---
name: mbr-kpi-briefing-author
description: "Load when an operator, GM, or Chief of Staff pastes a KPI metrics package — actuals, targets, and optional context notes — and asks for a monthly or quarterly business-review briefing, MBR narrative, ops review write-up, or 'what's working / what needs attention' summary. Not for chart generation, live data pulls, or freeform data exploration without targets. Returns a structured briefing: zone-classified table, materiality-ranked headline, what's working, what needs attention, recommendation with named levers, next-period focus list, and data-gap disclosures."
---

# Monthly Business Review Briefing

The table shows what moved; the notes say whether the move is real, borrowed, expiring or a symptom. Briefings go wrong in five places.

## Where briefings go wrong (read first)

**1. A borrowed or expiring beat gets celebrated.** Find each favorable number's mechanism in its note; each has its own forward profile and its own sentence:
- Promotion, holiday shift or early orders: demand pulled from next period; expect a payback dip. If the note gives an underlying figure below target, the metric is a miss in disguise and also belongs in Needs Attention.
- Rebate, refund or one-time gain: the true cost base is hidden. Cite the prior period as the reference level; do not compute an ex-credit figure yourself.
- Fixed-price contract, promotional rate or temporary capacity: the beat ends on a date. Give the date and the gap to the current market rate; renewal is a deadline decision.
- One exit or deal: not repeatable.
- Work or pipeline carried over from the prior period: this period's own engine is weaker than the total; quote the underlying count.
- Seasonal lift: say when it reverses and whether the de-seasonalized level clears target.
Different mechanisms never share one caveat.

**2. Direction is misread.** Lower is better for costs and expense ratios, durations and wait times, defects, complaints, returns, shrink, waste, churn, staff turnover, absence, delinquency, readmissions, infections, incidents, budget variances, workload ratios per staff member and prices paid for assets (purchase or entry multiples). Margins, growth, yield, on-time and completion rates, satisfaction, NPS, compliance and returns on capital are higher-is-better. Occupancy and utilization are levels to fill, unless the notes describe the excess as strain; then the target is a ceiling and anything above it is adverse.

**3. The headline goes to the biggest number or the mood of the month.** The headline is always an adverse row: off-track first, at-risk only if nothing is off-track. Exclude favorable or qualified beats, misses whose driver is already fixed and whose metric recovers mechanically (yield or uptime after a repair), and reported misses the notes call non-recurring with a recurring level inside target. Then:
- First, a row whose note records that earlier deals, batches or cohorts at this level underperformed.
- Next, a feedback loop a note states between two table rows (pressure that slows the very process that would relieve it); lead with its pressure metric.
- Next, an unresolved driver that will hit a named peak season, date, or share of supply or customers. Among several, the larger adverse % wins.
- Then the hub most other notes blame, then harm reaching customers, regulators or investors, and only then raw size.
- A metric approaching a regulatory or covenant trigger always gets a preventive action with the distance to the trigger, but headlines only if nothing above applies.

**4. Symptoms are reported as separate problems.** When a note cites another note's driver, it sits downstream. One driver behind several misses is one problem with several symptoms; say so once and link the items. A repair stops new damage but does not undo consequences already delivered (complaints from shipped product, renewals already lost, premiums already paid, backlog created); say which items recover by themselves and which do not. Link rows only where a note links them.

**5. Gaps get filled with plausible numbers.** A missing actual, prior period, cause, split or estimate does not exist. Write "not specified in the notes". Attribute drivers, dates and estimates to their note ID.

## Zones

For each row with an actual and a target:
- Equal to or better than target: **On-track**. Anything short of target, even 0.1%, is a miss.
- Miss: gap % = |actual - target| / |target| x 100, one decimal, from unrounded values. **At-risk** if 5.0 or less; **Off-track** if above 5.0. The comparison is arithmetic only: an improving trend, a non-recurring cause or a small point gap never softens it. Exactly 5.0 gets a footnote saying it sits on the boundary.
- Zero target: any adverse value is Off-track; give the gap in units.
- No actual or no target: **Unclassified**, no zone, no estimate; both Δ cells read n/a. Missing prior: "n/a (no prior data)", never a dash.
Qualifications never move a zone: a qualified beat stays On-track with a marker (*, †, ‡); a reported miss called non-recurring stays Off-track with its recurring level stated.

Read `references/operator-reading-guide.md` for the full set of note-reading rules and lever catalogue before drafting any item.

## Working notes, then the briefing

Start the reply with a section titled **Working notes**: one table line per row with direction, Δ vs target in units and %, the verdict written exactly as in the table rule (favorable at any size, else ≤5 adverse or >5 adverse, always on the reported actual) and the zone it implies; then the headline candidates, each tagged with the first test above it passes (past cohorts underperformed, self-feeding loop, dated hit, hub, customer harm, size); the earliest tag wins; then each beat's mechanism in one line; last, for each miss blamed on a one-month event, whether last period was already short of target (this check never changes the headline). Under 210 words. Then a line `---` and the briefing, beginning with the packet's title line.

## Shape of the briefing

Read `references/page-template.md` for the complete seven-section briefing template with table format rules before writing the briefing.

The first line is the packet's title.

Under 900 words with the table, each fact once; at-risk items get one or two sentences.