Customer Story | OneSignal

Manual fraud detection with no reliable system: Julia was spending one to two hours per week manually checking LinkedIn profiles, connection counts, and account creation dates for every suspicious application — with no consistent way to confirm or rule out fraud.
Fraudulent candidates slipping deep into the process: Without an automated check, fake applicants were making it through multiple interview rounds. One candidate reached seven interviews and nearly received an offer before Julia's instincts flagged something was wrong.
Recruiter time and morale eroding fast: Back-to-back screens with fraudulent candidates were draining Julia's time and energy — time that could have been spent building relationships with real candidates.
Inbound pipeline losing credibility as a source: With fraud rates approaching 90% on some roles, the team was losing confidence in inbound entirely and defaulting to sourcing and referrals just to find trustworthy candidates.

Results with Gem

14+ hours reclaimed per role: By automatically flagging fraudulent applications before review, Julia eliminated hours of manual vetting and avoided screening calls with fake candidates entirely.
Fraud caught before it reaches the interview stage: Gem's Fraud Detection Agent flagged 112 out of 127 applicants (88%) as high risk on a single open role — stopping fake candidates before they wasted a single interviewer's time.
Recruiters empowered with evidence, not just instinct: With clear risk levels and signal explanations surfaced automatically in Gem, the team moved from gut-based decisions to confident, data-backed ones — with full control over every final call.
Inbound restored as a viable recruiting channel: With fraud filtered out automatically, OneSignal rebuilt trust in inbound as a real pipeline source — ensuring strong candidates who apply directly aren't lost in the noise.

Applicant fraud is growing, and OneSignal was fighting it manually

At OneSignal, Julia Graham had become something of a human lie detector. As a tech recruiter, she was fielding back-to-back candidate screens from 8 a.m. to 6 p.m. — and increasingly, her gut was telling her something was wrong before the data could confirm it.

"It was exhausting. I was constantly on calls where you could tell someone wasn't real, or if they did slip through, I was having to backtrack — a ton of wasted time for me, for interviewers, for everyone."

Julia Graham

Senior Technical Recruiter

The volume of fraudulent applicants had grown so significant that Sam Perlman, VP of People at OneSignal, had started to worry about more than just wasted hours. "When Julia came out of a recruiter screen saying 'another fake one,' I started thinking: what happens to our company data if someone like that crosses the threshold? The risk is terrifying."

That fear nearly became a reality. A candidate made it through seven interviews, impressed the entire hiring team, and came close to receiving an offer — all while Julia suspected something was off. There was no tool to confirm it. Just instinct.

OneSignal's fraud challenge wasn't unique — it was a sign of what's coming. Gartner predicts that by 2028, one in four candidates will be fake, driven by mass bot-generated applications, fabricated identities, and increasingly, state-sponsored fraud attempts. Julia was seeing it play out in real time: profiles impersonating real engineers at real companies, sophisticated enough to pass a first screen, organized enough to clear multiple interview rounds.

Julia was catching most of it, but the process was manual, inconsistent, and unsustainable. Checking LinkedIn connection counts. Looking up when accounts were created. Cross-referencing email domains. For every suspicious application, that's one to two minutes of investigation — before even picking up the phone.

"It's a scary market. Is it someone who'll just take the money and run? Someone trying to multi-job? Or now — literally — someone with different intentions entirely? That's a risk I can't accept, and Julia should not be our only line of defense. It's just not safe, and it's not scalable."

Samantha Perlman

VP of People

Finding Gem's AI Fraud Detection Agent

OneSignal first heard about Gem's Fraud Detection Agent through the Higher Community, which announced the partnership with Tofu and caught Sam's attention immediately. She followed up directly to get an intro to Jason Zoltak, co-founder of Tofu, the fraud-detection technology that powers Gem’s agent.

He demoed the product live, using OneSignal's near-miss candidate as a real-world test. The results were immediate: the system surfaced exactly what Julia had suspected, backed by evidence. When Gem announced Early Access for Fraud Detection Agent, OneSignal was first in line.

"Gem's AI Fraud Detection Agent didn't even take five minutes to figure out. It was just seamless — intuitive with my existing flow in Gem. When you're learning new tools, it can take time to really get the hang of things. This was immediate."

Samantha Perlman

VP of People

88% of at-risk applications flagged and 14+ hours reclaimed on application review

The impact was visible from day one. On one open tech role, Julia pulled up her applicant queue in Gem: 127 total applicants. 112 flagged high risk.

Her new workflow is more streamlined: Review the high-risk applications with a quick spot-check, bulk reject with confidence, then give meaningful attention to medium-risk candidates — digging into LinkedIn profiles, connection counts, account creation dates — the things the tool flags as worth a closer look. From there, she moves to low-risk applicants and advances the best.

The time savings are significant. For one role, Julia estimates one to two hours saved just on applicant review, plus roughly 12 hours of screening time avoided by not advancing fraudulent candidates to phone screens. That's 14+ hours reclaimed just on one role — not including downstream interviewer time.

One unexpected insight emerged during early access: When Gem's AI application ranking and Fraud Detection Agent run together, candidates matching 80–100% to a job description are almost always fake. Bots auto-apply by tailoring profiles to match job descriptions exactly. Real candidates, who have company-specific experience and their own backgrounds, rarely hit that mark.

"If Julia saved two hours reviewing, then avoided screening 25 fake candidates, you're already at 14 hours. Add scheduling, hiring manager time, follow-ups — in one role alone, it's substantial. And that's just one role."

Samantha Perlman

VP of People

"Now we can trust inbound again"

The downstream effect of Gem's AI Fraud Detection Agent went beyond the application queue. High fraud volume had been quietly eroding OneSignal's confidence in inbound recruiting as a channel at all. Why invest in reviewing applicants when nine out of 10 are fake?

Gem changed that calculus. With fraud filtered out automatically, inbound is a viable pipeline again — and the real candidates who apply there aren't getting lost in the noise.

"Now we can trust inbound again," Sam said. And for Julia, the change was personal. "It was getting pretty dark there for a second," she admitted. "I could not talk to another fake candidate. It's draining. Now I'm actually spending my time on people who are real."

From gut instinct to evidence-backed decisions, powered by Gem’s all-in-one platform

For Sam and Julia, Gem's AI Fraud Detection Agent isn't just a time-saving tool — it's a foundational change to how they think about recruiting security.

"Relying on one person's instincts isn't a system. Gem's AI Fraud Detection Agent gives us the evidence to make confident, consistent decisions — and makes sure we're not putting that burden on one person alone."

Samantha Perlman

VP of People

"Trust it. It will only make your life easier. It's an absolute no-brainer."

Julia Graham

Senior Technical Recruiter

Get started today

See how Gem can help you hire with remarkable speed and efficiency