A superintendent interviews well. He works through a tough phasing problem on a hospital job and tells you how he kept the inspector happy when the schedule slipped. You hire him, and for ninety days he confirms everything you saw in the room. Then, around month fourteen, the cracks show. A foreman stops bringing him problems. A job finishes three weeks late for reasons nobody can quite name. By the time you trace it back to a hiring decision from over a year ago, you have opened two more searches and made the same call twice more, and those seemed to go fine…
Learning from bad hires is almost impossible, and not because the people doing the hiring are careless. The teams making these mistakes care about hiring well. They are experienced. They run real interviews and ask real questions. The problem is that the environment they are learning in gives them almost nothing to learn from. The signal comes back late, mixed with noise, and disconnected from the decision that caused it. So the same movie plays again. A bad hire, a post-mortem where everyone agrees it will not happen again, and six months later, the same scene with a different face.
I have watched capable teams run this loop for years without ever naming it, and the reason it persists is that the standard explanations are wrong, and the wrong explanations buy the wrong fixes. The problem is not effort. Your people are not failing to try hard enough or care enough. Hiring, by its structure, is one of the hardest environments to build real skill in that a working professional will ever operate in, and construction makes a hard environment harder.
Why learning from bad hires is so hard
There is a useful distinction from the psychologist Robin Hogarth that explains the whole problem. He separated learning environments into two kinds: kind and wicked. A kind learning environment gives you feedback that is fast, accurate, and clearly tied to what you did. Chess is kind. You make a weak move, you lose a piece, and the connection between the two is visible inside a minute. A surgeon closing a wound sees the result. A free-throw shooter watches the ball go in or rim out. In environments like these, practice works the way intuition says it should: you do the thing, you see the result, you adjust, and over thousands of repetitions you get measurably better.
A wicked learning environment gives you none of that. The feedback is slow, or contaminated, or missing, or impossible to trace back to its cause. And here is the part that should stop you: in a wicked environment, experience does not build skill. It builds confidence. The practitioner gets surer of their judgment while their judgment gets no better, because the environment never corrects them. The reps accumulate, and without a scoreboard not one of them teaches anything.
Hiring is a wicked learning environment in four specific ways, and it is worth walking through each one because the fix has to address all four.
The feedback is delayed. You make a decision in the interview room, and the real signal about whether you were right takes weeks, often months, sometimes more than a year to emerge. By then the conditions of the decision are cold. You cannot feel the connection between what you weighed and what you got, because too much time and too many other events sit between them.
The feedback is noisy. When a hire works out, was it the person, or the project, or the manager who happened to coach them well, or an estimating department that handed them clean numbers, or a market that gave them room to ramp? When a hire fails, same question in reverse. Performance depends on a dozen variables, and the candidate is only one of them. You almost never get to isolate the part you actually decided on.
The feedback is incomplete. You hired one person. You will never see how the other finalist would have run that same job under those same conditions. There is no control group. You are grading your decision against an outcome you can observe, but you are missing the comparison that would tell you whether the decision was actually good or merely survived.
And cause and effect are nearly impossible to connect. Who, in the middle of a bad third quarter, traces the problem back to an interview from fourteen months earlier? The bad quarter has its own immediate explanations: the weather, the owner, the subs, the takeoff. The hiring decision that seeded it is so far upstream that it never enters the conversation.
Put those four together and you get the sharpest finding in this whole area, which comes out of Daniel Kahneman's work and the research that followed it. The unstructured interview, the kind nearly every construction leader runs and trusts, predicts actual job performance at a level barely better than chance. A coin flip in a sport coat. Decades of data say this, and it is not close. Yet the people running those interviews feel more accurate every year, not less.
Experience in a wicked learning environment does not build expertise. It builds confidence. And confidence with no scoreboard is just a feeling that hardens over time.
That is the trap. The people in the room are not bad at their jobs. The job, as structured, simply cannot teach them to be good at it. The feeling of getting better is real, and the improvement that should sit underneath it never arrives.
Why construction makes a hard problem harder
If hiring is a wicked learning environment everywhere, construction tightens every screw.
Start with volume. A project executive might make three to five real hires in a year. A surgeon does hundreds of procedures. A chess player runs through thousands of positions. You cannot build statistical intuition on three samples a year, and across a thirty-year career you might see a few hundred decisions, most of them clouded by every problem above. There are simply not enough reps to overcome the noise, even in principle.
Then there is the length of the performance cycle. Whether a superintendent actually runs a job well does not fully reveal itself for twelve to eighteen months. A project has phases. Someone can look strong through structure and reveal themselves at closeout, or coast through the easy stretch and come apart when the schedule compresses. By the time the verdict is in, the interview that produced it is ancient history, filed under a decision you can barely reconstruct.
