Kill Criteria (Premortem Plus Pre-Commitment)
A premortem only works if each predicted failure signal is bound to a pre-committed action
- Difficulty
- Easy
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 95%
Premortems — imagining the project has already failed and asking why — feel powerful but, in Duke's research with Maurice Schweitzer and Len Lodish, rarely change anyone's plan or behaviour. The value is not in improving the plan; it is in harvesting the early warning signals and converting each one into a kill criterion: a specific signal paired with a pre-committed action to stop, probe, or pivot. Without the pre-commitment you are just hoping you will act rationally when the signal appears — a hope that reliably fails, which is why people climb Everest in a blizzard.
Origin
The premortem itself is Gary Klein's technique (popularized via Daniel Kahneman). Duke's contribution is the finding that premortems alone don't change behaviour, and the pairing of premortem output with pre-committed kill criteria — research conducted with Maurice Schweitzer and Len Lodish at Wharton.
Core principles
- 01By the time you're doing a premortem you're almost certainly going to launch anyway — so use it to arm the exit, not to redesign the plan
- 02Sunk cost makes in-the-moment quitting decisions unreliable; the decision must be made in advance, when you're cool
- 03A signal without a bound action is not a kill criterion — it's a note
- 04Kill criteria also protect the operator: they give a salesperson or PM a defensible reason to stop
How to run it
- 1
Run the premortem prompt individually
Send each contributor a written prompt: imagine it is N months from now and this project/deal is dead. Looking back, there were early signals it was going to fail. What were they? Collect answers independently.
Pro tip Set the time horizon to the realistic full run of the effort (Duke used six months for an enterprise sales cycle).
- 2
Consolidate the signal list
Pool the signals into a single list of concrete, observable early indicators — things you could actually notice in the field, not vague fears.
Watch out Reject signals you cannot observe. 'Bad market fit' isn't a signal; 'the buyer refuses a demo and only asks about price' is.
- 3
Attach a pre-committed action to each signal
For each signal decide now what you will do when you see it: kill outright, or take a specific probing action whose answer determines kill-or-proceed.
Pro tip Two-tier actions work well: signals that are decisive get an immediate kill; ambiguous signals get a scripted probe with a kill branch.
- 4
Publish and enforce the criteria before launch
Make the criteria a visible team artefact before work starts, so that acting on them is following policy rather than an individual admitting defeat.
Pro tip Explicit criteria stop leadership from punishing people who correctly walk away from a doomed deal.
Watch out If leadership overrides the criteria the first time one fires, the whole mechanism is dead.
In the wild
Duke prompted a sales team's ICs to imagine a lead from an RFP/RFI that they had worked for six months and was now dead, and to list the early signals. Three that emerged: the RFI was clearly written with a competitor in mind; the customer would not take a demo and only wanted to talk price; and after the first few meetings they could not get a decision-maker in the room. Each was bound to an action — kill the price-only box-ticking deal outright; ask the competitor-shaped RFI directly whether they're already working with a competitor and how far down the road, then kill or proceed; offer executive alignment at the next meeting, and kill if they decline.
→ The team stopped pursuing dead deals indefinitely, and reps had a sanctioned, leadership-backed reason to walk away rather than getting in trouble for stopping.
Common mistakes
Running a premortem and stopping there
Research shows premortems on their own generally don't change the plan or the behaviour. The exercise feels like diligence but changes nothing unless you deliberately use it to alter the plan or set kill criteria.
Trusting yourself to act rationally when the signal fires
In the moment, sunk cost, endowment and identity all push you to continue. Hoping you'll behave rationally later is a hope that will not come true — hence the climbers who summit-push in a blizzard.
Is it for you?
Best for
Product teams, sales orgs and founders committing months of effort to a bet where quitting late is the expensive failure mode
Not ideal for
Exploratory research or early discovery work where the whole point is open-ended learning and premature kill rules would cut off optionality
From the transcript
“yes okay so a premortem is great but only if you attach a pre-commitment”
“what a premortem allows you to do is to set up kill criteria so kill criteria are just uh a set of signals that you…”
“the RF RFI was clearly written with a competitor in mind”
“use the portum to now create structure around those signals that you've spotted and commit to actions that you're going to take if you see…”
From the episode
This will make you a better decision-maker
Annie Duke (author of “Thinking in Bets” and “Quit,” former