1. LandOffer
  2. Blog
  3. All articles
  4. Run a Project Premortem Before the Risks Become Real

Onboarding & Teamwork

Run a Project Premortem Before the Risks Become Real

Turn independently imagined failures into evidence-checked priorities, funded responses and early signals. Follow a complete fictional board and planning review that keeps residual risks visible.

9 min readLandOffer team

Run a project premortem by asking people to imagine that the planned project has failed, then write possible reasons independently before discussing them. Treat those reasons as hypotheses, check them against what is known, and select responses the team has authority and capacity to carry out. Give each selected risk an owner, an early signal, and a review point while preserving the uncertainties that remain.

This LandOffer workplace guide develops a fictional premortem board and a completed response-planning review for an internal test-data catalog pilot. All people, project facts, hypotheses, capacity agreements, and follow-up observations are invented examples. No actual failure, project workshop, or risk-reduction experiment was observed while preparing this article.

Project planners independently examining possible weak points in an unfinished glasshouse model before sharing their hypotheses

Frame an imagined failure without inventing an incident

A premortem looks forward from an imagined unsuccessful outcome. A postmortem examines an event that actually happened. A retrospective reviews how completed work went. A risk register maintains ongoing risk information. The premortem is a bounded discovery activity that can feed planning and later risk tracking; it does not replace those other practices.

Gary Klein described the project premortem in Harvard Business Review in September 2007. Label each failure statement as imagined, and distinguish it from the current evidence used to assess it.

In the fictional case, facilitator Maren is preparing a limited internal pilot of a test-data catalog. The catalog would help engineers find approved synthetic examples and identify their owners and update dates. The team is considering a metadata feed, but has not approved production rollout or access to real customer records.

Maren's prompt is: “Imagine we reviewed the pilot after its first month and decided it could not continue usefully. What might have made that happen?” This is a hypothetical scenario, not a prediction that the pilot will fail or evidence that it has already done so.

Define what the exercise can influence. This team can change pilot scope, metadata requirements, support arrangements, and investigation priorities. It cannot guarantee another team's delivery, approve a corporate budget change, or assign work to people who have not accepted it.

Let people generate concerns before debating them

Give participants a short description of the proposed scope, current dependencies, and open decisions. Ask them to write independently first, including concerns that contradict the favored plan. Provide a written contribution option for people who cannot join the live session or prefer time to formulate their thoughts.

Atlassian's premortem play uses silent idea generation, grouping, prioritization, discussion, and actions with owners and deadlines. Its prompts and voting mechanics are options for facilitating the exercise, not measurements of how likely each failure is.

In the fictional workshop, Diego writes that catalog entries could look current when nobody owns their refresh. Iris worries that requests for help could accumulate without agreed coverage. Owen imagines a metadata-feed change invalidating the integration. A fourth participant worries about a wider organizational funding freeze. Maren collects each concern before the team discusses solutions.

Do not immediately ask the author of a concern to defend a probability or volunteer to fix it. First ask what the imagined failure means and what conditions would make it plausible. Preserve the independent inputs before debate so an early confident response does not replace the concerns others brought.

The AHRQ Premortem Tool also uses independent future-failure thinking before shared prioritization and intervention planning. That tool is designed for a healthcare improvement program. Its clinical setting and prescribed participants do not become requirements for this software project; the relevant method is creating space for concerns before choosing responses.

A completed hypothesis-to-priority board

Here is the fictional board after clarification. “Known basis” means a current fact inside the invented case, not a real-world observation by this article's author. The failure statements remain imagined outcomes throughout.

Imagined future failure Known basis and current uncertainty Team influence and priority judgment Early signal to watch
Catalog entries appeared usable but nobody could establish their freshness In a three-entry planning sample, one entry has no confirmed update owner. The team has not checked the wider catalog High influence over the pilot's required fields and included entries; select for action Any pilot entry lacks an owner or as-of date at the Day 3 readiness check
Help requests accumulated because support responsibility was unclear No named pilot support owner or absence backup is recorded yet. Request volume is unknown Team can agree a bounded support route with accepted owners; select for action Coverage or the escalation owner remains unconfirmed at Day 3
The metadata feed changed and the integration stopped matching its output The current planning note does not identify a versioned feed contract. No actual feed change has occurred Team can investigate compatibility and set an integration gate, but cannot control the upstream team's changes; select for action Contract owner or supported version remains unconfirmed at Day 4
A company-wide funding freeze stopped the pilot No freeze has been announced in the case; this is a possible external event without a current supporting signal Outside this team's control; retain as an escalation condition without inventing a local fix An actual funding or priority decision affecting the pilot is communicated

The first three risks are selected because they matter to the pilot and have an actionable next step. The fourth is retained rather than dismissed, but it does not consume the team's limited investigation capacity without a current signal. This is an editorial judgment within the fictional planning exercise, not a calibrated risk model.

If participants vote, record what the vote means: attention priority among the concerns considered. A popular card is not automatically the most probable failure, and a low-vote concern can still require specialist or governance review. Do not let voting erase a constraint the team lacks authority to waive.

