Nguyen Le Phong

Pre-Mortems: Ask Why the Plan Failed Before You Start

A pre-mortem asks your team to imagine the plan already failed and explain why.

Thursday afternoon, and the migration plan looks great on the slide. Dual writes for a week, a backfill over the weekend, cutover on Monday. Everyone nods. Two people in the room each have a small worry: one isn't sure the backfill can finish in a weekend, the other remembers a reporting job that reads the old table. Neither says a word. The meeting is running late, the mood is good, and nobody wants to be the one who rains on it.

A pre-mortem is a short exercise built for that exact room. Before the work starts, you tell everyone: it's three months from now and this plan failed badly. Write down why. Those two worries stop being an awkward interruption. They're the assignment.

This post covers where the idea comes from and why it works, how to use it on code and systems, how the same question helps with promotions, interviews and personal goals, and two ways to run one: the original version and a 15-minute version.

Nguyen Le Phong goes over a work plan at his desk in the evening, a notebook open next to the screen, before the next decision is locked in.
The best time for a pre-mortem is right before you commit: before the deploy, the interview or the review cycle, while changing the plan is still cheap.

Where the idea comes from, and why it works

Psychologist Gary Klein described the method in a 2007 Harvard Business Review piece, Performing a Project Premortem, and he still calls it the Pre-mortem Method of Risk Assessment. His starting point is blunt: projects fail a lot, and one reason is that people with doubts during planning keep them to themselves. A pre-mortem makes it safe for them to speak, because finding a reason for failure is now the job.

The second reason is about how our heads work. Klein cites a 1989 study by Deborah Mitchell, Jay Russo and Nancy Pennington on "prospective hindsight", which means imagining that an event has already happened. In his summary of their work, that framing increased people's ability to correctly identify reasons for future outcomes by 30%. "What might go wrong?" tends to get a polite, generic answer. "It went wrong, what happened?" gets a story, and stories come with specific causes: the dependency nobody owned, the rollback nobody tested, the stakeholder nobody asked.

Klein's article has a nice example. A research project held a 90-minute kickoff meeting where nobody mentioned a time constraint. In the pre-mortem, a senior executive said the project had "failed" because there wasn't enough time to prepare a business case before an upcoming corporate review. The project manager reworked the plan around that review cycle on the spot. An hour and a half of planning missed something that a few minutes of imagined failure caught.

He lists a few side effects worth having too. The exercise cools down the full-speed-ahead attitude of people who are heavily invested in the plan. The person who spots a weakness nobody else mentioned feels their experience counted. And the team gets better at noticing early signs of trouble once the work is underway.

The pre-mortem loop: state the plan, assume it failed, name the causes, change the plan, then watch for early warning signals. PRE-MORTEM: ASK WHY IT FAILED WHILE YOU CAN STILL CHANGE IT Plan what we intend Assume failure it already went wrong Name causes the doubts people kept Mitigate reduce or prepare Watch signals You won't predict every failure. You can catch the ones somebody in the room already half-sees.
The loop in five steps. The dashed line matters: the signals you pick feed the next round of planning once the work is running.

Is this just pessimism with a nicer name? Pessimism says "this will fail" and stops there. A pre-mortem keeps going: which failures can we already see, and what do we change today? You run one because you want the plan to work.

In software: questions to ask before you ship

The same question scales from one pull request to a whole platform.

At PR size, ask: "This change caused a production bug next week. What did I assume?" Maybe the input was never empty in your test data. Maybe the third-party API returns a field with a different type than the docs say, once in a while. Maybe your retry loop retries too hard and turns a two-minute outage into a traffic spike against a service that's already struggling.

At release size, follow one request through the system. The payment provider is slow for five minutes at peak. Do we time out, queue the work and show the user something useful? Or do we hold threads, drain the connection pool and take the database down with us? A few try/catch blocks won't answer that. You have to look at the whole path.

At architecture size, try: "Traffic went 10x overnight and the platform fell over. Where did it break?" The usual suspects are the database connection pool, a synchronous call hiding in a hot path, a cache stampede when a popular key expires, or a background job that was fine at small volume and now floods the queue. There's an opposite failure worth asking about too: the system didn't die from load, it died from complexity. So many services, abstractions and config layers that tracing a real incident takes longer than the incident.

