Take-Home Practice: Seven Templates and the Red Lines

The three prompts in lesson two are the main line, but in real work you will not always run the whole thing — sometimes you only need to check whether your materials are enough, sometimes you just want the root causes.

This lesson breaks the method along the retrospective timeline into seven templates: prep → G restore goals → R compare results → A analyse root causes → I build action, plus a one-line starter for when materials are complete. Then the blank canvas, tool selection and the red lines.

Use: the pre-flight check. Two minutes here saves you from discovering mid-retrospective that the data sheet is missing.

Use: standing up the exam paper. "The measurement basis must quote the original" is a hard constraint — once AI quietly polishes the basis, the achievement rates can no longer be reconciled with anyone.

Use: producing the comparison table. The row and word limits are not fussiness — a table over 8 rows means you have not separated core metrics from process data.

Use: digging for causes. That last sentence is the soul of this template — four delays may be four links in one chain, and viewed separately they become four separate "tighten up management" notes.

Use: turning analysis into action. Note the phrase about estimating whether next round's goals become achievable — it forces you and AI to check whether the actions actually work, instead of writing promises nobody can keep.

Use: producing what goes upward. Do not delete "conclusion first" — management wants to know whether it worked, why, and what happens next, not to read the process in chronological order.

Use: the fast lane when materials are complete and time is short. Do not use it for your first retrospective — run templates 1–6 once first, so you know what each step's output should look like.

Add one line under the table for experience assets: practices that proved effective this round and should be scaled next round (≤3 items).

This line is the one most often skipped, and it is where the compounding lives — root causes stop you repeating mistakes; experience assets make you faster each round.

The third row is why this is worth doing once: consistent formats are what let two retrospectives be compared. Otherwise each one is an isolated document and you cannot see a trend.

1. AI makes things up with a straight face: numbers, dates and names must be checked by hand — every figure in a retrospective has to trace back to the source data sheet;
2. Never feed it confidential or sensitive information: anonymise first, or do not feed it at all — client lists, internal pricing and unpublished data all get anonymised;
3. The output is not the final version: your name is on it, and so is the responsibility — review it once yourself before it goes out;
4. Retrospectives target processes, not people: root causes must point at processes and mechanisms, and AI output must never become a case file against a colleague; before it reaches your manager, confirm the evidence chain behind every root cause.

The fourth deserves a second look: the root causes AI writes are reasoned conclusions with evidence, not verdicts. Once a retrospective report is used as evidence against someone, nobody will ever put the real situation into the minutes again — that is how retrospectives actually die.

Check two things before sending: can every root cause be traced back to a specific source? Does it point at a rule you can change?