A system built a hundred years ago to sort people fast, not to find the right fit.

Every recruiter has heard both complaints in the same week. A candidate says no one will hire her, not for lack of trying. A hiring manager says he cannot find anyone, not for lack of budget. Neither one is wrong. The reason both are right at the same time is not a coincidence. It is the design.

A hundred years ago, the U.S. Army faced a problem no company faces today: sort roughly a million and a half recruits into the right roles, fast, with a war underway and no time to get to know any of them. A team of psychologists led by Robert Yerkes, working under the Army’s Surgeon General’s Office, built the Army Alpha and Army Beta tests to do exactly that. By the war’s end, more than 1.7 million recruits had taken one or the other. It worked, in the sense that the Army stopped guessing and started sorting at scale. It was never built to find the best match for each role. It was built to move volume under scarcity.

When the war ended, the people who built that system did not set it aside. Walter Dill Scott, who had run the Army’s own personnel classification committee, founded one of the country’s first personnel consulting firms within months of the armistice, and carried the same testing and classification methods straight into corporate hiring. That tradeoff, speed over fit, is the direct ancestor of the hiring process running today. The tools changed. The tradeoff did not.

For most of the century since, a person sat somewhere in that process and could notice when the sorting logic was obviously wrong: a gap on a resume that had a good explanation, a nontraditional path that still added up. Automating the filter took that person out. A 2021 Harvard Business School and Accenture study found that automated screening rejects more than 27 million workers who were otherwise qualified for the jobs they applied to. Nearly half of employers using resume sorting software reject any candidate with a six-month gap, no matter the cause. A meaningful share of rejections trace back to nothing more than a parsing error, the software failing to read a resume correctly and scoring it incomplete. None of this is a bug in an otherwise sound system. It is the original wartime logic, unsupervised.

This is why the two complaints land in the same week. A system optimized for throughput will always produce a flooded inbox on one side and an empty shortlist on the other, because it was never measuring fit in the first place. It is doing exactly what it was built to do, at a scale nobody designed it for. The frustration is not evidence the system is broken. It is evidence the system is working.

Then it got worse

The next move was predictable. If the filter reads for keywords, the fix is to write for keywords, and candidates started doing exactly that, at scale, with the same kind of tools employers use to screen them. Applications are up sharply as a result. Recruiters now spend real time just separating spam from substance before they can evaluate anyone. Neither side is wrong to be doing this. Each is responding rationally to a system built on a proxy that turned out to be gameable. But an arms race between two automated systems does not produce a better match. It produces more noise on both ends, and less value in the middle.

Germany’s answer to this problem is well known: a dual apprenticeship system where, according to OECD figures, roughly half of upper secondary students split time between classroom and paid, on-the-job training, often for years, inside a single company. It works, and it works because it is coordinated at an industry and national level, built over decades, with unions, employer associations, and government all pulling in the same direction. Almost no individual U.S. company can replicate that on its own, and for most, technology is exactly the wrong function to try it with. Technology needs shift every year or two. A training pipeline built around this year’s skill gap is aimed at a target that will have moved by the time it is ready. Building an apprenticeship program for a function that is not the actual business a company runs is a full second job, for a need that will not sit still long enough to justify it.

What still works is the underlying idea apprenticeship was built on: watch someone do the work before you commit to them. Most hiring managers already do a version of this informally, through their own network. It is a good instinct. A trusted referral, a trial project, a day spent working alongside the team, all of it produces a better signal than a resume ever will. But a personal network has a ceiling. It works when hiring is occasional. It stops working the moment hiring becomes constant, because the pool of people already known runs out long before the need does.

That is the gap TM Floyd Talent is built to close. Before a single person is placed, the work is scoped: the skill mix, the seniority, the judgment the role actually requires. The team is built to that scope, not pulled from whoever is next in a queue. When a federal compliance mandate required a 43-person specialized team across multiple locations, it was standing and working within 60 days. The people a client evaluates are the people who do the work. There is no junior team behind a senior face.

The hundred-year-old triage problem was never really about interviews, or tests, or software. It was scale outrunning judgment. The fix is not a smarter filter. It is a team built for the specific work in front of you, evaluated the way you would evaluate anyone about to be trusted with something that mattered.