The premortem, and the room where it fails
Kahneman called it his favorite debiasing tool. It takes twenty minutes and it stops working the moment the most senior man in the room speaks first.
$ eldr --cite klein-2007
The tool
Gary Klein published it in the Harvard Business Review in 2007, two pages, no apparatus. The procedure is one sentence long: before you commit, assume the plan has already failed, and have everyone write the story of how.
Not "what could go wrong". That question produces a list of risks, ranked by how easy they are to say out loud, and it changes nothing. The premortem fixes the outcome as a fact and asks for an explanation, which is a completely different cognitive task — you are no longer defending a plan, you are performing a postmortem on one, and people are dramatically better at explaining a failure that has happened than at imagining one that might.
Kahneman has praised it repeatedly, which is a high bar from a man who spent a career cataloguing why the rest of them do not work. I have not found a primary source for the widely repeated claim that he called it his favorite, so I am not making it.
Why it works, mechanically
Two effects, both documented, neither mysterious.
- Prospective hindsight. Framing an event as already settled increases the number and specificity of reasons people can generate for it. You get named mechanisms instead of categories of worry.
- Permission. In a normal planning meeting, the man who raises the fatal objection is the man slowing everyone down. In a premortem, raising the fatal objection is the assignment. The social cost of dissent drops to zero and the dissent arrives.
Everything the counter-case block in this product does is this, run against our own position before you see it. That is not decoration or a legal hedge. It is the one part of a read that can be checked, and it is a premortem with a smaller sample size.
The claim
A premortem's entire value comes from the fact that nobody has spoken yet, and it is destroyed by the first sentence out of the most senior man in the room.
Written independently, then read aloud, it works. Opened as a discussion, it produces a polite version of whatever the boss said first, and a document that will be cited later as evidence that the risks were considered. That second version is worse than not running it, because it manufactures the paperwork of diligence without the content.
Falsifiable and cheap to test: run it both ways on two real decisions. Count distinct failure mechanisms named. In my experience the written-first version produces two to three times as many, and the ones it produces are the ones nobody wanted to say. If you run this and get the same output both ways, the claim is wrong in your organization and I would like to know why.
Where it fails on a man deciding alone
Here is the limitation that matters for anyone using this product, and it is rarely mentioned in the corporate write-ups: a premortem you run on yourself inherits your own blind spot. The whole mechanism depends on other people having information and incentives you do not have. Alone at a kitchen table at 11pm, you will generate the failures you were already worried about, in more detail, and mistake the detail for coverage.
The partial fixes, in order of how well they work:
- Get one other person to write theirs before hearing yours. Even one. A colleague, a brother, anyone with any exposure to the situation.
- Write the failure story in third person and in the past tense. "In November he ran out of money because…" Distance measurably improves the quality of the reasons; this is the same effect that makes people give better advice to others than to themselves.
- Require a mechanism, not a mood. "It didn't work out" is not a premortem entry. "The two engineers I am relying on resigned and I had no budget to retain them" is, and that exact sentence appears in a debrief in this library where it turned out to be correct.
The case against the tool
A premortem is an anxiety generator with a nice reference. Give a naturally cautious man a formal procedure for enumerating failures and he will produce twelve of them, rate them all as plausible, and do nothing — and now his inaction has documentation. There is no published evidence that premortems improve decision outcomes; the evidence is about the quality and quantity of risks named, which is not the same thing and never has been.
Both halves of that are true. The evidence base really is about surfaced risks rather than realized outcomes, and anyone claiming otherwise is overselling a two-page article from 2007.
The defense is narrow and I will keep it narrow. A premortem is not a decision procedure and should never be used as one. It is an input generator, and it needs a stopping rule bolted to it: enumerate the failures, pick the two with the highest product of likelihood and cost, do something concrete about those two this week, and then proceed. Without the stopping rule it becomes exactly what the objection says it is.
That is why every read here pairs the counter-case with a window and a date. The argument against your position is only useful if something closes afterward.
The reference
Klein, G. (2007). Performing a Project Premortem. Harvard Business Review, 85(9), 18–19. Two pages. Free everywhere. If you read one thing on the reading list, read that one and skip the rest.