Weekly Review — the numbers, the diagnosis, the one focus

Marketing drifts not from lack of effort but because nobody sits the numbers next to the plan each week and asks what changed. This is the 30 minutes that prevents four confident weeks in the wrong direction. Same shape every week, one page, so week twelve reads against week two.

Inputs

• This week's numbers, or where to pull them from: $ARGUMENTS
• From marketing-brief.md if present: numbers that matter (north star, drivers, baselines), channels & funnel, any live experiments, and last week's focus if recorded. If no metrics tree exists, review on pipeline basics — reach, leads, conversations, sales, spend — and note that /marketing:kpi-tree is overdue.

Do this

1. Lay out the table: each driver metric with actual, plan, and last week, marked up, flat, or down. Missing numbers get written down as gaps, not skipped — an unmeasured driver is a decision being made blind.
2. Find the biggest gap and follow it down to the stage that actually broke: fewer sales, because fewer calls, because reply rate halved. Name the stage, not the symptom.
3. Separate weather from climate. One bad week is noise; three moving the same way is a trend. Only trends and cliffs justify changing the plan — reacting to weekly noise is its own failure mode.
4. Close last week's focus. Did it happen, did it work, what did it teach? Unfinished focus rolls forward or gets killed out loud — never silently dropped.
5. Check the live experiments from the brief's Learnings & experiments section: any that hit their decision rule get their verdict called now, not "when there's time".
6. Set the ONE focus for next week: the single highest-impact action against the stage that broke, with an owner and what "done" looks like on Friday. If two feel essential, the diagnosis is not finished.
7. Sweep for brief updates. Any new number, learning, or dead assumption from this week gets a dated line in marketing-brief.md so the department's memory stays current.

Output

The one-page review, identical format every week — the value compounds when reviews are comparable:

• The numbers table — actual, plan, last week, direction per driver.
• The diagnosis in three sentences, naming the stage that broke.
• Last week's focus verdict, and any experiment verdicts that came due.
• Next week's one focus, its owner, and what done looks like on Friday.
• The data-gaps list — what could not be measured this week.

Rules

• Numbers first, narrative second. "Feels slow" is not a finding until a number agrees.
• Evidence over invention: data gaps are reported as gaps, never patched with estimates dressed as facts.
• One focus. A review that produces five priorities has produced none.
• Write misses plainly. A review that always reads well is being written for an audience, not for a decision.
• Thirty minutes, hard stop. If it takes longer, fix the measurement, not the meeting.