that once you get a bit more data on one of the candidates, say they've done a couple of jobs through the platform, the reviews are pretty valuable. But when it comes to that first almost, you know, zero shot, one shot approach of sort of classifying this candidate, how did you guys sort of test this? Was it just like pray that this works and hopefully the companies like it? Or how did you go about that? It started with a lot of doing things that don't scale and was sort of, you know, of doing human processes and think how we can replace them.

18:19where my co-founders and I have done thousands, not exaggerating ourselves. And we obviously would be able to reach the scale that the models are at now for us, hundreds of thousands of interviews. But you gain an immense amount of appreciation for all the different ways to optimize an interview experience and the importance of automating all of those processes. And so it started with manually reviewing resumes and then figuring out how we can automate at least the short listing, even if not the final review. And then I remember for a time, we automated conducting the interview, but not yet evaluating the interview, because it would actually be way faster.

18:59We could just like watch interviews on 2x speed and then, you know, and stop an interview if someone wasn't good and then pushing towards complete automation, which is where we're at. Hey, we'll continue our interview in a moment after a word from our sponsors. Hey everyone, Eric here. In this environment, founders need to become profitable faster and do more with smaller teams, especially when it comes to engineering. That's why Sean Lanahan started Squad, a specialized global talent firm for top engineers that will seamlessly integrate with your org. Squad offers rigorously vetted top 1 % talent that will actually work hard for you every day.

19:37Their engineers work in your time zone, follow your processes and use your tools. Squad has front-end engineers excelling in TypeScript, React, and Next.js ready to onboard to your team today. For back-end, Squad engineers are experts at Node.js, Python, Java, and a range of other languages and frameworks. While it may cost more than the freelancer on Upwork billing you for 40 hours but working only two, Squad offers premium quality at a fraction of the typical cost, without the headache of assessing for skills and culture fit. Squad takes care of sourcing, legal compliance, and local HR for global talent.

20:10Increase your velocity without amping up burn. Visit choosequad.com and mention Turpentine to skip the waitlist. I see. What would you say was the most challenging part when it came to building this technology? I mean, it could either be one of these specific features, whether it be the AI interviewer, the resume analyzer, or maybe it was just at a broader level, maybe data collection. Like what was really hard to get right for this to work? Yeah, I think the largest challenge, which will continue to be the largest challenge for probably a long, long time in the business is in predicting who's going to perform well for what reasons.

20:50It's just like such a challenging problem because humans aren't very good at it. And there's so many problems within that of collecting the right data. Data is like super subjective and you want to be very thoughtful about, you know, minimizing bias and having peer review on all of that data and having experts that have deep expertise in the domains that the person is discussing. all the way to model architectures, to different approaches and training them. And just optimizing that model to be as predictive as possible is, I think, the most interesting as well as the most challenging problem in the business.

21:31And what would you even start to lay out as potentially a moat? I assume that maybe some of these other players that we previously mentioned might be thinking of, hey, we should maybe use some AI here as well. I believe I've heard maybe a couple other startups have at least looked at the space, maybe none with as much traction as you guys. But how do you think about that, like defensibility? Yeah. So there's two modes that come with a business. And then I'll give a spiel about like my broader take on why labor aggregation is such a large opportunity. The first one is the marketplace, which is that when you're facilitating transactions and you build up a large amount of supply side and a large amount of demand side, there's network effects in strengthening each other.

22:19And so as we get larger and larger, that makes facilitating transactions even easier and it makes it even more challenging for our competitors. The second mode is around a usage data flywheel, where we're able to learn from not all, but most of our customer interactions to predict based on the previous thousand and roles that we placed, who is likely to perform well at the next role. And that can be really, really powerful, especially as we start reaching for placing millions of candidates. Because if others don't have that performance data that's structured in the right way to train an LLM, then they're not going to be even close to competing on the same dimension.

23:01I think at a high level, the reason that I'm so interested in the labor market is that it's the largest market in the world and also the most inefficient. And I really strongly believe that building the infrastructure to collect proprietary data on everyone on earth and why they're a good fit for any role and matching people with those opportunities is not only the largest opportunity of the economy, but also the most defensible one because it's predicated around really strong network effects and proprietary data to support them. And so I think the defensibility of the business is actually one of the things that made me so excited about Mercor.

23:45And from the perspective of the employers who are using you guys, beyond just doing the vetting for them, which therein allows them to just, you know, in theory, see a lot more candidates. Is there other things you offer on the platform that's attractive and makes them use you guys rather than other talent marketplaces? Yeah. So we also facilitate all of the payments and international compliance associated with these transactions. The reason that we do that is twofold. The first is that this is a solution customers want. They don't want to have to deal with all of the overhead of using different tools, onboarding people, having their own documents.

24:23they just want to press a button and have that person ready to go on their team, which is the experience that we deliver for them. And the second thing is that through seeing the entire life of the transaction of how long a customer is working with a candidate, how that experience is going, it allows us to collect really valuable data. And that speaks to the performance data flywheel that we were talking about earlier. If we were to say, just be like LinkedIn and, you know, have a candidate that a recruiter can search for and not get any data beyond that, we wouldn't get the most important part of that transaction, which is the ability to predict how well a candidate is going to do.

25:04And so really being with a customer through the complete life cycle of seeing their problem in the initial steps and how the solution ends up playing out, obviously in this case, you know, talent is immensely important for the data that informs all elements of it. And for these companies, where does a lot of the cost effectiveness for them come from? Yeah, so reducing the reliance on recruiters is huge. Another big thing is allowing them to unlock a lot of undiscovered talent internationally. One thing we've found is that many people like the idea of hiring someone in India or South America or Europe or wherever because of better price performance in terms of hiring a better engineer or technical project manager that can do a good job or work super hard and not cost as much money.

25:59But then they'll have experiences that are more mixed than people they might find in Silicon Valley. And our core belief is that there is incredible talent internationally, but it's not as dense as the talent you find in Silicon Valley. And when you're able to automate the process of assessing that talent, solving this talent density problem, then you're able to find the diamonds in the rough. You're able to find the people no one else is considering and just create incredible experiences for both sides. Like I can't tell you how many stories we've had of people we've hired from, you know, parts of India in particular that are phenomenal engineers, honestly, like much better than many engineers we work with in Silicon Valley, aren't given opportunities, even remotely in the same ballpark.

26:48And through allowing them to discover these opportunities, it changes their lives. And the idea of doing that for a given individual is really fulfilling. But the idea of scaling that for thousands or millions of people is even more meaningful. How do you kind of show to these companies that the technology that you have built for vetting is up to parts, up to standard? I'm curious what that sales cycle almost looks like. You know, it's funny because there's so many companies out there now, especially that say that they're AI companies, right? And how they're using AI to help solve a customer problem.

27:25And we took this really unique approach, which is instead of just saying we're an AI company and then sending them a resume, we showed them what we were doing. We showed them like why we're selecting candidates. We showed recording of the AI interview that's being conducted of the candidate. So clients can watch that interview, see that it's a thorough interview, and decide whether or not our vetting was thorough enough based on that. We show them the analysis of the work experience of that individual that has the GitHub contributions that we're assessing, as well as a summary of why we selected that individual for their query.