Accountability scatters, too. An owner, a PM, an operations director, sometimes an HR coordinator all touch a single hire. When it goes well, everyone remembers their part in it. When it goes wrong, the responsibility diffuses until no single person owns the miss clearly enough to learn from it. A decision that belongs to everyone belongs to no one, and a miss that belongs to no one teaches nobody.
There is a structural gap underneath all of it. The people running the interview are often evaluating candidates for roles whose daily reality they sit one step removed from. The interview room rewards presentation: the candidate who frames a story cleanly, who reads confident, who has the vocabulary. The field rewards execution: the person who holds a schedule when three trades are stacked on top of each other and the owner is calling daily. Those are different tests. The interview measures the first and pretends it measured the second.
And construction carries one more thing that most industries do not, which is the deep cultural legitimacy of the gut call. "I know a good super when I see one" is a real belief, held by real people who have been in the trade their whole lives, and who are sometimes right. That is what makes it so hard to dislodge. A belief that is wrong every time is easy to kill. A belief that is right often enough to keep its credibility, and wrong often enough to cost you, is the most stubborn kind there is. It cannot be falsified from the inside, because every hit confirms it and every miss has an alibi.
The miss always has an alibi
Even when feedback does come back, something quietly corrupts it on the way in. The technical name is outcome bias, and the plain version is this: people judge the quality of a decision by how it turned out, not by how it was made.
Watch how that plays out in hiring. A hire fails, and the story writes itself around the candidate. He was not who he said he was. He interviewed better than he worked. The market was thin, so it was a defensible risk. The process is left standing, untouched, exonerated. A hire succeeds, and the process gets the credit, regardless of whether the interview actually predicted anything or the person simply landed in a situation that made them look good. Heads, the candidate's fault. Tails, the process was sound. The process never loses.
That is a self-sealing loop, and it is the reason the same mistakes survive decade after decade. The process is never put on trial, so it is never convicted, so it is never changed. The interviewer who trusted his gut and got it right once carries that win forever. The three times the same gut missed get rewritten in memory: there were concerns, there was an override, the situation was unusual. Hindsight edits the record until the instinct comes out looking sharper than it was. Nobody is lying. Memory just files things in the direction that flatters the judgment doing the remembering.
You cannot learn from a process you never allow to fail. And outcome bias guarantees the process never fails, because the outcome always gets pinned on something else.
What learning would actually require
Picture what the feedback environment would have to look like for real learning to happen. Not the solution yet, just the conditions. What would have to be true for a hiring team to actually get better over time?
You would need performance standards defined before the hire, not after, so the bar is set independently of whoever you end up choosing and cannot drift to fit the person you picked. You would need the interview itself preserved: the scores, the concerns, the specific reads, written down and kept, not dissolved in the rush to get an offer out the door. You would need attribution, which is the uncomfortable one. When a hire underperforms, someone has to go back and ask what the interview actually said about this person, and whether anyone listened to it. And you would need a decision log: what you knew, what you weighed, what you decided, and on what basis. So that a year later, when the cracks show, there is an actual record to audit instead of a fog of competing memories.
Almost no company does any of this. Not out of negligence. Because nobody ever told them that this is what learning from hiring requires. They were handed the belief that good interviewers are simply good judges of people, and that experience would do the rest. The environment was never built, so the learning never compounded, and a competent team kept running the same play because the field never gave them a reason it would not work.
What structure actually does
This is the work, and it is worth being plain about what it is and is not. A structured search, behavioral interviewing anchored to the real demands of the role, an assessment layer that measures what the interview cannot, and a documented debrief that survives the decision: none of that is bureaucracy. It is the infrastructure that turns a wicked learning environment into a kind one. It is the difference between a framer who snaps a chalk line and one who eyeballs every cut and trusts it will come out plumb.
A decision log means you can audit the decision later. Behavioral anchors mean there is something concrete to trace a failure back to, instead of a vague memory of a good conversation. Preserved assessment scores mean that across enough hires, patterns start to surface that no single decision could ever show you. None of this replaces your judgment. It builds the scoreboard that lets your judgment improve, rep over rep, the way it already does in every kind environment you have ever gotten good in.
This is what Ambassador Group builds for the clients it works with, and it is also what a well-run internal process should build for itself. The aim is to surround the call with enough recorded reality that the next call can learn from this one. The goal is to compound judgment over time, not to take it out of anyone's hands.
This is fixable, and it is fixable from where you sit. The teams that struggle here are not careless or short on instinct. They have spent their careers working inside an environment that was never built to let good decisions compound.
The problem was never that you could not see people clearly. It is that nobody ever built you a process where seeing clearly was allowed to add up.