Which tool to choose
Any generalist assistant will do; what matters is how you set it up. Claude and ChatGPT handle the role of a coach who asks questions instead of patting you on the back well. The critical point isn't the tool but a single instruction: explicitly tell it not to praise you out of politeness. The models tend to please whoever's writing, and a retrospective that's all compliments is useless.
How to do it
The path is identical on computer and on phone: you recount, the AI structures, you decide what to change.
- Choose a concrete, closed case. A finished project, a lost deal, a bad week. On the vague you learn nothing; on a real episode you do.
- Ask for the demanding retrospective. The working syntax:
Act as a demanding but constructive coach. I'll tell you how a recent project of mine went. Help me do an honest retrospective: 1) three things I did well; 2) three real mistakes, and for each one the root cause, not just the symptom; 3) for each mistake a concrete, verifiable action to apply next time. Don't sugarcoat and don't praise me out of politeness: if a mistake is serious, tell me clearly. Here's what happened: "..."
- Dig into the cause. When the AI points out a mistake, ask "why did it happen?" two or three times in a row. You often arrive at a cause different from the one you imagined (not "I got the figures wrong" but "I had no fixed moment to check them").
- Turn the actions into rules. Close by asking: "summarize this in three rules I can pin above my desk." Those stick, the analysis is forgotten.
A concrete example
Marco runs small events and the last one ran at a loss. Instead of brooding, he tells ChatGPT everything with the retrospective prompt: budget, suppliers, timing, what fell through. The AI recognizes two things done well (the venue, the communication) and three mistakes. The biggest wasn't the late supplier, but the fact that Marco had no safety margin in the quote.
Marco asks "why?" and reaches the root: he always accepts the client's first price for fear of losing them. The concrete action becomes a rule: no quote without a 15% margin, and if the client haggles, you remove a line item, not the margin. Next time he won't repeat the same mistake, because now he has a rule and not just a regret.
When it does NOT work (and how to fix it)
If AI consoles you instead of criticizing you
It's the most common flaw: a gentle analysis comes out that dances around the problems. Force it: "play devil's advocate, write the retrospective the way someone who wants to prove where I went wrong would do it." The harsh version contains the things the gentle one hides.
If the analysis stays on paper
A retrospective without a change is wasted time. For each mistake demand a measurable action, with a when and a how ("every Friday I review the figures," not "I'll be more careful"). Without action, next time you reopen the same identical retrospective.
If the facts involve sensitive data
If the episode involves confidential numbers, client names or personal data, anonymize them before pasting (roles and round figures in place of names and exact amounts). The analysis works just the same, and you don't entrust sensitive information to a chat.
A tip from someone who really uses it
Keep a single "logbook" chat and come back to it each time. After a few retrospectives ask the AI: "looking at all the cases I've told you about, what's the mistake I repeat?" That's where the pattern invisible from a single episode emerges: the same stumble in different forms, which on your own you wouldn't have connected.
Frequently asked questions
Do I have to recount everything in detail?
The more concrete you are, the more useful the analysis, but the essential facts are enough: what you wanted to achieve, what happened, where you got stuck. You don't need a novel; you need the points where things took a wrong turn.
Can AI help me with the things that went well too?
Yes, and it's underrated. Understanding why a success worked makes it repeatable. Ask "what did I do right and how do I turn it into a fixed method": replicating what works is worth as much as correcting what doesn't.
Can an AI really understand my mistakes if it doesn't know me?
Here's the misunderstanding to clear up: AI doesn't know you and understands nothing about you. It works only on what you tell it, and that's exactly why it's useful — it forces you to put the episode into words, and half the clarity already arises in the writing. The tool asks the right questions; the answers, and the courage to look at them, stay yours.