Where we are
In a1 you split the delivery job into what AI drafts and what only a person decides. In a2 you set the governed boundary: what goes into a workspace, what never does, who is allowed to connect what. In a3 you built a review rubric for AI-drafted plans, and in a4 you did the same for risk, dependencies and escalation. Those four sessions were about artifacts your team produces. Tonight is different. Reporting is the artifact the organisation consumes, weekly, forever. It is where your function is judged, and it is the single most common first place AI lands in a delivery team - which makes it the most common place to go fluently, confidently wrong at scale.
The week, as a loop 6 min live
Most reporting is not a process, it is a Friday afternoon. Somebody assembles four threads, a ticket export and a spreadsheet into a document, ships it at six, and starts again on Monday. That works until the person is on leave, or the programme doubles, or the sponsor asks for a second one. A cadence fixes it: five small stops instead of one large heroic one, each with an owner, and only two of them needing a human in the room.
LiveThe five stops, and who owns each one4 min▶
The point of naming stops is that each one becomes small enough to survive a holiday. Nobody has to be heroic on a Friday, because Friday is now fifteen minutes of reading and signing.
- Monday - refresh the context pack. Swap in the current plan, the current risk register, and last week's report. Five minutes, usually a PMO analyst. A stale pack is the most expensive failure in this whole cadence, because it produces confidently wrong answers about last month's programme, which is far harder to catch than obvious nonsense.
- Tuesday - risk and dependency sweep. What moved: gate forecasts, new risks, risks that closed, dependencies that slipped underneath you. On Northwind this is the stop where R-03, the data-engineer hiring gate, gets its honest forecast before it becomes a surprise.
- Wednesday - draft plus the gaps block. The draft is cheap. The gaps block is the valuable half: the list of things the material could not support. Read that first, then go and fill it from people.
- Thursday - steerco and decisions. The ask gets answered, or it does not, and either way it gets logged with an id and a date. No model participates in this stop.
- Friday - log and actions out. Decisions written into the log, actions distributed with one owner and one date each, report signed by a named human and sent.
The programme that survived a holiday. A head of delivery ran a portfolio where one PM assembled every weekly pack alone. She took two weeks off in August and the pack simply did not go out - twice. Nobody noticed until a dependent team built a sprint on a date that had moved. After the cadence went in, the Monday refresh and Wednesday draft ran without her, and the cover person's whole job was Thursday and Friday: attend, decide, sign. The pack went out both weeks she was away, and the delta section is what caught the moved date.
Self-studyWhy the two human stops are Thursday and Friday2 min read▶
You could put a person at all five stops. Plenty of organisations do, call it oversight, and get nothing for it - because review that happens everywhere happens nowhere. PMI's 2026 AI standard is unusually direct about this: oversight needs clear structures, roles, authority and guardrails, with real intervention triggers, rather than a perfunctory review layered over everything.
So concentrate the human where intervention is actually possible. Thursday is where a decision gets taken, and a model cannot take one. Friday is where accountability attaches: one name, and that person read it, especially the numbers.
Monday, Tuesday and Wednesday are assembly. If something goes wrong there, Thursday and Friday catch it, which is what an intervention trigger means in practice. Two real checkpoints beat five ceremonial ones every time.
The six checks, and what each one buys you 5 min live
Your PMs build these in the PM track. You need something different: the ability to pick up any report in your portfolio and know in thirty seconds whether it can be acted on. Six checks, and each missing one produces a specific, recognisable failure - not a vague drop in quality. Learn the failures and you can review by symptom. Send your PMs to session b6 in the PM track to feel it: the same six checks are scored live there, a Northwind week 9 report climbing from 24 out of 100 with every lever off to 100 with all six on. Watching the score jump when the RAG rule goes in does more for adoption than any memo you will write.
| The check | What it buys you | The failure when it is missing |
|---|---|---|
| Status by a written RAG rule | A colour that two people read the same way, stated in the report itself | The colour is a mood. Amber runs for eleven weeks and nobody acts, because nobody agreed what it meant |
| The delta since last week | Movement, which is the only thing a sponsor can actually steer | A 14-day slip reads exactly like a stable plan. The photograph is fine, the film is missing |
| One named owner and one date per action | A commitment a specific person can be held to next week | Orphan actions. "Follow up on data quality" reappears for a month, reworded each time |
| Decisions taken, logged | An audit trail, and an end to relitigating settled things | The same decision gets taken three times, slightly differently, by three different groups |
| Evidence behind every claim | A report that can be challenged, which is the only kind that can be trusted | Unfalsifiable prose. "Quality is looking reasonable" cannot be argued with or relied on |
| The ask | A steerco that decides something instead of nodding at something | It is a broadcast. The real escalation then happens in a corridor, off the record |
LiveReading a report by symptom, in thirty seconds3 min▶
You are not grading prose. You are looking for six structural features, and their absence is visible before you read a sentence.
