Your ATS Is Full of Qualified Candidates You Never Called Back
Companies spend heavily to reach strangers while hundreds of people who already applied sit unscreened. The economics of inbound screening have quietly changed.
There's a strange asymmetry in how companies spend recruiting budget.
Outbound sourcing gets tools, headcount and agency fees, all aimed at finding people who've never heard of you and aren't looking. Meanwhile a few hundred people applied to your last three job postings, and most of them were never read properly by a human.
Those applicants raised their hand. They already want to work at your company. And they're sitting in your ATS going stale.
The Arithmetic Nobody Admits To
This isn't laziness. It's maths.
A reasonably visible role attracts several hundred applications. Reviewing one properly — reading the CV, checking the trajectory, forming a view — takes three to five minutes. That's fifteen to twenty hours per role, for a pool where most applicants genuinely aren't a fit.
So teams do what anyone under time pressure does: skim the first fifty, hire from outbound instead, and let the rest expire. The applications keep arriving. The backlog grows. Eventually someone suggests turning the posting off.
The hidden cost is worse than the wasted effort. Every unreviewed application is a candidate who applied, heard nothing, and formed a view about your company. Multiply that by every role, every year.
Keyword Filters Made It Worse
The first generation of ATS screening tried to fix this with keyword matching and knockout questions. It failed twice over.
It rejected good people for cosmetic reasons — the candidate who wrote "React" where the filter wanted "React.js," the career changer whose relevant experience sat in the third paragraph rather than the job title. And it let through everyone who'd learned to game it, which by now is most professional applicants.
The result was a screen that was simultaneously too aggressive and too permissive, in ways nobody could audit. The appearance of a filter without the substance of one — and a significant contributor to the doom loop both sides now complain about.
What Changes When a Model Reads the CV
A language model reading a full resume against written criteria does something categorically different from a keyword match. It can follow a career across titles that don't line up. It can notice that a stated role understates actual scope. It can weigh a non-obvious background against the specific problem the role exists to solve.
It isn't better than a good recruiter. It's roughly as good as a rushed one, and it doesn't get tired at applicant number 180 — which is the comparison that actually matters at this volume.
Two design decisions separate useful from black box. Score against the same criteria as outbound: if inbound applicants are graded on a different bar, you can't compare them, and you'll keep defaulting to outbound. One scorecard per role, applied to every candidate regardless of how they arrived. And push results back where the work happens: recruiters live in the ATS, so grades and reasoning need to land there as notes and tags, not in a separate tool nobody opens.
The Archive Is Bigger Than the Inbox
Screening today's applicants is the obvious use. The larger opportunity is the back catalogue.
A company that's been hiring for a few years has thousands of past applicants sitting in its ATS. People rejected for a role that no longer exists, or who were strong but arrived two weeks after an offer was signed, or who applied to the wrong team. They're cheaper to reach than anyone in a cold pool. They know the company. Many have since gained exactly the experience they were missing the first time.
Re-scoring a historical talent pool against a role you opened this morning is a genuinely different capability from screening a live pipeline, and it's where the return is highest.
One practical note once scores exist at volume: the average is close to meaningless. The shape of the distribution is what tells you something. Bunched in the middle means your criteria are too generic to discriminate. Almost everything low means the posting is attracting the wrong audience, or the scorecard encodes an expectation the market can't supply. A clear tail at the top means it's working — review the tail.
You already paid to acquire these candidates. The only question is whether you ever look at them.
Umamy connects to your ATS, grades every applicant against the same scorecard you use for sourcing, and writes the results back as notes and tags where your team already works.