Keep evidence, assumptions, and triggers separate

The missing owner in one sample entry supports checking ownership requirements. It does not establish that the whole catalog is stale. The lack of a documented feed contract supports resolving that dependency before integration. It does not prove that the upstream team will break the interface.

A trigger is an observable condition that prompts a response. “People lose trust” is too vague for this plan. “An included pilot entry has no confirmed owner by Day 3” gives someone a check they can perform and a decision they can route.

Likewise, a response should state whether it prevents exposure, investigates uncertainty, reduces the pilot's scope, or escalates a decision. These are different actions. Completing a contract inquiry is not the same as obtaining a supported interface, and writing a fallback plan is not the same as proving that it works.

Maren asks each risk owner to read the proposed wording back. Diego confirms that his concern is freshness evidence, not a claim that every entry is wrong. Owen asks to retain the upstream dependency as residual risk even if a local compatibility example is prepared. The board keeps both corrections.

A planning team assigning specific preparations around the unfinished model while keeping an unresolved dependency visibly separate

A filled response plan with funded ownership

The fictional response plan is approved for investigation and preparation only. Team lead Sabine confirms the allocations by deferring the named optional work below. The people involved accept their tasks; a name in the table is not used as a substitute for agreement.

Selected risk Owned action and capacity Trigger response and authority Due and review evidence
Freshness cannot be established Diego has two hours to define owner/as-of requirements and inspect the three planning entries; his optional icon cleanup is deferred If an entry lacks the required evidence, Maren keeps it out of the pilot candidate set until its owner resolves the gap Day 3: retain the three-entry check and unresolved fields
Support coverage is unclear Iris has 1.5 hours to document the pilot support route; Luca accepts 0.5 hour to confirm planned-absence cover. Their optional FAQ formatting and inventory-label cleanup are deferred respectively If coverage is unconfirmed at Day 3, Maren does not open the pilot to users. Urgent questions use the existing escalation route; no new always-on obligation is created Day 3: named owner/backup acceptance, coverage window, and escalation contact
Feed compatibility is unconfirmed Owen has 2.5 hours to request the supported contract/version and prepare a nonproduction compatibility example if the required information arrives; his optional benchmark cleanup is deferred If the contract remains unconfirmed at Day 4, Maren holds feed integration and brings the scope decision to Sabine. Owen cannot commit the upstream team to a date Day 4: response received, evidence still missing, and whether the example could be prepared

The allocated preparation totals 6.5 hours across four people. It is not added invisibly to their original plans, and it does not include unlimited investigation. If a task cannot fit its allocation, the owner records what remains and asks for a planning decision rather than quietly extending the work.

Maren schedules a Day 5 planning review. The team will examine completed preparation and unresolved gates, then decide what can proceed. This is not the pilot's release approval or a promise that the imagined failures have been prevented.

Review preparation without declaring the risks gone

Here is the completed fictional Day 5 review. The review records two eligible entries, bounded weekday support, and an unconfirmed feed contract.

Diego records owners and as-of dates for two of the three planning entries. The third planning entry still has no confirmed refresh owner, so only two of the original three can enter the candidate set. Maren excludes that entry from the candidate set. The remaining two entries satisfy this planning check; the team has not established that they will remain current throughout the pilot.

Iris and Luca confirm a weekday support window and a planned-absence handoff route, with the existing escalation contact named. Weekend coverage has not been offered or approved. Maren retains that boundary in the pilot proposal instead of describing support as available at all times.

Owen has sent the contract inquiry, but the upstream owner has not confirmed a supported version. The compatibility example is therefore not prepared. He records the dependency rather than inventing expected feed behavior to mark his action complete. His bounded inquiry is done; the integration condition remains unsatisfied.

Sabine's Day 5 decision is to keep feed integration and user-facing pilot opening on hold. She allows a separate internal desk walkthrough using the two eligible synthetic planning entries, with no live feed and no invitation to users. This limited preparation does not change the original pilot's acceptance criteria or convert a blocked pilot into a successful launch.

Owen will bring the unresolved contract question to the next dependency planning discussion, and Maren will revisit the opening decision when evidence arrives. No response date is promised on the upstream team's behalf. Diego's unowned entry and the weekday-only support boundary remain in the planning record.

Preserve the distinction between action and outcome

The team completed some preparations, exposed one unresolved dependency, and limited what could proceed. It did not measure failure probability before and after the exercise. No numerical reduction in risk, delivery improvement, or prevented incident can be inferred from the imagined scenarios and planning review.

Save the independent inputs, clarified board, funded actions, and Day 5 review together, including the held integration gate and unowned entry. Move continuing risks into the team's actual tracking process, retaining owner, next signal, and decision boundary. Keep the workshop bounded, then carry its open questions and preparations into the existing planning process.

Before your next project begins, define the imagined failure, collect independent explanations, and separate evidence from speculation. Select responses with accepted owners and real capacity, then review what was actually learned before authorizing the next step. More workplace guides are available through LandOffer.

Sources and Further Reading