On-Profile, Not High-Volume: What Candidate Quality Actually Means
Every sourcing tool promises quality candidates and almost none define it. Here's a definition you can measure, argue with and miss — and why more volume makes it worse.
Every sourcing tool, agency and platform in this category promises quality candidates. The word appears in roughly every pitch deck in hiring, including, at various points, ours.
It's almost never defined, which makes it unfalsifiable and therefore useless as a claim. So here's an attempt at a definition you can measure, argue with, and miss.
The Only Test That Matters
A candidate is on-profile if the hiring manager would take the call.
That's the whole test. Not "is this a good engineer." Not "does the profile match the keywords." Would the person doing the hiring, looking at this profile cold, agree it's worth thirty minutes.
It's binary, it's checkable after the fact, and it maps directly onto the thing that actually costs money in hiring: the hiring manager's attention. Everything else is a proxy. Match scores are a proxy. Keyword overlap is a proxy. Years of experience is a bad proxy.
Once you accept that, the metric becomes obvious. What percentage of the profiles you put in front of a hiring manager do they accept? At 20%, you're outsourcing your screening to them. At 90%, their problem changes from filtering to choosing — which is a much better problem to have.
Why Volume Works Against You
The instinct when a search is going slowly is to widen it. More profiles, more messages, more top of funnel.
This makes things worse in a predictable way. Review capacity is fixed — a hiring manager has maybe an hour a week. Doubling the pipeline doesn't double the review; it halves the attention per profile. Reviews get shallower, and the strongest people in the pool are the ones most likely to be lost, because they're also the ones being contacted by four other companies that week.
High volume degrades outreach too. A message that could have referenced something specific becomes a template, because there's no time. Reply rates drop, which gets read as a top-of-funnel problem, which triggers more volume. That's the doom loop running on the employer side.
What Separates On-Profile From Adjacent
After reviewing a lot of rejected shortlists, the misses cluster into a handful of patterns — and almost none of them are about skills.
- Environment mismatch. Strong profile, wrong scale. Someone who's run a function inside a large structured org, applied to a twelve-person company, or the reverse.
- Trajectory, not tenure. Two people with the same seven years look nothing alike if one has been compounding responsibility and the other repeating year one.
- Adjacent domain. B2B, but the wrong motion. Sales, but transactional rather than complex. The keywords match and the work doesn't.
- Unstated exclusions. The competitor you won't poach from. The background that hasn't worked out three times. Real constraints nobody wrote down.
- Hard blockers. Location, work authorisation, language, compensation.
Most of these are exclusions rather than requirements. That's the practical insight: in mature searches, defining what disqualifies a profile discriminates far better than defining what qualifies one. Requirements are broad and everyone passes them. Exclusions are specific and cut hard.
Why Boolean Can't Express This
Boolean strings match strings. They'll find every profile containing "TypeScript" and not "Java." They can't evaluate whether someone actually owned a system end to end, or whether a title reflects real scope.
They also fail in both directions at once — too narrow to catch the person who describes the same work in different words, too broad to distinguish a skill listed from a skill exercised.
The useful model is layered. Hard filters for what's genuinely binary: location, language, graduation timing, employment type, excluded companies. Cheap, deterministic, no judgement required. Semantic assessment for what requires reading: scope, trajectory, environment fit. This is where a model earns its place, because it reads a career the way a recruiter does rather than pattern-matching on tokens. Then human review on whatever survives, at a volume small enough to review properly.
Each layer should do work the layer below can't. Most tools collapse all three, which is why their output feels either mechanical or arbitrary.
Measure It Or Don't Claim It
Three numbers, per search. Acceptance rate on shortlist — of profiles presented, how many the manager agreed to speak with. Time to first qualified candidate, not first message. And rejection reasons, logged, every time.
The third is the one teams skip and the one that compounds. A rejection without a reason is a wasted signal. A rejection tagged "wrong company stage" is a criterion you can add. Get the brief right and all three move together.
Quality isn't a volume problem solved harder. It's a definition problem, and you either wrote the definition down or you didn't.
Umamy is built to a single standard: around 90% of the candidates we surface should be ones the hiring manager wants to meet, with the first qualified profiles in roughly 48 hours.