Fraud has become the problem talent acquisition cannot stop talking about—Stacy Zapar and Magen Gicinto’s keynote this morning at the ERE Recruiting Innovation Summit will dive into the cloak-and-dagger side of it—and it has turned screening, never the most glamorous corner of recruiting technology, into a hot segment of the market.
Greenhouse CEO Daniel Chait told Forbes last month that applications on the platform are up 20 percent year over year, with the average open role now drawing 270 applicants. In 2022, by Greenhouse’s own benchmark data, that figure was only 116. Over the same period the average recruiting team shrank by more than half.
An application, which once at least indicated interest in an open role, now costs next to nothing and often does not even mean that the candidate is aware that the job exists. The top of the funnel has filled with resumes written by AI, tailored by AI, and often submitted by AI on behalf of someone who may or may not exist. Hari Kolam, co-founder of Findem, put it in one sentence: “The volume of applicants and the [surface] quality of applicants essentially increases because generating English has become a commodity.”
Every screening vendor I spoke with at HR Tech last month agreed that the application document recruiters use to screen no longer carries the necessary information. Where they disagree is where trustworthy information comes from instead. One camp believes the answer is more information directly from the candidate. The other believes the answer is more data about the candidate, from external sources.
The evidence camp starts at the bottom of the funnel, with the background check, and its case is that the check was never really checking. As Yardstik CEO Andrew Johnson describes it, a traditional check gathers a candidate’s information at the point of hire, never asks whether the Social Security number belongs to the person typing it, and never looks again once the person starts work. “You’re just assumed good forever.” Yardstik, which closed a $30 million Series B in September, has built its business around those gaps. According to Johnson, it matches a driver’s license to a face and the face to the Social Security number before it looks at anything else, reads device and network signals such as an email address created last week or an application for a job in Minnesota from a phone in the Dominican Republic, and keeps what it verified, so that one in ten applicants it now sees has been checked before. In gig and remote roles the company sees fraud rates of 10 to 15 percent, including one early customer that discovered a ring selling stolen Social Security numbers to people who used the identities to clear screening.
Findem is applying the same logic to the resume itself. On October 20 the company launched Authenticity Suite Pro, built on Glider, the assessment and proctoring company it acquired this spring. It checks an applicant’s claims against Findem’s database of patents, licenses and company histories, then flags what does not line up: conflicting employment dates, a company that does not exist, or repeated applications from one location. Flagged candidates get a consent-based, proctored verification session with a liveness check and a government ID, and the recruiter gets a report with a plain-language explanation and an outcome of successful, unsuccessful or inconclusive.
Fama, which screens candidates’ social media, looks for evidence in a third place: what the candidate has already said in public. Lens, which Fama launched in October, starts from a written policy, the customer’s own or one drafted with Fama’s help. A set of agents screens against it, with every decision traced back to a specific policy element. CEO Ben Mones gave the real-life example of a social media company that screens influencers for under-21 campaigns and does not allow any reference to alcohol. Lens flagged something unlikely: an ice cream cake. The agent had looked up the recipe, which contained rum. As innocuous as that sounds, it was consistent with the policy, and exactly what the customer wanted to flag.
The other camp has stopped looking for evidence about the candidate and started asking the candidate to make some. Greenhouse acquired Ezra, a voice AI interviewer, about six months ago, announcing the deal at the last ERE Recruiting Innovation Summit, and its founder Ophir Samson now runs voice AI for the company. He knows the technology has a reputation problem. “The conversation right now on Reddit and LinkedIn threads is: this is bad,” he said. “Which honestly, if I received the same information that many of these people received, I would feel the same way.” His answer is that the real choice is not an AI interview versus a human one. It is a written application that goes into a black hole against a thousand others, most of them bots, versus a 10- to 15-minute conversation in which Ezra asks the questions the recruiter cares about for that role, the candidate answers aloud and in real time, and the candidate can ask questions back. “You don’t have to guess what they’re interested in. They’re asking you,” Samson said.
XLOT, a two-person startup from Munich, has pushed the effort idea as far as anyone. Instead of a work test, which candidates treat as homework, it drops them into a simulated first week on the job: one task, incomplete information, and a simulated manager and peers to question. The company filed for a patent in October that covers the mechanism: information about the business appears only when the candidate asks for it. The platform scores the solution, the reasoning behind it, and the questions asked. It also scores the questions not asked. A candidate who never asks about customer acquisition cost builds a plan without it, and the system notices. Co-founder Ilya Billig sees the extra effort as the filter. “Instead of the current state where people apply to everything that remotely looks like a job they may take, they will be doing this quest only for the jobs they like,” he said. “Companies will be getting a short list instead of a stack of resumes. They will get a short list of motivated candidates that demonstrated that they have the right skills.”
Whichever camp a company picks, the candidates who get through will arrive at an interview. Katherine Loranger, chief people officer at Safeguard Global, is retraining her managers to run interviews differently for the AI era. “It’s not going to be about learning what they did, but learning why they did it, what they were thinking, how they would do things differently, and what they’ve learned. Things that they can’t get out of that AI chat module in their interview prep.” She acknowledged it makes candidates uncomfortable. “But that’s where we figure out where they are.”