What you get out of this is a short list of guardrails to add before the stress arrives: timeouts, circuit breakers, idempotency keys, alerts, a rollback plan, a load test, a feature flag, a runbook. Sometimes the best guardrail is a less clever design.

A drone prototype with exposed wiring sits next to a laptop on a lab bench.
While the prototype is still open on the bench, ask which assumption breaks first and what safeguard it needs before real load shows up.
ContextPre-mortem questionWhat it often turns up
Code change"This PR caused a production bug. Which assumption was wrong?"Missing edge cases, thin validation, weak tests, hidden coupling.
System design"Traffic grew 10x and the platform fell over. Where was the bottleneck?"Database pressure, queue backlog, cache stampede, synchronous dependencies.
Migration"We rolled the migration back at midnight. What did we miss?"Data shape mismatch, half-tested rollback, slow backfill, no way to see progress.
Promotion"The review cycle ended and I wasn't promoted. What was the real gap?"Impact nobody could see, thin stakeholder trust, no successor, scope too small.
Interview"I walked out knowing it didn't land. Which part gave me away?"Design answers without tradeoffs, vague behavioral stories, unclear ownership.
Personal goal"I quit after one month. What made the plan too hard to keep?"Goal too big, too much friction, no restart plan, running on motivation alone.

For your career: promotions and interviews

Klein built the method for project teams, but the question works fine for one person. Careers rarely fail in the final review meeting. They lose ground months earlier, when expectations are fuzzy, evidence is thin or good work stays invisible.

Try this one: "Six months from now, the review cycle is over and I didn't get promoted. Why?" The honest answers are often not technical. You shipped plenty of tasks but never tied them to business impact. You became the person who fixes urgent problems, not the one who changes how the team works. You're still the bottleneck for daily operations, so nobody gets to see you working at the next scope. Or you did strong work and the people who decide never heard about it.

Writing that down stings a little, but it beats finding out in the review. Each answer points to a move you can make now: ask your manager what the next level looks like in concrete terms, pick one project with impact people can see, keep a running note of decisions and outcomes, hand one of your daily responsibilities to a teammate, and ask for feedback before the official cycle starts.

Interviews work the same way. Before a senior or lead loop, picture walking out knowing it didn't land. Why? Your system design stayed in implementation details and never got to tradeoffs. Your incident story sounded heroic but didn't show what you learned. You said "we" so often that the interviewer couldn't tell what you owned. You drilled algorithms and skipped the judgment questions the role is really about. Now your prep has a target: rehearse the weakest story, redraw the design with real constraints, and practice explaining one tradeoff out loud.

Nguyen Le Phong and two teammates lean over a laptop in an outdoor courtyard, going through a plan before they commit to it.
A career pre-mortem is mostly an early, honest conversation that names the gaps: thin evidence, fuzzy ownership, a story other people can't repeat.

For personal goals: plan for the week you'll want to quit

Most personal plans don't fail on day one. They fail because they were designed for a version of you with endless energy, no interruptions and permanent motivation. Learning a language, running, writing, saving money, a side project: all easy to picture, all hard to keep going.

Before you start, ask: "A month from now, I stopped. What happened?" The answers are usually boring. Sessions were too long. The gym was too far. The side project had no small milestone. One busy week broke the streak and there was no way back in.

Once you can see the reason, the plan gets smaller and sturdier. Fifteen minutes a day beats a heroic two-hour plan that collapses on Tuesday. A restart rule ("if I miss two days, the next session is only five minutes") beats guilt. A public promise helps some people and a private routine helps others. The aim is a plan that survives your actual life.

Nguyen Le Phong stands on a mountain summit looking out over the sea, taking a break before the next stretch of the climb.
A personal pre-mortem is a short pause to admit what usually breaks your rhythm, then plan the next stretch around it.

