A project plan describes how the work should go. A premortem asks the team to examine a hypothetical version that failed.
The exercise becomes useful when a failure scenario changes a commitment. That might mean reducing scope, assigning an approver, checking an assumption, or stopping before work starts.
A list of things to worry about leaves that work unfinished.
The method below adapts a planning discussion for a small service team. The worksheet, decision rules, and content-library example are editorial proposals. The example describes no real client or completed project.
Define the Project Before Imagining Its Failure
Write a short project statement before inviting scenarios. Name the intended output, its users, the proposed boundary, and the person who can authorize it.
For a fictional content-library launch, the statement might be: prepare a public library of approved service guides for prospective buyers. Exclude private support records, account-specific advice, and unapproved product claims.
Then define the failed outcome as a hypothetical prompt: “The library opened, but readers could not identify which guides were current, and the team could not maintain them.”
That sentence does not predict failure. It gives participants a shared situation to examine. Replace it if it describes the wrong project.
Atlassian's premortem play asks participants to consider what could go wrong, write answers silently, group similar ideas, and identify mitigation actions. Its guidance also includes owners, deadlines, and later review.
This adaptation turns the discussion into explicit project decisions. It does not assign risk probabilities or claim that completing the exercise makes the launch safe.
Invite People Who Can Challenge the Assumptions
Include the people responsible for producing, approving, using, and maintaining the output. For a small team, one person may hold several responsibilities.
Identify those responsibilities before the meeting. A writer cannot confirm publishing authority merely by attending. A project lead cannot assign an absent maintainer without checking availability.
Give participants the same short project statement and the evidence already available. Mark unresolved questions. Avoid presenting the preferred plan as an established fact.
For the fictional library, invite a content owner, a subject reviewer, a publishing owner, and a maintenance owner. A buyer-facing colleague may contribute likely reader tasks. Their account remains input to check, not a substitute for reader evidence.
Ask each participant what falls within their authority. Record approval or staffing questions that require another person. The premortem can expose those gaps. It cannot settle them by assumption.
Gather Independent Scenarios Before Discussion
Start with a silent writing period. Ask each participant to describe a plausible route from a project condition to an unwanted result.
Use this prompt: “The project reached the failed outcome. What condition could have contributed, and what would we observe first?”
Keep one scenario per entry. Describe the work or dependency involved. “Communication broke down” is too broad to assign. “The publisher received a guide with no named approver” gives the team something to inspect.
Collect entries before opening discussion. Preserve different accounts even when they point toward the same concern. A quiet participant's specific dependency should remain visible beside a project lead's broader concern.
Ask writers to label existing evidence separately from conjecture. A missing approval field in the draft workflow can be inspected. A prediction that nobody will approve anything remains a scenario.
Do not require numerical estimates when the team lacks a basis for them. The purpose here is to decide what to check or change.
Group Scenarios by a Cause the Team Can Address
Read the entries together. Group similar causes while keeping the original wording available. Do not group everything under “people,” “process,” or “technology” and lose the actual problem.
For the fictional library, useful groups might include approval authority, content ownership, claim evidence, release scope, and continuing maintenance.
Separate causes from outcomes. “Buyers read an outdated guide” is an outcome. “Nobody owns the guide after launch” is a proposed contributing cause.
A cause outside the team's control can still affect the decision. Record the dependency and identify what the team can do about it. That may mean waiting, narrowing scope, or preparing an alternative.
Avoid treating repeated mentions as measured likelihood. Several participants can share the same incomplete information. Record disagreement when the evidence does not resolve it.
Select discussion topics by their effect on the proposed commitment. An absent approver can block launch. A cosmetic preference can wait. Explain the ordering in words rather than inventing a precise risk score.
Use a Premortem Worksheet That Ends in a Decision
Complete a row for each scenario the team chooses to act on. Leave evidence gaps visible. Name the person responsible for obtaining the missing information.
| Scenario | Existing evidence | Early signal | Mitigation | Owner | Proceed, change, or stop decision |
|---|---|---|---|---|---|
| [Condition leads to unwanted result] | [Checked record, date, and limit; or unknown] | [Observation that starts action] | [Prevention or detection action] | [Named responsible person and action deadline] | [Choice, condition, and decision owner] |
Use the terms consistently. Existing evidence describes what is available now. An early signal describes something the owner will watch for. A mitigation describes an action, not an aspiration.
Give each mitigation a deadline before the next commitment it controls. Separate the action owner from the person authorized to change project scope.
“Maintain quality” cannot be completed. “Require an accepted source record before a guide enters the release list” gives an owner a condition to check.
Proceed means the authorized owner accepts the defined scope under recorded conditions. Change means the team revises the plan before commitment. Stop means the proposed work cannot proceed under the current conditions.
