Where we are
You have a workspace (b1), a charter (b2), a milestone architecture (b3), a dependency map (b4) and a live risk register (b5). Northwind is now nine weeks in and M3 - raw ingest signed off - has slipped from 10 August to 24 August. This is the week the reporting either earns trust or quietly loses it. Status reporting is also the single most common first use of AI in delivery teams, which makes it the most common place to go fluently, confidently wrong.
Why the AI-drafted status report fails 6 min live
Ask any model to "write a weekly status report" from a pile of programme noise and you get something readable, well-structured, and useless. Not because it is stupid - because you asked it to compress ambiguity, and compression without a rule produces the average of everything. "Amber. Good progress, some blockers." Nobody can act on that, and worse, nobody can argue with it.
LiveThe four fluent failures3 min▶
Every weak AI status report fails in one of four recognisable ways. Learn to spot them and you can review a draft in thirty seconds.
- The mood ring. A RAG colour with no rule behind it. Two readers disagree about whether the programme is in trouble, so nobody acts.
- The flat photograph. A picture of this week with no comparison to last week. A 14-day slip and a stable plan look identical.
- The orphan actions. "Follow up on the data-quality issue." Owned by nobody, dated never, and it will appear again next week, reworded.
- The unfalsifiable claim. "Quality is looking reasonable." No ticket, no number, nothing to challenge - and nothing to trust.
Amber for eleven weeks. A programme reported amber for eleven consecutive weeks with no agreed definition of amber. Every week the steerco nodded. In week twelve it went red and the sponsor asked why nobody had flagged it - and the honest answer was that everybody had, in a colour nobody had defined. The report was not lying. It just could not carry information.
Self-studyAgreeing the RAG rule before you need it2 min read▶
Write the rule at kickoff, when nothing is on fire and nobody is defending anything. A workable default:
| Colour | Means |
|---|---|
| Green | Every live gate forecasts on or before its date, no critical risk open above the agreed threshold |
| Amber | A gate forecasts 2 or more weeks late, and there is a credible recovery plan attached |
| Red | A gate is unrecoverable without a decision from the steerco: scope, money, or the date moves |
Then restate the rule inside every report. It costs one line and removes the entire "what does amber mean here" conversation - which is the conversation that eats the first ten minutes of a steerco that should be spent deciding something.
Watch the report climb: 24 to 100 ★ 12 min · everyone plays
Below is Northwind week 9, fed to a model with everything switched off. Read the report first - it is fluent and it is worthless. Then turn on one lever at a time and watch two things: the score, and the specific failure that disappears. Turn on the scorecard to see which of five real programme weeks come out steerco-ready with your current setup.
LiveWhat each lever actually is, on your desk4 min▶
- The RAG rule - one line in your house rules and one line restated in every report. Agreed once at kickoff.
- Owner and date per action - comes from your owner list (b1) and your RACI. If the model cannot find an owner, it must write "TBC - ask [role]", never "the team".
- The decision log - a running list with ids: D-14, 31 Jul, sponsor, what was decided, what it cost. Session b7 builds it from meeting transcripts.
- Evidence - ticket ids, dates, and the actual metric. This is the lever that makes the report challengeable, which is what makes it trustworthy.
- The ask - the decision you need, the options, the cost of each, and the date after which it stops being available.
- The delta - forecast dates, open criticals, and one throughput number, this week against last week.
Run your own last week through the rubric ★ 8 min · everyone builds
Open the last status report you actually sent - not a good one, the real most recent one. Score it against the same six checks. Most people land between 40 and 60 the first time, and the missing points are always the same two.
Check 1 - is the status colour defined by a written rule that appears in the report? If the rule lives only in your head, score zero.
Check 2 - can a reader see what changed since last week, with at least one number?
Check 3 - does every action carry exactly one named owner and one date?
Check 4 - are the week's decisions recorded with who took them and when?
Check 5 - can every claim be traced to a ticket, a metric, or a document?
Check 6 - is there a clear ask, with a deadline and the cost of not deciding?
Total your score out of six. Whatever you scored, fix the single cheapest gap this week rather than all of them. Usually it is the RAG rule, which takes one meeting.
Four out of six, and the two missing were the expensive ones. A senior PM scored her own reports and found she was strong on actions, decisions, evidence and the delta - and had never once written the ask or defined amber. Nine months of reports, and not one had told the sponsor what decision she needed. She had been escalating in the corridor instead, which worked until the sponsor changed. Two lines fixed it permanently.
The report prompt that refuses to write "good progress" ★ 10 min · build your own
Now encode the rubric so you never grade by hand again. This prompt lives in your workspace; in b9 it becomes a skill that runs every Thursday against your real tools.
Paste the prompt below into your programme workspace, with your context pack loaded.
Feed it a genuinely messy week - raw notes, exported tickets, unedited threads. Do not tidy the input first; tidying it is the job you are trying to move.
Read the "gaps" block it returns before the report itself. That block is the model telling you what your week is actually missing - usually an owner or a real number.
Fill the gaps from the team, not from the model. Then re-run and send.
Try it yourself - this week ◐ 30-45 min total
- Write your RAG rule in one sentence per colour and get it agreed by your sponsor. This is the highest-value fifteen minutes in the whole track.
- Run this week's real report through the prompt. Send the version you fixed, not the version it drafted.
- Keep the GAPS block. After three weeks, look at what keeps appearing - that is a structural hole in your programme, not a reporting problem.
- Score three of your past reports out of six and note the pattern. Most people are missing the same two checks every time.
- Optional: bring your worst-scoring report to session b8, where the escalation and the hard message get built.
Official sources covered
Taught from the delivery reporting canon, PMI's AI guidance on oversight, and the Northwind case carried from the PMO course. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · In the simulator, which single lever is worth the most points - and why?
The RAG rule and owners tie at 17 points, and the rule is the floor: an undefined amber is the PM's mood, so two readers disagree about whether the programme is in trouble and nobody acts.
2 · Your draft says data quality is "looking reasonable". What is wrong with it?
A claim nobody can check is not a safe claim. The evidence lever exists to make the report challengeable; 96.4% against a 99% gate is a fact, "reasonable" is a feeling with a suit on.
3 · The GAPS block says an action has no owner. What do you do?
The gap is the finding. An inferred owner is an invented commitment, and "the team" is how an action survives untouched for a month. Fill gaps from people, never by inference.