How to run one

Klein's original format is short. The team gets briefed on the plan. The leader announces that the project has failed spectacularly. For the next few minutes, everyone writes down every reason they can think of, alone, including the things they'd normally keep to themselves to stay polite. Then the leader goes around the room, starting with the project manager, and each person reads one new reason until every list is empty. Afterwards, the project manager goes through the list for ways to strengthen the plan.

Atlassian's Team Playbook stretches this into a 60-minute session for 3 to 11 people. It adds a few useful pieces: ten minutes of silent brainstorming, three votes per person for the biggest threats, a column for what could go right, ten minutes each on the top three topics with at least one action per risk, and an owner and due date on every action. It also suggests running it asynchronously for teams spread across time zones, or as a condensed version inside sprint planning.

For a single release or a personal goal, fifteen minutes is often enough:

  1. Name the plan precisely: ship this release, pass this interview loop, finish this side project milestone.
  2. Write one sentence: "It's three months from now and this failed."
  3. Give everyone five minutes to write reasons alone. Silence stops the first confident voice from setting everyone else's agenda.
  4. Go around the room, one reason each, until the lists run out. Group the duplicates.
  5. Pick the two or three causes that are both likely and expensive.
  6. For each one, agree on a change you can make now, an owner, and an early warning signal that says it's starting to happen.

That last step is what separates preparation from worry. A risk list with no actions just becomes background stress. A short list with owners and signals is something you can steer by.

One question to keep handy

"If this fails later, what will I wish I'd noticed today?" It works before a code review, an architecture decision, a stakeholder meeting, an interview or a new habit.

What a pre-mortem won't do

It won't make a plan bulletproof. No exercise sees every failure, and some surprises are real surprises. But plenty of failures aren't mysterious at all. They showed up as weak signals long before they got expensive, and someone in the room usually half-saw them.

It also goes wrong in a few predictable ways. The most senior person talks first and everyone else edits their list to match. The session turns into a contest over who can sound most doomed. Or the team produces a great list and nobody owns any of it. Writing alone first, going round one reason at a time and ending with owners fixes most of that. And when something still breaks after launch, the post-mortem is the other half of the loop. That one works best when it's blameless from the first minute.

Try this week

  • Pick one thing you're about to start: a release, a migration, an interview, a new habit.
  • Write "It's three months later and this failed" at the top of a page, then five minutes of reasons.
  • If it's a team plan, have everyone write alone first, then go round one reason at a time.
  • Keep the top two or three causes. Give each one a change you make now, an owner and a warning signal.
  • Put the warning signals somewhere you'll see them again, like the release checklist or your calendar.

Start small. Fifteen honest minutes on next week's deploy will show you whether the habit is worth bringing to the bigger plans.

What did you think?

Frequently asked questions

What is a pre-mortem?
It's a planning exercise Gary Klein described in Harvard Business Review in 2007. After the team hears the plan, the leader says the project has failed. Everyone writes down reasons on their own, the team shares them one at a time, and the plan owner uses the list to strengthen the plan before the work starts.
How is a pre-mortem different from a post-mortem?
A post-mortem happens after something went wrong and looks at what actually happened. A pre-mortem happens before the work starts and assumes the failure has already happened, so the team can change the plan while changes are still cheap. Klein calls it the hypothetical opposite of a post-mortem.
Why does imagining failure help a team find risks?
Two reasons. It makes raising doubts the assignment, so people who would normally stay polite speak up. And it uses prospective hindsight: Klein cites a 1989 study by Mitchell, Russo and Pennington showing that imagining an event has already happened increased people's ability to correctly identify reasons for future outcomes by 30%.
How long does a pre-mortem take?
In Klein's version, the writing step takes only a few minutes. Atlassian's Team Playbook runs a 60-minute session for 3 to 11 people with silent brainstorming, voting and owners for each action. For a single release or a personal goal, 15 minutes is often enough.
Can I use a pre-mortem for a promotion or a job interview?
Yes. Ask: "The review cycle ended and I wasn't promoted, why?" or "I walked out of the interview knowing it didn't land, which part?" The answers usually point to specific gaps, such as impact nobody saw, no successor for your current work, or design answers without tradeoffs, and each gap suggests something you can fix now.