She lost a £220,000 client in March. By the end of April she had told the story three times, twice over a glass of wine and once on a coaching call, and each version was slightly different. By the end of May the version in her head had settled, and the version in her head was already wrong in two specific places. She had not written any of it down. When the next pitch with similar shape arrived in October, she went into it with the polished story rather than the actual one, and the lesson she had not quite learned in March was waiting for her again.
This is the case for a personal post-mortem you can run in 45 minutes on a Saturday morning, with AI sitting in the second chair as the peer who asks the questions you would not ask yourself.
What is a personal post-mortem with AI as your retrospective partner?
A personal post-mortem with AI as retrospective partner is a private 45-minute written review of a single recent setback, structured in three fifteen-minute phases, with AI playing the role of a trusted peer who asks the questions you would rather not ask yourself. The output is a dated two-page document with three lessons and two open questions, written in your own words and kept where you will find it again.
The model is engineering’s blameless post-mortem, codified by Google’s Site Reliability Engineering teams. The discipline is to investigate why the available information looked the way it did at the time, not to allocate fault for the outcome. Annie Duke’s work on decision-making under uncertainty makes the same separation between decision quality and outcome quality, which is the second move the practice asks of you. James Pennebaker’s research on expressive writing is the third anchor: writing about a difficult experience produces measurable cognitive and physiological benefits that thinking about it does not. The military After Action Review is the fourth, evidence that time-bounded structured review is a transferable leader practice rather than an engineering peculiarity.
Why does it matter for your business?
It matters because owner-operators carry the most consequential setbacks in a small firm and process them on the move. A lost deal, a hire that did not work, a launch that landed flat, these get worked through the next three days of meetings, and the lesson, if there is one, settles into a feel rather than the specifics. By the time the next similar situation arrives, the feel is what shows up.
Memory decay is faster than many leaders assume, and hindsight bias rewrites the version you carry forward inside days. Without a written record pinned at the time, the post-mortem you do in your head is the story you can live with, rather than the one that would change your behaviour. The financial maths is straightforward. Setbacks are inevitable in services businesses, and the question is whether each one teaches you something or simply costs you. A founder who runs four written post-mortems a year and acts on what they find compounds learning that a founder who processes the same setbacks only in conversation does not. The artefact is the asset, and writing it down is what lets you stop carrying it.
Where will you actually meet it?
You will meet it on the Saturday morning after a setback you cannot stop thinking about: the deal that closed against you on Friday, the resignation letter on Tuesday, the launch that opened to a flat inbox on Monday. The 24-to-72-hour window is what produces the most useful document, because you have enough distance to think but not so much that the details have been smoothed.
The format is deliberately light. A single dated document, two pages, written in plain prose rather than a corporate template. The first page is the timeline, what happened in what order, with the specific words and numbers you remember. The second page is the lessons and the open questions. The AI sits next to it as a chat window, fed the timeline as context, asked to take the role of a trusted peer asking three challenging questions. You answer the questions in writing inside the document. That is the practice. No retrospective software, no team meeting, no template land, no formal action register. The lighter the practice, the more often it actually happens, and frequency is what produces the compound effect. This sits inside the personal AI category covered in AI for your own work, not just your business and inside the Do quadrant of the EAD-Do framework recast for AI, where AI is a thinking partner rather than a tool.
When to ask vs when to ignore
Run it when the setback is consequential enough that you would want a younger version of you to know what you learned. The lost deal worth chasing again next year. The hire that revealed something about how you assess people. The launch that exposed an assumption you did not realise you were making. These earn the 45 minutes back several times over.
Ignore it when the setback is genuinely small, when nothing in the situation will repeat, or when you are still inside the emotional weather of it and any honest writing would slide into self-flagellation rather than reflection. The practice avoids therapy, confessional, accountability ritual, and corporate retrospective in personal clothing. If the only lesson available is “I should have known better”, the post-mortem has not been done properly yet, because that conclusion is hindsight bias rather than learning. Wait a day, come back to it, and ask the more useful question: what information would have changed the decision you actually made. The companion practice on the other side of the timeline is the pre-mortem run with AI before a major decision, which uses the same structure forwards rather than backwards.
Related concepts
Five anchors sit underneath the practice. Google SRE’s blameless post-mortem culture is the methodological foil and the source of the “good intentions, available information” framing. Annie Duke’s decision-quality versus outcome-quality separation in Thinking in Bets is the move that keeps personal post-mortems from collapsing into self-blame. Amy Edmondson’s basic, complex, and intelligent failure classification gives you a useful question to ask in the second phase.
Was this a basic failure that a checklist would have caught, a complex failure that needed a different process, or an intelligent failure where you tested a real hypothesis and learned it did not hold. James Pennebaker’s expressive writing research is the reason the artefact has to be written rather than thought, and Roediger and Karpicke’s retrieval-practice work is why the six-month re-read does the heavy lifting. The military After Action Review is the institutional precedent that the format scales. None of this is new. What is new is that AI now plays the second-chair role credibly enough that the practice no longer needs a peer or coach in the room to work, which is what brings it inside reach for an owner-operator on a Saturday morning.
If this is the kind of practice you want to make a habit of, Book a conversation.
