Exec Update

One page. Read in 90 seconds. Five numbers, three calls, one ask, one kill line. Everything else below the fold. Weekly or biweekly, matching your exec team's rhythm. The five scoreboard numbers never change between updates.

The Template

# [Team / Product Area] Update, [date]
From: [name]   Cadence: [weekly / biweekly]   Read time: 90 seconds

## Scoreboard
| # | Metric | This period | Trend | One line of context |
|---|--------|-------------|-------|---------------------|
| 1 | Direction metric: [the number your strategy says should move, e.g. weekly active accounts using X] | [value] | [↑ ↓ →] | [why it moved, one line, no spin] |
| 2 | Eval trend: [score on your most important AI-touched surface] | [value] | [↑ ↓ →] | [what changed in quality, one line] |
| 3 | Cost per outcome: [margin or unit cost, e.g. cost per resolved ticket / per generated report] | [value] | [↑ ↓ →] | [driver, one line] |
| 4 | Adoption: [uptake of the current bet, e.g. % of target accounts activated] | [value] | [↑ ↓ →] | [one line] |
| 5 | Top customer-signal theme: [the most frequent theme from calls/tickets this period] | [theme name, n mentions] | [new / rising / steady] | [one line] |

RULE: same five metrics every update. A flat number with an honest
explanation beats a swapped-in green one. Changing metrics is metric
shopping and the reader knows it.

## Three calls I made
1. **Call:** [decision, one line]
   **Evidence:** [what you knew that drove it, one line]
   **Reversibility:** [two-way door, walked through it / one-way door, why]
2. **Call:** [...]
   **Evidence:** [...]
   **Reversibility:** [...]
3. **Call:** [...]
   **Evidence:** [...]
   **Reversibility:** [...]

RULE: decisions, not activities. "Shipped the redesign" is activity.
"Chose to ship the redesign before the pricing test because the funnel
data pointed at onboarding" is a call. Fewer than three real calls?
List fewer. Do not pad.

## One thing I need
**Decision needed:** [the yes/no or A/B, one line]
**By:** [date]
**Case:** [one short paragraph, the strongest honest version]
**Default if I hear nothing:** [what you will do absent an answer]

RULE: an ask is a decision, not an FYI. No real ask this period?
Write "No ask this week." A fabricated ask trains readers to ignore
the real ones.

## What I killed or declined
[One line: the request, the no, the reason. e.g. "Declined the
one-off integration ask for [prospect]: no signed commitment, no
second customer asking. Revisit on LOI."]

---
## Below the fold (appendix)
- [Item, max 3 lines. Nothing here requires action; actions live above.]
- [Item]
- [Item]
- Dashboard: [standing link for anyone who wants raw numbers]

Worked Example (sanitized fragment)

## Three calls I made
1. **Call:** Paused the integrations workstream two weeks to put both
   engineers on eval coverage.
   **Evidence:** Eval score on the summary feature fell a second
   straight week while integration demand stayed flat in call themes.
   **Reversibility:** Two-way door. Integrations resume on the 18th
   regardless.

## One thing I need
**Decision needed:** Yes/no on shifting the open contractor budget
to the eval pipeline this quarter.
**By:** Friday EOD.
**Case:** Eval coverage is the bottleneck on shipping the next two
bets; the contractor req has been open six weeks with no pipeline.
**Default if I hear nothing:** I shift it Monday.

The Agent Prompt Block

Run on your schedule (e.g., Thursday 4pm for a Friday-morning send). Wire to your analytics export, decision log, and call-summary store. Then do the 10-minute human pass below. Do not send unreviewed.

You draft a one-page exec update in a fixed format. You will receive:
- ANALYTICS: this period's values for 5 named metrics, plus last
  period's values.
- DECISION LOG: rows added since the last update.
- CALL THEMES: theme rollups from customer call summaries this period.
- LAST UPDATE: the previous update, for continuity and trend arrows.

PRODUCE, in this exact order, nothing else:

1. SCOREBOARD: the same 5 metrics as LAST UPDATE, in the same order.
   Value, trend arrow (↑ ↓ →) computed against last period, and one
   line of context per metric. Never substitute a metric. If a value
   is missing from ANALYTICS, write "MISSING, verify" instead of
   guessing.

2. THREE CALLS: select the 3 most significant rows from DECISION LOG
   (significance: team-time committed, reversibility, disagreement).
   For each: the decision (one line), the evidence field verbatim
   from the log, and the door type. If fewer than 3 rows exist, output
   fewer. Never invent a decision.

3. ASK PLACEHOLDER: output exactly "[ASK: written by human]". Do not
   draft the ask.

4. KILL LINE: if the DECISION LOG contains a declined/killed item this
   period, summarize in one line. Otherwise output "[none logged,
   human confirm]".

5. BELOW THE FOLD: up to 5 items from remaining material, max 3 lines
   each, no item that requires reader action.

STYLE: short sentences. No adjectives of enthusiasm. No "great
progress." Numbers carry the tone.

The 10-Minute Human Review Pass

1. (4 min) Verify every scoreboard number against source. One wrong number costs a quarter of credibility. Anything marked MISSING gets resolved or the update does not go out.
2. (3 min) Rewrite the three calls in your own voice. The agent gets the facts; you supply the judgment framing. Check the reversibility labels yourself.
3. (2 min) Write the ask. From scratch, every time. Decision, deadline, default. The ask is the one part of the update that is pure judgment, so the agent never drafts it.
4. (1 min) Read the whole page in 90 seconds. If you cannot, cut from below the fold first, then from context lines. The fold is a promise. Keep it.

That's the format. Five numbers that never change, three calls with evidence, one ask with a default, one line of nos. The agent does the mechanical 80%, the 10-minute pass does the judgment, and the time series of ten updates in a row tells the story no single update can.