Three hires, three different roles, possibly three different recruiters. Each one impressive in the process, each one gone or going within a year, and afterwards you can list the warning signs you talked yourself past.
At that point it is not bad luck. Something in your selection is stable, and it is you.
The mechanism
Interviews are supposed to be evaluations. Mostly they are recognition events.
Within a few minutes you have decided whether this person is any good, and everything after that is evidence-gathering for a conclusion you already reached. This is well documented and knowing about it helps less than you would hope.
What decides the first few minutes is not competence. It is whether the candidate matches something in you — and there are two versions of that, which fail in different ways.
The mirror. They are like you. They think in the same order, they find the same things obvious, the conversation is unusually easy. You come out of it energised and call it chemistry. What you have actually found is someone with your strengths and, more importantly, your blind spots. It works well for a year and then a whole category of problem goes unnoticed by both of you.
The compensation. They are what you wish you were. They have the certainty you lack, or the polish, or the calm. This one is more dangerous, because the trait you are drawn to is exactly the trait you cannot evaluate — you do not have it, so you cannot tell whether theirs is real. Confident people get hired by uncertain people at a rate that has nothing to do with their ability.
The tell
Look back at the three hires and find what impressed you in the first meeting each time.
Not what was on the CV. What made you sit forward. If the same quality appears all three times, you have found the thing that gets past your judgement, and it will do it again next month unless something in the process changes.
Common ones: a particular kind of directness, an ease with senior people, evident intelligence, a story that resembles yours, or straightforward enthusiasm about you and the company.
What actually changes the outcome
Structure helps, and I would rather give you the boring version that works than something clever.
Write the criteria before you meet anyone, including what you will not compromise on, and score against it afterwards.
Have someone unlike you in the process with real veto power. Not to be consulted — to be able to stop it.
Ask the question you avoid. There is usually one — about a gap, a departure, something inconsistent — and the reason you skip it is that the conversation is going well and you do not want to break it. That reluctance is the signal. Ask it.
Wait a day before deciding. The recognition effect fades. What is left the next morning is closer to your actual assessment.
Where the general answer stops
All of that is real and it will improve your hit rate. What it does not do is tell you which of the two patterns is yours, and that determines what you are correcting for.
If you hire mirrors, you need someone in the process who thinks differently, and you need to stop reading ease of conversation as evidence of quality. If you hire compensations, you need a way to evaluate the specific trait you are drawn to, because your read on it is unreliable in a predictable direction.
Those are near-opposite corrections. Getting it backwards is worse than doing nothing.
Which one you run comes down to what you value in yourself and what you quietly believe you are missing — and most people are considerably clearer on the second than the first.