- Scan for a colour and a rule on the same page. If the rule is not restated, it lives in someone's head and you have a mood ring.
- Scan for two numbers of the same kind. This week and last week. One number is a photograph.
- Scan the actions column for names. Any instance of "the team", "engineering" or "TBC" that has been TBC for three weeks is a finding.
- Scan for ids and for the last paragraph. D-14, R-07, NWD-412, a metric with a denominator - no ids means no evidence, whatever the adjectives say. And if the closing paragraph does not name a decision, options with costs and an expiry date, the report was written to inform, not to move anything.
Amber for eleven weeks. A programme reported amber every week for a quarter with no written definition of amber. Every steerco nodded. In week twelve it went red, and the sponsor asked why nobody had flagged it. The honest answer was that everybody had - in a colour nobody had defined. The report was never lying. It simply could not carry information, and no amount of AI drafting would have changed that, because the missing thing was a rule, not a paragraph.
Self-studyWhat the checks look like on Northwind week 92 min read▶
Northwind is nine weeks in. M3, raw ingest signed off, was gated for 10 August and now forecasts 24 August. Six of nine sources are landing, 812k rows a night, and data quality is 96.4% against a 99% gate. Here is the same week with the checks on and off.
| Check off | Check on |
|---|---|
| "Amber - some challenges with ingest" | "Amber: M3 forecasts 24 August against a 10 August gate, with a recovery plan attached. Amber means a gate is 2 or more weeks late and recoverable" |
| "Data quality is improving" | "DQ 96.4% this week against 96.1% last week, gate is 99%. Ticket NWD-412" |
| "The team will chase the vendor" | "Priya N. to hold the Veridian escalation call by 6 August, VN-2291" |
| "We discussed contract cover" | "D-14, 29 July, Elena V.: hold contract cover pending the 6 August vendor call" |
| "We may need more resource" | "Decision needed by 8 August: 2 contract engineers for 4 weeks, about 38k, or M4 moves. After 8 August the option lapses" |
Nothing in the right column is longer prose. It is the same length with facts in it. That is the whole shift.
When the reports get too good 5 min live
Here is the failure mode that only appears after AI drafting works. Every report in your portfolio becomes well-structured, correctly formatted, evenly toned and pleasant to read. And you quietly lose a signal you had been using for years without ever writing it down: the sound of a PM who was worried. PMBOK 7's stakeholder and measurement domains both treat the report as a two-way instrument - it informs, and the reaction to it informs you back. Uniform tone attacks the second half, so you have to replace that channel on purpose.
LiveThree habits that keep the signal3 min▶
You cannot un-improve the writing, and you should not want to. What you can do is stop relying on tone as your early-warning system and replace it with three deliberate habits.
- Read the gaps block, not just the report. The report is what could be supported. The gaps block is what could not - the action with no real owner, the number nobody could source, the date that came from an assumption. Three weeks of gaps blocks tells you more about a programme than three months of reports, and it is the part that has not been smoothed.
- Ask one question no report can answer. Every week, one PM, one question: "what are you worried about that is not in here?" or "what would you tell me if this were not written down?" It takes ninety seconds and it restores the channel the tone used to carry.
- Watch for the report nobody reads. Drafting is cheap now, so reports grow. A five-page pack that is beautiful and unread is a worse instrument than the scruffy one page people argued with. If nobody has challenged a line in a month, that is not quality. That is silence.
The PM who stopped sounding worried. A delivery lead had one PM whose reports she always read twice, because when he was anxious his sentences got shorter. After the team standardised on a drafted format, his reports read exactly like everyone else's - clean, complete, calm. Two months later a vendor issue surfaced late, and when she went back through the packs the warning was there every week, in the gaps block, in a line about an owner nobody could confirm. She had been reading the report and skipping the gaps. Now the gaps block goes first in her portfolio review, and the reports go second.
