Run a Five-Minute Pre-Mortem Before the Action Items

The launch review ends eight minutes early. The team has chosen a date, the AI recap has named six owners, and a polished follow-up is ready before anyone closes the call.

It feels like an excellent meeting.

Three weeks later, the launch slips because the migration plan assumed that customers could export a file they do not actually have permission to access. The issue was visible in an old support thread, but nobody went looking. The meeting moved so smoothly from decision to execution that the plan never had to survive a serious challenge.

This is an increasingly ordinary risk. AI can reduce the friction between talking and doing. It can recover context, summarize the choice, propose tasks, and make the next step feel obvious. That is valuable. It also means a meeting can accelerate a weak plan with impressive efficiency.

Before the action items harden, give the plan five uncomfortable minutes.

Faster execution raises the value of doubt

Microsoft's 2026 Work Trend Index captures the tension well. In its survey of 20,000 people who use AI at work, 50% named quality control of AI output and 46% named critical thinking among the human skills becoming more important as AI takes on more work. The report also says 86% treat AI output as a starting point rather than a final answer. Microsoft Work Trend Index

Those are self-reported attitudes, not proof that review happens consistently. In a busy meeting, "we will check it" often means a quick scan of the recap. That catches a wrong name or date. It rarely exposes the shared assumption beneath the whole plan.

A pre-mortem creates a more useful kind of review. Instead of asking, "Does this plan look right?" the group assumes a specific future failure and works backward.

The underlying idea is called prospective hindsight. In a foundational study, researchers asked people to explain future events from different temporal perspectives. Treating an outcome as if it had already occurred changed the explanations people produced, including their length and use of concrete, episodic reasons. Journal of Behavioral Decision Making

That small shift matters. "What could go wrong?" invites a familiar list of generic risks. "It is November 15, the launch failed, and the customer has escalated. What happened?" gives the mind a scene to explain.

Run the five-minute version

This is not a second planning meeting. It is a short interruption between agreement and task creation. Use it when the choice affects a customer promise, meaningful spend, a public date, hiring, security, or work that will be expensive to reverse.

Minute 1: Freeze the plan

Read one sentence that states what the group believes it is about to do.

We will move 40 pilot customers to the new workflow on November 1, with support covering the first weekend.

Do not begin the exercise until people agree that this is the plan under examination. A vague plan produces vague failure stories. If the recap contains several possible versions, that is already useful information: the group may not be ready for action items.

Minutes 2 and 3: Write silently

Ask everyone to complete this sentence on their own:

It is two weeks after launch. The plan failed in a way we should have anticipated. The main reason was...

Silence matters. If the most senior person speaks first, the room will tend to explore variations of that person's concern. Independent writing gives operational knowledge, customer knowledge, and quiet skepticism a chance to appear before the group starts negotiating.

Require one plausible cause, not a catalog of disasters. "Everything took longer" is too soft. "Support coverage was scheduled, but nobody had access to the migration logs" can be checked and addressed.

Minute 4: Find the earliest signal

Each person reads their cause. The facilitator groups duplicates, then asks one question for the strongest two or three:

What would we notice first if this were becoming true?

The answer should be an observable tripwire. It might be an approval still missing on October 25, two pilot customers unable to complete a test export, an error rate above an agreed threshold, or an owner who has not accepted the weekend rota.

This step keeps the pre-mortem from becoming theater. A risk with no visible signal is difficult to manage. A signal gives the team something it can watch before the imagined failure arrives.

Minute 5: Change the plan

Choose one of three treatments for each priority risk:

• Prevent: Add a step that reduces the chance of failure.
• Detect: Name a tripwire, owner, and check time.
• Contain: Decide how to limit the damage if the risk occurs.

Then edit the action list while everyone is still present.

The original task might be, "Mina prepares customer migration instructions by Friday." After the pre-mortem, it becomes, "Mina tests the instructions with two pilot customers by Friday. If either customer cannot access the export, Leo pauses the November 1 migration and opens an access review."

The task is longer, but the plan is more honest.

Do not turn caution into fog

A pre-mortem can fail in two directions.

The first is performative pessimism. Participants invent dramatic possibilities that nobody can observe or influence. The cure is to demand a mechanism and an early signal. "Customers will hate it" is not ready. "Three of five pilot customers abandon setup at the permissions step" is.

The second is treating every concern as a veto. The exercise is not a search for certainty. Most worthwhile plans retain risk. Its purpose is to reveal which risks the group knowingly accepts, which ones deserve a guardrail, and what evidence should cause the plan to pause.

Keep the ritual proportional. A reversible internal experiment may need one minute and one tripwire. A contractual commitment or difficult migration may deserve a separate risk session with people who were not part of the original plan. Five minutes is a floor for consequential work, not a magic dose.

Also separate plan failure from AI failure. The migration can go wrong even if the recap is flawless. Conversely, the plan can be sound while the generated action list drops a condition or assigns the wrong owner. Review both layers: first challenge the plan, then verify that the record represents it.

Give AI the evidence job

AI can make the pre-mortem sharper without running it on the team's behalf. Ask it to retrieve relevant past incidents, surface contradictions in connected material, show unresolved dependencies, or suggest a question the current discussion has not answered. Those are evidence tasks.

People still decide which failure is plausible, which tradeoff is acceptable, and who has authority to pause the work. They also notice what the available record cannot contain: a strained partner relationship, a fragile handoff, or a commitment that was technically made but not genuinely understood.

A meeting copilot is most useful here when it supports the pause as well as the acceleration. Caspi can provide live recap, suggested questions, contextual chat, proactive flags from connected tools, post-call action items, and persistent meeting memory. That combination can help a team freeze the actual plan, bring relevant context into the risk discussion, and preserve the tripwire beside the commitment it protects.

The goal is not to make every meeting anxious. It is to make confidence earn its next step. When execution is only a click away, five minutes of disciplined doubt can be some of the most productive time in the room.