A premortem tells a team the plan has already failed and asks each person to write down why, before the plan starts. It is one of the most widely adopted decision techniques in business and one of the least tested, and the gap between what it is known to do and what it is credited with deserves stating plainly.
The answer, in one line
A meeting held before a plan starts, in which the team is told to imagine it is a year later and the plan has failed, and each person writes independently why.
Definition#
Premortem: a meeting held before a plan is executed, in which the team is told the plan has failed and each member writes down why, so that doubts are stated as explanations rather than as objections. The method is Gary Klein's, set out in 2007.
The mechanism it proposes#
The claim is social before it is cognitive. Asking a room what could go wrong invites objection, and objection has a cost: the person raising it looks negative, or unsupportive of a decision the senior people have already made. Asking why it failed makes the failure a premise of the exercise. Producing a reason is then compliance rather than dissent, and the people who would otherwise say nothing have cover to speak.
The cognitive half comes from prospective hindsight. Mitchell, Russo and Pennington found in 1989 that asking people to explain an event described as having already happened produced about 30 per cent more reasons, and more specific ones, than asking them to explain an event described as possible. Certainty appears to license detail that uncertainty suppresses.
What is known and what is not#
What is known is the 1989 finding about reason generation. What is not known is whether any of it improves a decision or a project. Klein's Harvard Business Review piece is a procedure set out by the person who devised it, with reasoning and adoption behind it and no trial. No study has compared projects that ran a premortem against projects that did not.
That places it in the same company as most of the structured techniques reviewed by Chang and colleagues in 2018, where the psychological rationale was found weak and the accuracy evidence absent. The premortem has a better underlying experiment than most of them. It still has no outcome evidence.
How to run one#
- State the failure as fact: it is a year from now, this went badly, and everyone knows it.
- Everyone writes independently, before anybody speaks. The independence is the part most often dropped and the part that keeps the first speaker from setting the range.
- Collect reasons one at a time round the table rather than in open discussion.
- Keep the list and revisit it. A premortem whose output is never looked at again is a ritual with a good origin story.
Where it fits with AI decisions#
It is most useful where a recommendation arrives looking finished. A machine's output carries no visible uncertainty, and the fluency reads as confidence, which is the condition under which teams stop generating reasons against it. Running the exercise on the recommendation rather than on the plan is a small adaptation: it is a year from now, we took this recommendation, and it was wrong. The related question of when to add deliberate delay is at which decisions should become slower.
What this does not establish#
That a premortem improves outcomes, reduces project failure, or surfaces the reasons that matter rather than simply more reasons. Generating more explanations is not the same as generating better ones, and a longer list can create its own false confidence that the risks have been covered.
Key sources
- Klein, G. (2007). Performing a project premortem. Harvard Business Review.
- Mitchell, D. J., Russo, J. E. and Pennington, N. (1989). Back to the future: temporal perspective in the explanation of events. Journal of Behavioral Decision Making, 2(1).
- Chang, W. et al. (2018). Restructuring structured analytic techniques in intelligence.
Related SuperSkills research#
Explainer · SS-2026-322 · Graded against the published rubric
Hirji, R. (2026). What is a premortem?. The SuperSkills evidence base, SS-2026-322. https://thesuperskills.com/research/what-is-a-premortem. Last reviewed 26 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work