Imagine it's nine months from now. Your AI initiative has failed. Revenue targets were missed, the team is demoralized, and leadership is asking what went wrong.

Now — before any of that happens — write down exactly why it failed.

That's the premortem. And it's the most powerful tool most leadership teams never use.

Why premortems work

The premortem, developed by psychologist Gary Klein, works because it sidesteps the optimism bias that infects most planning processes. When you imagine failure as a certainty and reason backwards, you surface risks that a forward-looking plan will almost always miss.

In AI initiatives, this matters enormously. Most AI project teams are excited about the technology, motivated by leadership attention, and optimistic about outcomes. That optimism is energy — but it's also a filter that makes it easier to dismiss legitimate risks as edge cases.

A premortem forces you to take those risks seriously before they become real.

How to run a 60-minute premortem

Step 1 (10 minutes): Future failure framing. Tell the team: "Assume it's [date]. This initiative has failed — significantly, not slightly. What happened?" Everyone writes their failure scenarios silently and independently.

Step 2 (20 minutes): Read aloud and capture. Each person shares their top failure scenario. No debating, no defending. Just capture. You're looking for patterns — the same risk showing up in multiple people's scenarios is your highest-priority signal.

Step 3 (20 minutes): Countermeasures. For the top three to five risks, define the specific, named action that prevents each one. Who owns it? By when? This transforms the premortem from a conversation into a prevention plan.

Step 4 (10 minutes): Kill risk scoring. Rate each identified risk: how likely is it (1–5), and how catastrophic would it be (1–5)? Anything scoring 16+ is a blocker — it needs to be addressed before launch, not after.

What the premortem tells you

The scenarios your team generates will usually fall into four categories: data quality, organizational resistance, unclear ownership, and misaligned success metrics.

If all four show up in the first 10 minutes, you're not behind. You're further ahead than 90% of organizations attempting this. You now know your kill risks before they kill you.

Run the premortem. Then fix what you find.