When a company is looking for a candidate for an open role, the hiring manager is probably going to rattle off a bunch of qualifications that they’re looking for to a recruiter — and Kanjun Qiu says recruiters will probably just run with that when the manager’s requirements might not actually be so rigid.
It’s that intent from the manager — the idea that the actual boundaries for a qualified candidate are more opaque — that sparked the idea for Sourceress. Instead of just hunting down candidates based on a bunch of keywords, Sourceress works with hiring managers to understand the kinds of attributes they need in a potential hire and builds a model to find someone who would fit what a hiring manager is looking for, even if they don’t fit the bill explicitly. To do this, Sourceress has raised $3.5 million in new financing from Lightspeed Venture Partners, OpenAI researchers, Y Combinator, Dropbox founders Drew Houston and Arash Ferdowsi, as well as other smaller investors.
“The advantage is that when you source and you go outbound as a company, people feel like, oh, you want them,” Qiu said. “You’re extending a hand out to them, and then they can choose to take your hand or not take your hand. It makes you feel like you’re wanted, that you have these options, that you could go somewhere. The problem today is sourcing is so transactional, you hire sourcers who are on contract or not on contract. It’s hard for you as a sourcer to spend time personalizing and customizing an approach, and the tools aren’t really there.”
For example, just because someone doesn’t have experience with a specific programming language doesn’t mean they can’t be trained in that language. So, rather than completely ignore a candidate because they don’t have experience working in JavaScript, Sourceress should pick up on a candidate with years of experience using Python and flag them as someone worth flagging as a potential hire. The same might be true of a qualified candidate with experience using that language, but fewer years than what a company sets for its initial standards.
The problem starts with a phone call with a hiring manager, where that person will detail to Sourceress what they want in a candidate. Sourceress then builds a model based on that information and starts scouring for candidates on the avenues that you might expect, trying to bend the boundaries so they aren’t so rigid in their search for candidates. Each additional hire tunes those algorithms over time to better look for candidates. Right now, Sourceress focuses on engineering and product — because, for now, it makes sense to be working in an area where the team has experience.
It’s that tuning part which is probably the most critical aspect of Sourceress’ future. Having to take a call with a hiring manager every time can be a pain, especially as more and more hiring managers call in and are really looking for candidates with very similar profiles. As Sourceress matches the right candidates, its idea of what a manager that wants when they ask for “a Python expert” will start to better understand the intent behind their search for a candidate, rather than just taking the qualifications at face value. The models become more abstract, and eventually, once Sourceress has enough data, it can automatically divine the right candidate profile.
Right now, there’s no candidate-side part of the service, as the low-hanging fruit is more on the recruiting side. But it would make sense to use such a model to slot into the spots that Indeed, Hired, or even LinkedIn, have tried by giving candidates a hub to go and find potential job matches. Most potential hires are passive candidates that aren’t looking, and it’s hard to determine who to reach out if they aren’t raising their hand, Qiu said.
Taking this kind of approach by looking for potential attributes — and not just qualifications — is something Qiu said would help surface up more diverse candidates, which she said tend to have a higher response rate. Qiu also said the percentage of our hires for women and minorities on Sourceress is between 30% and 40%.
“Women, when they look at a job description, they tend to disqualify [themselves],” Qiu said. “So if you’re reaching out they’re more comfortable talking to you. If we’re able to actually assess for merit, and we’re able to fill the top of the funnel with more women or minority candidates, your likelihood of hiring someone goes up. If you’re not getting diverse candidates into the pipeline, it’s hard to make diversity hires. The problem is most pipelines, they’re referral based. Coming into this, we thought, if we can make finding candidates getting in touch with them much easier, we should be able to change.”
Since it’s a language problem as much as it is an unstructured public-facing data problem, it’s going to be an area with intense competition. There are startups like Headstart looking to help analyze candidates, though that process more deeply involves the candidate side in order to determine the right fit. There are, indeed, a lot of startups getting funding in this space — and it’s likely that plenty of the bigger companies are working on such tools.
The end goal would be, for example, for Sourceress to be able to find a student at a college in the midwest that will either immediately or one day fit the needs of a hiring manager. That might require scouring a Github account, or published papers, or what kinds of posts they put up on Stack Overflow. But the point is to come up with a diverse set of information sources that can help the company identify candidates that a recruiter might not find if they were just digging through LinkedIn for potential leads. All this data would naturally be public-facing, which means it could be up for grabs for anyone, but in the end, it’s the approach that matters more, Qiu said.
“The actual data itself doesn’t matter, it’s how you post-process it and the features you extract,” She said. “That’s our meta processing layer, that’s the difference.”