TutorialsStart Here: What a Retrospective Really Does
Four core concepts (retrospectives, the GRAI method, workflows versus one-off conversations, materials and information gaps), seven scenarios, and the course formula.
Many people think a retrospective means retelling the project from the beginning. A retrospective only does one thing: take the exam paper you set at kickoff and mark it again.
This lesson sets the target: you are not producing a process record, you are producing a table that shows which rule needs to change.
Four concepts, once and for all
A retrospective targets processes, not people
It is not about finding someone to blame. It asks questions of the process and the mechanism: where did the result diverge from the goal, and which rule caused the gap. The object is always the work, never the person.
The GRAI four-step method
The standard playbook: G restore the Goal, R state the Result, A Analyse the root cause, I build Insight into action. Everything today runs on these four steps — the three hands-on prompts are G+R, A, and I in relay.
Workflow vs one-off conversation
A one-off conversation is like ordering takeout; a workflow is the fixed recipe in your own kitchen — fixed ingredients, fixed steps, fixed output. Do it every time and it gets faster.
Materials and information gaps
The five raw material types you feed AI: goal records, meeting minutes, group chats, task lists, data sheets. What is missing must be flagged as an information gap — never invented.
Taking each one further
Seven high-frequency scenarios
"Retrospect a project" can mean very different work. Match yours in this table:
| Scenario | What this work is | How AI helps |
|---|---|---|
| Rebuild the timeline | Process restoration | Assemble records from minutes, chats and task lists into one project timeline, flagging what is missing |
| Restore the goals | Goal alignment | Distil the kickoff goals into an exam paper, so the retrospective is not a contest of impressions |
| Compare the numbers | Achievement comparison | Align goal versus actual item by item, compute achievement rates, spot the weak layer |
| Dig for causes | Root-cause analysis | Three layers — phenomenon → direct cause → root cause — rather than "be more careful next time" |
| Save the experience | Knowledge asset building | Working practices become SOPs, pitfalls become checklists, reused next round |
| Produce the report | Writing it up | Directly output a presentable retrospective report and slide outline |
| Track the actions | Closing the loop | Action items carry owner, deadline and acceptance criteria, then become trackable to-dos |
Lesson three gives seven templates matching this timeline. Nothing to memorise now — just hold on to one idea: a retrospective is not one big prompt, it is a production line from materials to action, where each step's output feeds the next.
The formula: role + task + context + format + constraints
The course teaches exactly one formula and uses it throughout. Get all five parts right and the first draft is usable:
| Role | Task | Context | Format | Constraints |
|---|---|---|---|---|
| Project retrospective expert | Restore goals, compare results, find root causes | All project materials fed in | Tables / three-layer analysis / report | Flag missing data as information gaps, never invent |
The single most valuable trick
Add one line at the end of any prompt:
开始前，先问我三个问题，确认信息后再开始。This follow-up is the most easily skipped and most valuable step in the whole lesson.
AI will ask: which material sets the goal measurement basis? When the group chat and the minutes disagree, which one wins? Which figures have not been backfilled yet? — questions you will usually need a moment to answer.
Where you cannot answer is exactly where the retrospective is most likely to be written wrong. Letting it ask first is far cheaper than fixing it three times afterwards.
The principle behind it all
In one sentence: AI writes the first draft and finds what you missed; you judge and decide. Treat it as a very smart subordinate — you assign the work, and you make the call.
Key takeaways from this lesson
| Remember this | Because |
|---|---|
| Retrospectives target processes, not people | Conclusions aimed at a person cannot be improved; those aimed at a rule can |
| GRAI: goal → result → root cause → action | Skip one and a retrospective degrades into retelling or emotional debate |
| Push root causes to process / mechanism / rule level | Stop at the direct cause and you will hit the same pit next round |
| Five material types: goals / minutes / chats / tasks / data | Miss one and that step runs on impressions |
| Five formula parts: role, task, context, format, constraints | Miss one and AI gains one more degree of freedom |
| End with "ask me three questions first" | It surfaces exactly where you are most likely to get it wrong |
| The compounding comes from what is fixed | Fixed checklist, fixed prompts, fixed output — speed arrives on the second run |
Project Retrospectives with AI
One GRAI four-step method, 3 hands-on prompts and 7 templates — turn a pile of scattered project records into a report and a slide outline you can present, in 30 minutes.
Hands-On: Materials In → Analysis Out → Action Forward
Using the 'AI Open Class · Back-to-School Special' case, three prompts in relay turn five scattered materials into a timeline, a goal-versus-actual table, three-layer root causes and a presentable report.