These labels are proposed worksheet conventions. They do not replace the team's established approval process or grant permission to publish.
Work Through a Fictional Content-Library Launch
The following rows illustrate possible findings for the fictional library. Every evidence statement belongs to the invented scenario. None reports a real team's records or performance.
In this example, the project lead can pause work and reduce scope. Publishing still requires a confirmed approver. Each action is due before the release list is approved.
| Scenario | Existing evidence | Early signal | Mitigation | Owner | Proceed, change, or stop decision |
|---|---|---|---|---|---|
| Guides wait for approval and miss the release boundary | Draft plan has no named final approver | A release candidate has no accepted approval record | Confirm an approver and keep unapproved guides outside release | Project lead | Change: confirm authority before setting the release list |
| Nobody accepts responsibility for a guide | Fictional inventory contains owner blanks | A candidate enters review without an owner | Assign ownership or remove the guide from scope | Content owner | Change: launch only guides with accepted owners |
| A claim reaches the library without support | A sample draft has a claim with no source | Reviewer cannot locate supporting evidence | Remove the claim or obtain suitable evidence before approval | Subject reviewer | Stop that guide until the claim is resolved |
| Current status becomes unclear after launch | Maintenance assignment remains unknown | A guide has no review trigger or current-state label | Agree on maintenance ownership and a status rule | Project lead | Stop launch until responsibility is accepted |
The last row can stop the whole launch because it challenges the project's ability to stay usable. The unsupported claim can stop one guide while other approved guides continue. The decision boundary matters.
The team could revise the project to a smaller collection with confirmed owners. That change needs an updated project statement and approval. It does not inherit permission from the original proposal.
No row claims a probability, a revenue effect, or a quantity of avoided rework. The evidence supports a planning decision within the fictional conditions.
Distinguish Prevention From Detection
A prevention action changes a condition before the unwanted result. A detection action identifies evidence that a condition has appeared. Record which action the team is committing to.
In the library example, removing an unapproved guide from the release list is a prevention action. Checking release candidates for missing approval records is a detection action. The check still needs a consequence.
Write the response beside the signal. If a candidate lacks approval, the publishing owner withholds it and notifies the project lead. A flagged row without a response leaves the decision open.
Check that the owner can perform the action. Confirm access to the relevant inventory, source records, or publishing controls. Do not assign a check that depends on inaccessible evidence.
If the mitigation creates another obligation, record it. A source register needs maintenance. A new approval step needs an available approver. Revisit the scope when the proposed safeguard exceeds the team's capacity.
Keep Unresolved Assumptions in the Project Record
Some questions will remain open. Record the assumption, the evidence needed, the owner, and the next decision it affects.
For the fictional library, an unresolved assumption might be that existing guides can be made public. The team must confirm permissions and content suitability before treating those guides as release candidates.
Another assumption might concern reader tasks. A colleague's account can inform the initial plan while leaving validation incomplete. Label that limitation instead of promoting it into verified demand.
Keep a record of the resulting choice. Atlassian's decision documentation template provides fields for context, options, actions, and outcome. The unresolved-assumption register here extends that approach for this planning exercise.
Link actions into the active project management workflow. Keep the reasoning and evidence in the premortem record. Task completion alone does not show whether the underlying assumption held.
Set Triggers for Reopening the Plan
Name a trigger for each condition that could alter approval. A missing maintenance owner might block launch now. A later ownership change might require another review.
Useful triggers come from the project itself: scope expansion, an unresolved approval, a changed source, an unavailable owner, or evidence that a proposed mitigation cannot run.
Also set a scheduled check before the next commitment point. The owner should review open assumptions before approving production or release, according to the team's process.
Use the weekly review framework to surface due actions. Keep the premortem available when deciding whether the project can proceed.
If the project includes automation, use an operations automation audit to examine the affected workflow. The premortem does not establish that automation is feasible or that a tool performs correctly.
End With a Revised Commitment
Read the project statement again after the discussion. Record what changed, who authorized the change, and which conditions still block work.
Preserve the original scenarios and append dated updates. Later observations should show whether the team acted on the concern, changed its interpretation, or accepted a remaining uncertainty.
A completed premortem is a planning record. It does not prove delivery, reader usefulness, recovery readiness, or successful maintenance. Check those outcomes through the work itself.
Give each accepted concern a decision.
Sources
- Atlassian Team Playbook: Premortem. Silent scenario writing, grouping, mitigation actions, ownership, and revisiting the record. Checked 2026.10.28.
- Atlassian: DACI Decision Documentation Template. Context, options, action items, and recording the outcome. Checked 2026.10.28.