Design your own cadence ★ 12 min · everyone builds
Take your own reporting week and fill a five-day grid: what is drafted, what is reviewed, and who signs. Do it for one real programme, not the portfolio - the portfolio version is just this repeated. Most people find they have three stops crammed into one afternoon and no owner named for any of them.
Write down what actually happens today, honestly. Who assembles the pack, on which day, from how many sources, and how long it takes them.
Split it into the five stops. Refresh, sweep, draft, decide, sign. Some of yours will be empty, which is the finding.
Put a named person against each stop - a role is not enough, because a role cannot be on leave. Then name the cover person for each.
Mark the two stops that require a human in the room. If you have marked more than two, ask what intervention each one is actually capable of - review with no power to change anything is theatre. Then write the one thing that would break the loop, which is usually the Monday refresh, because it is small, boring and nobody's actual job.
Write the RAG rule with your sponsor ★ 10 min · build your own
This is the highest-value fifteen minutes in the leader track, and it is one meeting. An agreed, written RAG rule turns a colour from an opinion into information, and it is the only one of the six checks that a PM cannot install alone - it needs your sponsor's agreement, which means it needs you.
| Colour | A workable default to start from |
|---|---|
| Green | Every live gate forecasts on or before its date, and no critical risk is open above the agreed threshold |
| Amber | A gate forecasts 2 or more weeks late, and a credible recovery plan is attached to this report |
| Red | A gate is unrecoverable without a steerco decision on scope, money or the date |
Take the default above into the meeting already written. Never open with "how should we define amber" - you will get a philosophy session. Open with a draft and invite edits.
Test it against three real past weeks from your own portfolio. If the rule would have coloured a week differently from how you actually coloured it, that is the conversation worth having.
Agree the threshold numbers out loud: how many weeks late is amber, what counts as a critical risk, and who can overrule the rule. Somebody usually can, and it is better to name them than to discover them.
Agree that the rule gets restated in one line inside every report, then put it in the house rules file the same day, before anyone's memory of the meeting drifts. That one line removes the "what does amber mean here" conversation that eats the first ten minutes of every steerco.
Fifteen minutes, then nine months of arguments that did not happen. A PMO lead had spent two years watching steercos open with a debate about what the colour meant. She took a drafted rule to her sponsor, tested it against three past weeks, and got it agreed before the coffee arrived. The rule itself was unremarkable. What changed was that the first ten minutes of every steerco stopped being about interpretation and started being about the ask - and the programme that had been amber for a quarter was red within two weeks, on purpose, with a decision attached.
Try it yourself - this week ◐ 30-45 min total
- Fill the five-day grid for one real programme, with named people and named cover. Note which stop has no owner today.
- Book the fifteen minutes with your sponsor and take the drafted RAG rule in with you. Do not go in with a question.
- Pick three reports from your portfolio at random and score them against the six checks. Write down which check is missing most often - it is almost always the ask or the rule.
- Ask one PM the question no report can answer: what are you worried about that is not in here. Then do it again next week with a different PM.
- Set a hard length for the weekly report in your house rules, before drafting makes everything three pages. Then send your PMs to the b6 simulator and ask what their own last report scored out of six, without asking whose is whose.
Official sources covered
Taught from the delivery reporting canon, PMI's public AI guidance, and the Northwind case carried through both tracks. Certification and the full normative text of PMI's standards stay with PMI. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · You are designing the reporting week. How many stops should require a human in the room, and which?
Oversight that happens everywhere happens nowhere. PMI's standard asks for real intervention triggers, not a review layer over everything. Concentrate the human where intervention is possible: the decision, and the signature.
2 · Every report in your portfolio is now well-structured, evenly toned and pleasant to read. What have you lost?
You used to hear when a PM was worried. Uniform drafting removes that channel, so replace it deliberately: read the gaps block before the report, and ask one PM a week what is worrying them that is not written down.
3 · A programme has reported amber for eleven consecutive weeks. What is the first thing to fix?
Eleven weeks of amber is the classic undefined-colour failure. Without a written rule the colour is a mood, so nobody can act on it and nobody can argue with it. One fifteen-minute meeting fixes it permanently.