Marcus runs a family restaurant. Forty-five seats, three full-time staff, a few seasonal hires when summer fills the patio. When his lunch server gave notice in May, he posted the job and thirty-eight applications came back within a week. That was more than he expected, and more than he had time to read carefully between prep and the dinner rush.
So he paid for a month of an AI screening tool. The pitch was simple: feed it the applications, it reads them faster than you can, and it hands you back the strongest candidates. He let it run on the settings it came with, the way most of us install an app and tap "accept" without changing anything. It gave him a ranked list of eight people. He called those eight, interviewed four, and hired one. Start to finish, the whole thing took a fraction of the usual time. He was pleased with it.
A few weeks later he was at a community fundraiser and a woman he knew socially came over. She'd applied for the job. She wanted to know, friendly but direct, why she'd never gotten a call. Marcus told her the truth: the first cut had been automated. He said it the way you'd explain that the bank closed early, like it was just how things worked now.
On the drive home it started to bother him. He pulled over, opened the screened-out pile on his phone, and looked at her application. Then he looked at a few others the tool had ranked low. He couldn't tell why. The tool had given everyone a score, but it hadn't given him a single reason. If someone had sat him down and asked him to explain, in writing, why this woman had been ranked below the people he interviewed, he couldn't have done it. He hadn't made that decision. The tool had, and he'd signed his name to it.
And the woman from the fundraiser wasn't just a missed candidate. She has friends who eat at his place, family who books the back room for birthdays, a place in the same small circle Marcus depends on. A family restaurant doesn't run on foot traffic from strangers — it runs on being the place people in town trust. He hadn't only screened out an applicant he couldn't account for. He'd put a crack in the one thing his business actually stands on, and he'd done it without ever deciding to.
The problem isn't speed. It's the missing reason.
Here's the thing worth understanding. A screening tool that sorts people faster is genuinely useful. The trouble starts when it sorts them for reasons it won't show you. A score is not a reason. "The system ranked her eighth" tells you where she landed, not why, and "why" is the part you're on the hook for.
In hiring, you're expected to be able to account for your decisions. Not in a courtroom-drama sense, just in the ordinary sense that if a candidate, or eventually someone official, asks why one person got a call and another didn't, you should have an answer that holds up. "More relevant experience." "Available for the shifts I needed." "Already done the job somewhere similar." Those are answers. "The algorithm scored them lower" is not an answer. It's the absence of one, dressed up to look like a process.
Marcus didn't do anything reckless. His whole business runs on reading people, and it always has. He knows which regular wants to chat and which one wants to be left alone, which new hire will last and which one won't — all of it by instinct, across the counter, in a few seconds. When the tool offered to read the applications and hand back the strong ones, what he heard was someone taking the reading off his plate, the part he had no time for between prep and the rush. He wasn't choosing to let a machine judge people. He was getting out of a stack of paper he couldn't get through. The "fit" scoring was buried inside that, doing the judging quietly, and it never occurred to him to turn it off because he never understood it as judgment in the first place.
Two rules before you let a tool sort people
The fix isn't to throw the tool out. It's to use it only for the parts of the job you could have done by hand and explained afterward.
First, let it filter only on things you set yourself, in plain terms. Years of relevant experience. Available for the shifts you actually need covered. Has worked a similar role before. These are facts you can point to and defend in one sentence. Many of these tools also offer to rank candidates by "fit" or "potential," scoring people against criteria the tool decides for itself. Turn that off. If you can't say in a sentence what "fit" means, you can't stand behind a decision based on it, and you'd be standing behind it whether you understood it or not.
Second, before you call your shortlist final, spend ten minutes checking the rejections. Pull two or three applications the tool ranked low and read them yourself. For each one, see if you can finish the sentence: "I'm passing on this person because ______." If the reason makes sense to you, fine, that's a decision you made. If you can't fill in the blank, put that person back in the pile and read them properly. The whole spot-check adds maybe thirty minutes to a hire. What those thirty minutes buy you is the difference between a decision you made and a decision that was made for you while your name was on it.
That's the line. The tool can carry the reading. It can't carry the responsibility, and the responsibility was never the part that was slow.
Marcus kept the tool. He turned off the "fit" scoring the same night, and now he reads a handful of the no's before he finalizes anyone. The woman from the fundraiser didn't get the job the second time around either. He told her so himself, on the phone, and he had a reason.