| name | weekly-ops-review |
| description | Generate the weekly paid-ads operating review — the one-page document that captures health, decisions, and asks for the marketing team. Pulls from google-ads-manager, paid-channel-recap, attribution-modeling, and the profit table. Use when asked to "run the weekly review", "generate the Monday report", "weekly ops report", "what happened last week", or any weekly recurring growth/marketing standup. |
| metadata | {"version":"1.0.0"} |
Weekly Ops Review
The Monday-morning ritual. One page. Health, decisions, asks.
This skill is the conductor — it doesn't generate new analysis, it pulls the analyses from google-ads-manager, paid-channel-recap, attribution-modeling, and the underlying campaign_performance table, then synthesizes them into a single review document.
When to use
- Monday or Tuesday mornings, before the weekly marketing standup
- Before a leadership 1:1 where you're presenting paid-ads health
- After a major launch (compressed format — daily for first week)
- Anytime someone says "wait, how are ads doing?"
What it produces
A markdown document with five sections:
- TL;DR — three lines. The number, the action, the question.
- Account health — week-over-week metrics on the dimensions that matter
- Channel snapshot — performance per channel with the tier label
- Open proposals — what's awaiting approval from last week's run
- Asks for the team — specific things only humans can answer
Output target: one page printed. ~600-800 words. Anything longer gets skimmed and ignored.
The five sections
1. TL;DR
Three lines maximum. The skill picks from these patterns:
- Status line. "Account profit-to-spend is [X.X]x — [in band / above target / below floor]."
- Action line. "[Specific change happening this week] / [no change, holding]."
- Question line. "One thing I need from you: [specific question]."
If there's nothing to ask, skip the question line. Don't pad.
2. Account health
A small table — current week vs prior week vs 4-week rolling.
| Metric | This week | Last week | 4-week avg | Δ |
|---|
| Spend | $___ | $___ | $___ | __% |
| Profit (real) | $___ | $___ | $___ | __% |
| Profit-to-spend | __x | __x | __x | ±__x |
| Orders | ___ | ___ | ___ | __% |
| New-buyer % | __% | __% | __% | ±__pp |
| Refund rate | __% | __% | __% | ±__pp |
The skill flags any cell that's outside its expected band with a ⚠️.
3. Channel snapshot
For each active channel, a one-line health entry:
[Channel name] — $X spent, Y.Yx P:S, [PUSH/SCALE/TEST/PAUSE] tier. [One-line observation.]
Tier definitions (from google-ads-manager):
- PUSH — P:S above 2.0x. Budget increase proposed (with incrementality caveat).
- SCALE — P:S 1.5x-2.0x. Hold or modest shift toward.
- TEST — P:S 1.0x-1.5x. Investigate.
- PAUSE — P:S below 1.0x for 2+ consecutive weeks. Recommend pause.
The "one-line observation" is what humans actually use. Make it specific. "Brand search P:S dropped 0.3x because impression share lost to competitor [name] in IS auctions" beats "performance declined."
4. Open proposals
Anything from last week's change-proposal output that hasn't been actioned. Three states:
- Pending your approval: [link to proposal] — sitting in your queue since [date]
- Applied: [proposal] applied [date]. Outcome to date: [observation]
- Rejected: [proposal] you said no on [date]. Anything change?
The point of this section: don't lose changes in the workflow. The week-over-week loop only closes if approved changes actually happen.
5. Asks for the team
Specific things only humans can answer:
- "[Person] — can I get an updated COGS number for [SKU]? Last data point is from [date]."
- "[Person] — is the Q2 launch still on track for [date]? Affects the budget reservation."
- "[Anyone] — saw a 30% spike in branded search volume from [region]. Anyone know what's happening there?"
If there are no asks, skip the section. Don't invent them.
The data pipeline this skill consumes
| Source | What it provides |
|---|
attribution-modeling profit table | Real profit-to-spend per campaign per day |
google-ads-manager | Last week's circuit-breaker triggers, change proposals, search-term mining results |
paid-channel-recap (if active campaign window) | Per-channel performance rolled up |
paid-ads-context.md section 3 | Target P:S band, daily budget cap, refund rate band |
paid-ads-context.md section 7 | Cadence rules (when this report fires) |
Cadence
By default, runs once a week. Cron expression in paid-ads-context.md section 7.
Suggested setup:
- Monday 9am account-time: the skill fires. Output goes to a Notion page (or Slack channel) named "Weekly Ops Review."
- Monday 10am: operator reads, asks the team for clarifications.
- Monday 11am: operator runs
google-ads-manager change-proposal step for the week.
- Monday-Tuesday: approvals happen.
- Tuesday afternoon: changes applied.
Compressed alternative — for high-spend or high-velocity accounts:
- Daily mini-review (5 metrics, no channel breakdown) at 9am
- Full weekly review on Monday only
What this skill does NOT do
- It does NOT make budget decisions on its own. It surfaces them.
- It does NOT auto-apply changes. That's
google-ads-manager's job after human approval.
- It does NOT replace deeper analysis. Surface anomalies; the operator decides whether to dig in.
- It does NOT write a "weekly newsletter" style report. The output is operational, not communications.
Audience
The output is for ONE person: the marketing operator running the account. Not for stakeholders, not for partners, not for a wider audience.
If you need a stakeholder-facing version: use paid-channel-recap instead, which has the redaction policy baked in.
Output format options
- Markdown (default) — for Notion, internal wiki, or piped into Slack
- HTML dashboard — for a more visual version; uses the same template style as
paid-channel-recap
- Slack message — compressed to fit in a 1500-char message; TL;DR + bullets only
- Voice memo summary — 90-second audio summary for the operator's commute
Set the output format in the skill invocation:
Run weekly-ops-review. Output: markdown.
Run weekly-ops-review. Output: slack-compressed. Post to #marketing-ops.
Anti-patterns
| Anti-pattern | Why | Fix |
|---|
| Long-form weekly review | Nobody reads past page 1 | One page. Cut. |
| Generic observations ("performance was mixed") | Tells the operator nothing actionable | Force specificity: campaign name, exact number, suspected cause |
| No flags / no callouts | The whole point is to surface anomalies | Flag any metric outside its band |
| All metrics, no opinion | Data without interpretation isn't a review | Every section has at least one "here's what to do about it" line |
| Includes celebration / praise | The review is operational; celebration belongs in a different forum | Keep tone neutral and factual |
| Misses last week's open proposals | Decisions get lost | Section 4 is non-negotiable |
Related skills
| Skill | Relationship |
|---|
paid-ads-context | Provides targets (section 3) and cadence (section 7) |
attribution-modeling | Provides the profit table this skill reads |
google-ads-manager | Provides circuit-breaker results, change proposals, search-term mining |
paid-channel-recap | Different audience — stakeholders. Different format. Sister skill. |
ad-creative | Creative winners from last week feed forward as input here |
customer-research | If the review surfaces a question like "why did this audience tank," may trigger a research round |
n8n-recipes | The cron + Slack delivery pattern lives there |
References
