A friend applied to 47 jobs last quarter. She heard back from four. Two were automated rejections. One asked her to take a 90-minute coding test, then went silent. The fourth was an interview that vanished after the second round. None of these companies were bad employers. They were just operating a hiring funnel that is, structurally, broken.
The inbound funnel breaks at the volume step. A role at a 200-person company gets 250 to 400 applicants. The recruiter has, on a generous estimate, 30 seconds per CV. That is two and a half hours of CV-reading per role. Multiply that by 8 open roles and you have a full week of someone reading CVs before anyone speaks to a candidate. So the volume gets squashed: ATS keyword filters, top-of-pile bias, and a default of silence. About 90% of the funnel never gets a real signal back.
Three things broke. Candidates stopped trusting the application form. Engineers in particular started routing job hunts through their network and through Discord servers, because the public funnel felt like throwing a CV into a void. Employers responded by paying for outbound sourcing, which is what made LinkedIn Recruiter an 8k-per-seat product. And the candidate, the actual person trying to get a job, ended up doing the same 47-applications-to-4-replies maths my friend did, only worse, because the rejections stop arriving at all.
You can run a hiring funnel that respects the candidate. You cannot run it manually at 250 applicants. The volume math does not let you.
There are several public estimates of candidate response rates. The one we trust most is the SHRM 2024 talent acquisition survey, which puts the share of applicants who receive any individual response at 8% to 12%. A 2023 Indeed survey came in at 7%. Internally on the Neuroworx side, we have looked at applicant logs across 30-plus mid-market customers and seen response rates between 6% and 14%, with the median sitting at 9%. The headline rounds to 90% no-response.
A no-response is not always a rejection. Sometimes the company hires from another channel and the inbound queue is never closed out. Sometimes the applicant slipped past a keyword filter. Either way, from the candidate side, the experience is the same: silence.
When we built Picked, we picked the candidate experience as the load-bearing constraint, not the recruiter dashboard. Three product decisions follow from that:
The trade-off is real. Doing this manually would be impossible at 99 cents per candidate. Doing it with Claude doing the reasoning is, just about, the right unit cost.
The manager gets three vetted finalists with full transcripts on Friday morning. That is the deliverable. The 200-CV stage disappears, because the system has already done the vetting in a way the manager would have had to do anyway, except properly, scored against a calibrated rubric, with the audit trail.
The 90% number is not a marketing line. It is the size of the candidate trust hole. The whole product is an answer to that one number.
Founder of Neuroworx. Eight years building psychometrics into hiring. Writes about the unit economics, the candidate side of the funnel, and what shipping with Claude looks like in practice.