Classet came out of Bluecrew. Our three founders, Gino, Nick, and Coop, spent about a decade there building a digital staffing platform that got acquired by EmployBridge, the largest light industrial staffing firm in the country. They helped over a million hourly W2 workers find jobs. That gave them a very specific view of where hiring breaks. (I should also add that I was one of the OG interns back at BlueCrew in its early days!

One pattern kept repeating: sourcing was never their bottleneck. There were always enough candidates. Screening was. A distribution center posts four openings, gets 800 applicants, and two recruiters are supposed to call all of them. Most of those applicants never hear from anyone. Roughly 40% of candidates abandon an application because nobody got back to them.

So we built Classet around AI phone screening for high volume hiring. Our AI recruiter, Joy, calls an applicant within seconds of them hitting submit, conducts a structured interview, answers their questions about pay, shift, and location, then writes the transcript and summary back into whatever ATS the team already uses. She’ll screen 150 applicants in about ten minutes, at 2am, on a Sunday. That matters when more than half of hourly candidates apply outside business hours.

What makes us different is mostly what we left out. No video. No ranking candidates against each other. No scoring anyone on tone, personality, or how enthusiastic they sounded. No vocal or facial analysis of any kind. Those are the features where bias gets into hiring tech, and we made a deliberate decision not to build them. Joy gathers objective, job-related information and hands a structured summary to a person, who makes the call. AI does the volume. Humans do the judgment.

Four channels do most of the work: customer referrals, in-person events, paid search, and a growing number of people who found us by asking an AI assistant.

Referrals are the biggest and the best. Talent leaders in high-volume hiring all know each other. Someone running 3,000 hires a year at a distribution company knows someone with the same problem at a hospitality group, and one conversation between them does more than any campaign we could run.

Where I wish it came from is more of that. Referrals close faster, churn less, and cost us nothing. Paid search works, and we’ll keep spending there, but you’re renting attention from someone comparing five vendors at once. A referral shows up already believing you.

Handed over: account research. Before we talk to a company, an agent pulls their open roles, hiring volume, which ATS they run, and which jobs have sat unfilled for six months. That used to eat up a rep’s entire morning. Now it takes a few minutes, and it’s better than what we did by hand.

Went back to human: cold outbound. We tried to automate the whole thing end-to-end. Every other post online says SDRs are cooked. We ran that experiment and pulled it back hard.

Everyone is exhausted by AI slop. The second your email reads as if a machine wrote it, you’re done. What works is a person who did real research, used AI for the grunt work, and wrote the message themselves. The reply rates aren’t close. Same story with our content. We spent months working out what people actually find useful instead of handing the calendar to a model, and the pieces that we spend time researching, writing, and publishing are the only ones that perform.

It’s the same principle the product is built on. AI is excellent at volume and bad at judgment. Give it the repetitive work and keep a person on the part where taste matters.

Yes, and they turn into our best deals. Someone who books a demo after asking ChatGPT or Claude about AI phone screening closes 50-60% faster than our other inbound leads, and that holds even for enterprise contracts.

My read on why: an AI assistant doesn’t hand you ten blue links. It gives you a short list and a reason. By the time that person reaches us, they’ve already done the comparison, so the first call is about specifics, implementation, etc.

Most people show up well educated now, whether an LLM sent them or not. We’re deliberate about that. Our site spells out what we do, who it’s for, what we don’t do, and what the results were, because that’s what an AI assistant can read and repeat without garbling it. Vague positioning used to cost you a conversion. Now it costs you the recommendation.

Do: treat the candidate as an end user (not just the recruiters). Most people building for high-volume hiring optimize for the recruiter because the TA team signs the contract. You should do that AND prioritize the candidate too. The candidate is the one who decides whether your funnel works at all. 88% of candidates who complete an interview with us rate the experience positively, and that number is the reason our customers’ pipelines fill and our customers stick with us.

Spend ten minutes on Reddit reading what people say about one-way video interviews or video interviews in general. It’s brutal, and it’s earned. Candidates describe recording themselves for a camera and never hearing another word. If your process makes someone feel like they’re auditioning for a machine, they’ll drop out, and you’ll blame the labor market. Or if you’re making candidates log in to a computer or download software, you’re already discriminating against folks who may not have access to that technology.

A few specifics that matter more than they sound:

Tell candidates it’s AI, up front, every time. We find that less than 1% of candidates opt out of calls when we do that.

Don’t make candidates download an app or create an account. A phone call works on every phone that already exists.

Let the AI be patient and speak the language the candidate is most comfortable with. That one change does more for equity than most of what gets sold as bias mitigation.

Don’t score or rank people using AI. Judgment on hireablility should be left to a human.

Ignore anything sold on the promise of reading a person. Tone analysis, sentiment scoring, facial or vocal "signals," personality inferred from a 90-second answer. That’s the part of this category that regulators are already circling; it doesn’t predict job performance, and candidates hate it.