I’ve made three genuinely bad hires. Not “didn’t work out” bad — properly bad. The kind where you spend 3 months managing a situation that should never have existed.
Each one was different. Each one was avoidable. And looking back, each one had red flags I either didn’t see or chose to ignore.
Here’s what happened, what I missed, and how AI would have changed the outcome.
Bad Hire #1: The Brilliant Generalist
The situation: I hired someone with an extraordinary CV. Big companies, impressive titles, glowing references. In the interview they were articulate, thoughtful, and clearly intelligent.
What I missed: Everything on their CV was about scale — managing large teams, inheriting established processes, optimising existing systems. I needed someone to build from nothing. These are completely different skills and I didn’t screen for the difference.
What happened: They were paralysed without process. Every decision needed sign-off. Every ambiguity became a blocker. After 10 weeks we agreed it wasn’t working.
The prompt that would have caught this:
“I’m interviewing a candidate for [role] at an early-stage startup. Their background is primarily at large, established companies. Generate 4 interview questions specifically designed to reveal whether they can operate effectively in an unstructured, resource-constrained environment — not whether they’ve managed complexity, but whether they’ve created something from scratch. Include a scoring rubric.”
Bad Hire #2: The Culture Fit I Confused With Competence
The situation: I hired someone I genuinely liked. We had the same sense of humour, the same references, the same way of looking at problems. The interviews felt more like conversations than assessments.
What I missed: Liking someone is not the same as respecting their work. I had almost no evidence of their actual output — I had evidence that they were smart and fun to talk to.
What happened: The work was consistently mediocre. Not bad enough to fire quickly, good enough to keep hoping it would improve. It didn’t.
The prompt that would have caught this:
“Generate a work sample test for a [role] candidate. The test should: take no more than 2 hours, produce a tangible output I can evaluate, and require the specific skills the role needs day-to-day. Include a clear rubric for what good, average, and poor output looks like. The test should be hard to pass on intelligence alone — it should require actual experience.”
Bad Hire #3: The Reference I Didn’t Check Properly
The situation: I called two references. Both were positive. I made the offer.
What I missed: Both references were chosen by the candidate. One was a friend. One was a former colleague who owed them a favour. I asked soft questions and got soft answers.
What happened: Three months in I discovered a pattern of behaviour that multiple people in their previous network knew about. If I’d asked the right questions to the right people, I’d have found out before hiring.
The prompt that would have caught this:
“Generate 6 reference check questions for a [role] candidate that are difficult to answer positively if the candidate has significant weaknesses. Include questions that ask the reference to compare the candidate to others, describe a specific failure, and explain what kind of manager or environment brings out their worst. Also suggest how to find references beyond those provided by the candidate.”
The Pattern Across All Three
In every case I optimised for how the candidate made me feel in the room rather than evidence of what they could actually do.
AI doesn’t fix gut feel — but it forces you to generate evidence. When you use structured prompts to build your interview questions, your work samples, and your reference checks, you create a process that surfaces information your instincts would have filtered out.
The three hires I’ve made since rebuilding my process have all worked out. That’s not a coincidence.
Get the Full Hiring Prompt Library
15 prompts covering every stage of the hiring process — free download.