The Red Flags That Are Actually Green Flags

Published 2026-06-26 · Updated 2026-06-24 · By Alan Wizemann

Topics: AI & Machine Learning, Growth Leadership, Team Building & Culture

A couple of years ago I was on the buying side of this exact problem, and I want to start there, because it is the part I think about most now that the situation has reversed itself on me. A large distributor I was working with wanted to put an AI system in front of its warehouse and supply chain hiring, and the logic was genuinely hard to argue with on its face. The volume of applications was enormous, the recruiting team was underwater, and if a machine could read the first pass faster than a room full of people ever could, then why on earth wouldn't you let it. I understood the appeal completely (I still do), and I walked it back anyway, because the longer I looked at what we would actually be building, the clearer it became that we were about to construct a very efficient machine for rejecting the exact people we most needed to hire.

I think about that decision constantly now, for a reason that is not at all subtle, which is that I am on the other side of it. I am the candidate. And I am fairly sure that some model, sitting in front of a recruiting pipeline I will never see the inside of, is quietly filtering me out before a single human being ever reads my name.

Here is the thing that the people running these systems do not like to sit with, because it complicates the story they have told themselves about efficiency. The filters do not reject bad candidates, they reject candidates who do not match a checkbox, and those are two completely different groups of people. I do not have a degree, I do not have an MBA, and I have had a couple of stints that, measured purely on tenure length, look short. A model out front sees those three facts, flags them, and moves on, and it never gets anywhere near the part of the story where those same three facts are most of the reason I would actually be good at the job.

There is a story I heard that has stuck with me, because it is the whole problem in miniature. A company that takes in something like a million applications a month had a screening problem they could not crack, where people with twenty-plus years of relevant experience were getting bounced out of the process because somewhere in their history sat a three-month gap (not three years, three months) that the system could not categorize. A gap that, if you bothered to ask about it for the thirty seconds it would take, would turn out to be a sabbatical, a sick parent, a startup that did not survive, a move across the country – every one of them completely unrelated to whether the person could do the work in front of them. The machine does not ask. It simply does not advance you, and it does so without ever once raising its hand.

I tested this directly, because I wanted to know rather than assume. I was talking with a recruiter whose firm is growing quickly precisely because they have leaned all the way into AI screening, and I asked her the blunt version of the question, which was whether, with the red flags sitting on my profile (the short stints, the missing degree, the things that look wrong from a distance), her process would ever have surfaced me at all. She thought about it for a moment, and told me, honestly, that they probably would not have. I appreciated the candor, and it also told me everything I needed to know.

What makes this backwards, and not merely unfortunate, is that once you understand why those flags exist, most of them are not red at all, they are about as green as a signal gets. A short stint can mean someone took a hard turnaround job and left the day the work was finished. A gap can mean someone bet on themselves and built something that did not pan out, which is its own kind of education and usually a better one than the job would have been. No degree can mean someone was too busy doing the actual thing the degree is supposed to prepare you for. The entire value of a good hire lives in the context behind the bullet points, and context is precisely what a front-line filter is designed to throw away before anyone has to think about it.

People keep telling me the answer is the network, that I should simply know someone on the inside, and I used to believe that as firmly as anyone does. What I would say now is that networks do not carry the weight they used to. I pinged the five most connected people I know (one of them runs a genuinely enormous organization), and the most candid of them offered to make an introduction with a word of caution attached – the pipelines are almost entirely automated now, and the introduction lands in the same machine everyone else hits. Your champion can walk you up to the front door, but the door is still a model, and the model does not care in the slightest that someone vouched for you.

What all of that has done, in practice, is turn job hunting into something much closer to cold-call sales than anything I would have recognized a decade ago. You find the opening the day it posts. You work out who the actual hiring manager is, you write to them directly, and then you follow up, routing around the system rather than through it, because the system was never built to find you, it was built to shrink a pile. That is a reasonable survival tactic for one person trying to land somewhere, and a quietly terrible outcome for a company. The candidates with the energy and the cleverness to route around your filter are not a representative sample of the ones you are losing. You are selecting for persistence at gaming a process, which is not the same thing as quality of work, and on a long enough timeline those two qualities drift very far apart.

I do not think the answer is to rip the machines out and pretend the volume is not real, because it is real, and a team taking in a million applications a month cannot read them all by hand. The answer is to remember what the tool is actually for. A screen should exist to help an overwhelmed team surface and prioritize, not to silently delete, and the gap between those two jobs is the whole difference between a system that helps you and one that quietly costs you the best hires you never knew you missed. Use the model to raise its hand rather than to close the door, route the gaps and the short stints to a human being instead of letting the system resolve them on its own, and every so often take the people you actually hired and came to rely on (the best bets you ever made) and run them back through your own filter to see how many of them it would have thrown out before you ever got the chance to meet them.

Because this is the part that should keep anyone who owns one of these systems awake at night. A rejection you can see is a rejection you can still fix, but the candidate you never saw does not show up in any report, does not appear in any metric, and never registers as a cost at all. The expensive failure of a bad filter was never the people it turned down. It was the people it made disappear, one quiet auto-rejection at a time, while the dashboard up on the wall went right on insisting that everything was working exactly as intended.