You can usually tell in the first minute whether a review is going to be useful. If the opening slide is titled “Lessons Learned” and every lesson is something nobody has to change, the exercise is already finished. What follows is a ceremony. A real business post mortem analysis feels different in the room — slower, more specific, occasionally uncomfortable — because it is trying to establish something contested rather than confirm something agreed.
The contested thing is almost always the same. Was this a bad decision, or a bad outcome? Those are not the same event, and a review that conflates them will teach the organisation the wrong lesson with total confidence.
A Bad Outcome Is Not Evidence of a Bad Decision
Some decisions are well made and still lose. You had the information a reasonable person could have had, you paid a sensible price to get a bit more, you sized the downside, and the world went another way. Punish that and you train your best people to take only the bets that are easy to defend afterwards.
Other decisions are badly made and win. Nobody reviews those. That is the more expensive habit. An organisation that only convenes a review after a loss is running a biased sample, and it will slowly come to believe that its process is sound because the process has never been examined on a day when everyone felt good. I have seen decisions ratified as repeatable practice on the strength of a single fortunate result.
So the question that opens a review is not “did it work.” It is narrower and harder: given what was knowable at the time, and the cost of learning more before committing, would we make this call again? You can only answer that if the record of what was knowable still exists in a usable form. Usually it does not, and the reason is rarely dishonesty.
How the Record Gets Cleaned Up Without Anyone Lying
The first mechanism is narrative smoothing. Memory is a storyteller, and it prefers a straight line with one cause. So the written account describes a single path chosen for a single reason, when the actual moment involved three live options and a split room. Every account of a failure that reads as inevitable has had its uncertainty edited out.
The second is the vanished alternative. Reviews describe what was done in detail and what was rejected not at all. But a decision is a comparison, and without the option you passed on — and the reason it lost — you cannot evaluate the choice, only the result. I would rather read two paragraphs on the runner-up than ten on the execution.
Third is externalising to conditions that were already in the model. “The market turned” is a finding only if nobody named that risk beforehand. Go back to the original memo. If the risk appears there, was named, and was never sized or monitored, that is a process failure wearing the costume of bad weather. This is why I keep coming back to a small set of questions about concentration, reversibility and detection when a risk is first raised rather than after it lands. A risk that was identified but never made detectable is not an act of God.
Fourth is the uncaptured meeting. The real decision often happens on a call, in a corridor, or in the fifteen minutes before the formal session. The minutes then record a unanimous approval and none of the argument that preceded it. Anyone reading that file in two years will conclude the room agreed. It did not; it converged, and how it converged was the interesting part. This is one reason I ask to observe a board meeting before I accept a seat. Minutes tell you what was approved. Sitting in the room tells you whether disagreement is survivable there, which is the single best predictor of whether the written record will be worth anything later.
Fifth, and the quietest, is the survivor's edit. The person who objected has moved on, and the objection leaves with them. Whoever remains writes the history, not out of malice but because they are the only one still available to write it. The fix is to record dissent, with attribution, at the time it is voiced — not to reconstruct it afterwards from the memories of people who won.
Sixth is language drift. “We decided to pause the initiative” instead of “we cancelled it.” “Alignment was challenging” instead of “two executives wanted different things and neither would say so.” Passive, hedged prose is unreadable a year later, which is precisely when someone needs it.
Blameless Does Not Mean Subject-Free
The blameless retrospective is one of the better ideas to come out of engineering culture, and it is routinely misapplied. Blamelessness is a promise about outcomes: you will not be punished for a loss you could not reasonably have avoided. It is not a promise that no one's name appears in the document.
When a team hears “blameless” and produces a report with no human subjects — things happened, decisions were reached, an approach was taken — it has traded accountability for comfort. The result is a file that cannot be acted on, because nobody can be asked a follow-up question about it. Name the person who owned the decision. Then judge the process, not the person. Those two instructions only feel contradictory if the culture is weak, and if it is weak, that is the real finding.
Write the Memo Before You Know the Answer
Hindsight cannot be defeated by willpower. Once you know how something ended, you cannot reliably reconstruct what you believed before it ended. The only durable defence is a contemporaneous record, which means the quality of your post-mortems is set months earlier, at the moment of commitment.
A decision memo does not need to be long. It needs to say what we believe, what has to be true for this to work, what would tell us early that we are wrong, and what we expect to see by when. Then the review compares outcomes to the memo instead of to memory. Arguments about what people “always thought” end immediately, which saves the room for the argument worth having.
Two more things belong in that memo. The alternative that came second, and the estimated cost of delaying the decision to gather more information. Both are cheap to write down and impossible to recover later.
Who Runs It, and in What Order
Do not let the decision owner facilitate their own review. Not because they are dishonest — because they are the least able person in the building to hear the sentence that starts “the second option looks better in retrospect.” Someone with no stake in the verdict should run it.
Order matters more than most teams expect. Run the review in two sittings. In the first, reconstruct the timeline from artefacts only: documents, dates, messages, approvals, the numbers as they were reported at the time. No interpretation permitted, from anyone. In the second, interpret. Splitting them stops the first confident story in the room from organising all the evidence that follows, which is the default outcome when timeline and analysis happen in one session.
Then decide what actually warrants a review. Losses, obviously. But also near misses, unexpected wins, and the opportunities you declined. Reviewing a pass is unglamorous and disproportionately instructive, because the decision was made on the same information as any other and the outcome is observable from outside. Much of the useful work of governance happens in this register — the unglamorous reading and comparing that goes on between the meetings rather than in them.
Memory Has to Be Retrievable to Count
A post-mortem nobody can find is not institutional memory; it is a document. If yours live in a folder named after the project, they are effectively gone, because the next person to face the problem will not be thinking about that project. File them by the kind of bet instead. Build versus buy. Entering an adjacent market. Replacing a leader. Signing a customer whose requirements are unlike anyone else's — a decision I have argued elsewhere deserves more scrutiny than it usually gets, since choosing which customers not to serve is a strategic act and not a sales one.
Retrievability matters most where people rotate. In volunteer organisations it is the whole game; at Berkeley Florida, which I founded and chair, anything that lives only in the head of whoever is currently doing the work has a very short half-life. Companies have the same problem on a longer clock and notice it later.
What the Next Team Inherits
Learning from business failure is not really about catharsis or accountability, though both show up along the way. It is about tuition. Someone already paid for this lesson, and the only question is whether anyone else gets the benefit.
That is why I care more about whether a review is honest than whether it is flattering. The flattering version protects the people in the room for about a quarter. The honest version — dissent recorded, alternatives named, the decision judged on what was knowable and not on how it ended — protects whoever inherits the same choice three years from now. We will not remember the details. If we have done the work properly, we will not need to.