We used Round1 to screen our
own hires. Here's what happened.
When you build a hiring product, you either believe in it enough to use it yourself — or you don't. We used Round1 to hire for a Customer Success role. This is the unfiltered version of what happened.
Why we did this
Building a product you don't use yourself is a bad habit. Every time we told a customer "Round1 saves time and removes bias from first-round screening," we were making a claim we hadn't personally tested under pressure.
When we opened a Customer Success role in early 2026, we made a decision: no phone screens until Round1 had run the first filter. Forty-three applications came in over two weeks. We sent them all a Quick Apply link.
What the numbers looked like
31 candidates completed their interview. 12 either didn't open the link or abandoned it before finishing — a 72% completion rate, which we later learned is higher than our average across all users (typically 65–68% for self-scheduled links).
Score distribution: 4 candidates scored above 75 (strong yes territory), 9 between 55–74, 12 between 35–54, and 6 below 35. The AI flagged two candidates with suspicion scores above 65 — structured, comprehensive answers delivered too fast for the question complexity.
"There was one candidate we would have skipped entirely based on her CV — three years of experience, no brand-name companies, a non-linear path. She scored 82. Her answers were specific, structured, and showed exactly the kind of customer empathy the role needed. She's now on the team."
— Round1 founding team
What surprised us
Several candidates who looked impressive on paper (big company names, relevant titles) scored in the 40s. The AI doesn't care about brand names — it cares about what you can actually explain. We would have wasted significant time on at least 3 people we ended up deprioritising after seeing their scores.
We ran our raw job description through Round1's AI Optimize feature before going live. The rewritten JD was significantly more specific — it called out "de-escalation under pressure" and "proactive renewal conversation" as explicit skills. The AI used this to ask sharper questions than we would have thought to include manually.
One candidate (suspicion score: 71) had answers that were structurally perfect but rang hollow. Every response was a textbook STAR format with no personal texture. On a phone call, this might have come across as well-prepared. The transcript made it obvious — every answer was exactly 3 sentences: situation, action, result. No variation, no natural thinking.
Candidates interviewed on different days, at different times. Every single one got the same questions, the same depth of follow-up, the same standard. We've historically struggled with interviewer fatigue affecting our late-in-the-process assessments. That didn't exist here.
What we'd do differently
We set the minimum lead time to 30 minutes, meaning candidates could book a slot as soon as 30 minutes after applying. A few candidates rushed in underp repared — we'd set this to 2 hours minimum next time to give people time to actually read the JD.
We also didn't include the candidate's resume in the first pass — we were testing the AI's JD-only mode. In hindsight, for a CS role where specific industry experience matters, we'd attach the resume from the start to let the AI personalise questions to their background.
The honest verdict
The hire we made — the one who scored 82 and would have been overlooked — justified the entire process on her own. The time savings (roughly 15 hours of phone screens across two weeks) were significant but expected. What we didn't expect was how much we learned about our own hiring instincts by comparing them to the AI's assessments.
We've used Round1 for every role since. Not because we built it — because it's genuinely better than how we hired before.