Originated by Gary Klein (Performing a Project Premortem, 2007) on the prospective hindsight work of Deborah J. Mitchell, J. Edward Russo, and Nancy Pennington (1989).
Before launch, has each team member privately imagine the project has already failed and write why, then pools the stories to expose risks that enthusiasm was hiding.
A premortem on a sales-software rollout, from the imagined failure to the decisions
Play the session, or select a step. The team starts from a failure that has already happened, writes its causes alone, pools them without names, and clusters them; the sort below makes the decisions. Illustrative.
- Set the scene. Two weeks before the contract is signed, the project lead opens the meeting: it is next March, the new sales software has failed, and the reps are back on spreadsheets. Nobody is asked whether it might fail; it already has.
- Write alone. Each person writes, in silence, the story of how it failed. Writing alone keeps the most senior voice in the room from setting everyone else's story.
- Pool the stories. The facilitator reads every story aloud without names and lists each cause on the wall: 23 in all, several of them things nobody had said in a planning meeting.
- Cluster. The causes fall into three clusters: the reps never adopt it (9 causes), the migrated account data is too dirty to trust (8), and things outside the team slip (6).
- Decide. Each risk gets a decision and an owner: fix the biggest before signing, watch the unlikely but serious ones with a named trigger, and accept the small ones. The sort below puts six of them on the grid.
Six risks from the premortem, sorted by what to do about them
The first risk is placed as a worked example: likely and serious, so fix it now. Select each other risk, then the box for how likely it is and how much damage it would do; Show the sort compares yours with the team's. Illustrative.
- Not placedPlacedFix nowWatchAccept
- Not placedPlacedFix nowWatchAccept
- Not placedPlacedFix nowWatchAccept
- Not placedPlacedFix nowWatchAccept
- Not placedPlacedFix nowWatchAccept
- Not placedPlacedFix nowWatchAccept
Damage if it happens
How likely
Your sort matched 0 of 6.Your sort matched 1 of 6.Your sort matched 2 of 6.Your sort matched 3 of 6.Your sort matched 4 of 6.Your sort matched 5 of 6.Your sort matched 6 of 6.
- Fix now
- Likely and serious: change the plan before signing, with an owner and a date.
- Watch
- Unlikely but serious: name an early sign and who acts when it appears.
- Accept
- Minor either way: write it down, tell the people affected, and move on.
The sort
- Reps keep spreadsheets: Fix now. Five of six stories had reps working around the system. The fix: managers run pipeline reviews only from the new system, starting week one.
- Dirty account data: Fix now. Duplicate accounts would make every report wrong. The fix: a two-week cleanup before migration, owned by sales operations.
- Sales chief leaves: Watch. Unlikely this year, but the rollout has no other sponsor. The trigger: any sign of a search, and the finance chief steps in as sponsor.
- Integration vendor slips: Watch. The vendor has hit its dates before, but a slip would stall billing. The trigger: a missed milestone, and the team switches to the manual export.
- Training runs late: Accept. A two-week slip in training costs little; the old system runs in parallel until it is done.
- Dashboards look dated: Accept. Nobody will stop using the system over it, however likely it is. Logged for the second release.
An example
A company was two weeks from signing a three-year contract for new sales software, and the project team was confident. The project lead booked an hour with the six people closest to the rollout and opened it with one sentence: it is next March, the rollout has failed, and the reps are back on spreadsheets.
Everyone spent ten minutes writing, alone, the story of how it had failed. The facilitator read the stories aloud without names and listed 23 causes on the wall. Five of the six stories had reps quietly keeping their own spreadsheets, something nobody had raised in three months of planning, and four described duplicate accounts making every report wrong. The causes fell into three clusters. Before signing, the team added a two-week data cleanup and a rule that managers would run pipeline reviews only from the new system. It named a trigger for the two unlikely but serious risks, and logged the minor ones. The contract was signed a week late.
Why it works
Asked what could go wrong with a plan they believe in, people offer a few cautious worries and feel disloyal for those. Told that the plan has already failed and asked to explain why, they find reasons far more readily. That shift, called prospective hindsight, is the core of the method: explaining an event that has happened comes easier than imagining one that might.
The format adds two protections. The failure is a given, so raising a risk is the assignment and no longer a sign of doubt about the project; the people with bad news can speak without standing out. And writing alone before anything is shared keeps the first or most senior speaker from shaping the rest. Pooled, the independent stories show which risks several people saw without prompting, which is a stronger signal than any one person's warning.
The thinking behind it
Gary Klein Gary Klein, a research psychologist and a pioneer of naturalistic decision making, studied how fireground commanders and other experts decide under time pressure. He devised the premortem and set it out in a 2007 Harvard Business Review article, Performing a Project Premortem: before a project starts, the team imagines that it has failed and explains why. It fits his wider finding that experienced people test a course of action by imagining it played out.
Deborah J. Mitchell Deborah J. Mitchell, with J. Edward Russo and Nancy Pennington, ran the experiments on prospective hindsight reported in Back to the Future: Temporal Perspective in the Explanation of Events (Journal of Behavioral Decision Making, 1989). Imagining that an event has already happened made it easier for people to generate reasons for it. That finding is the mechanism the premortem puts to work.
Daniel Kahneman Daniel Kahneman, a psychologist, showed with Amos Tversky that intuitive predictions follow representativeness and neglect base rates, and proposed correcting them by starting from the record of similar cases, the outside view. Thinking, Fast and Slow (2011), his popular synthesis of that work, also carries Klein's premortem, which sits beside that correction as a second check a team can run on a plan it already believes in.
How to run it
- Hold it before the commitment. Schedule it after the plan is clear and before the contract, launch, or hire is final.
- Set the scene as a fact. Give a date and say the project has failed, then stop; nobody debates whether it might.
- Have everyone write alone. Allow ten minutes for each person to write the story of the failure in silence.
- Pool the stories without names. Read every story aloud, or collect them, and list each cause where everyone can see it.
- Cluster the causes. Group the causes into a few themes and note how many people raised each one.
- Decide each risk. Mark each one fix now, watch with a named trigger, or accept, and give every fix an owner and a date before the commitment.
Where it goes wrong
The session slides into a debate about whether the plan will fail. Someone senior defends it, and the room goes back to polite worries. The facilitator has to hold the frame: the failure has happened, and the only question is why.
A second failure is skipping the silent writing to save time. Spoken aloud in turn, the first two stories set the theme and the rest echo them. Keep the ten minutes of writing, even in a short meeting.
The third is a long list and no decisions. The team leaves with 23 risks on a wall and nothing changed. End the hour with each cluster marked fix now, watch, or accept, and the fixes added to the plan before anyone signs.
When to use something else
- Red Teaming. When the plan faces an adversary, such as a competitor or an attacker, and needs a team playing that side.
- Failure Mode and Effects Analysis. When the process is known in detail and each step's failure modes need scoring for severity, likelihood, and detection.
- Inversion. When one person wants to test a goal quickly by asking what would guarantee failure, without a session.
Where it sits in SOLVE eX
Go to the source
- Performing a Project Premortem, Harvard Business Review, September 2007, 2007
- Back to the Future: Temporal Perspective in the Explanation of Events, Journal of Behavioral Decision Making 2(1), 1989
- Thinking, Fast and Slow, Farrar, Straus and Giroux, 2011