Weekly performance digest

Weekly Performance Digest

A skill for producing publication-ready weekly Google Ads performance reports.

What this skill does

Given two weeks of campaign data (this week vs. last week, plus same week last month for MoM context), produces a digest with:

1. Executive summary (3 sentences)
2. Key metrics table with WoW and MoM comparisons
3. Campaign highlights (top 3 + bottom 3)
4. Anomalies and flags
5. Prioritized action items for next week

Output is plain-language, non-technical - sendable directly to stakeholders.

Required inputs

1. This week's campaign data (last 7 days)
2. Last week's data (the 7 days before that)
3. Same week last month (for MoM comparison; optional but recommended)
4. Known changes context - any new campaigns, paused ads, budget shifts, creative updates the user made

Required columns

For each period:

• Campaign name
• Spend, Impressions, Clicks, CTR, CPC
• Conversions, CPA, Conv. rate
• Conv. value / Revenue (if tracked)
• ROAS (if tracked)

Workflow

Step 1: Validate data alignment

Both periods must have:

• Same set of campaigns (or note any added/removed)
• Same column structure
• Comparable date ranges (7 days each)

If a campaign was added or removed between periods, note it explicitly - comparisons across asymmetric periods are misleading.

Step 2: Compute aggregates and deltas

Account-level totals (this week vs. last week):

Metric	Formula
WoW change %	(this - last) / last × 100
MoM change %	(this - same_week_last_mo) / same_week_last_mo × 100

Flag any metric where:

• WoW change > ±15% (notable)
• MoM change > ±25% (significant trend)

Step 3: Identify campaign highlights

Top 3 performers: ranked by absolute conversion or revenue gain WoW Bottom 3 performers: ranked by absolute drop in conversions OR CPA spike

For each, write a 1-sentence "what changed and why" - tied to the data, not speculation. If the cause isn't clear from the data, say so:

"Generic Search lost 18 conversions WoW (down 22%). Cause unclear from current data - recommend checking search terms report."

Step 4: Anomaly detection

Flag campaigns where:

• Spent >20% of weekly budget with zero conversions
• CPA jumped >25% WoW
• Conv. rate dropped >20% WoW
• IS lost to budget jumped >10 percentage points

For each anomaly, include:

• Campaign name
• The specific metric that moved
• Possible cause (if obvious from data)

Step 5: Action items

3–5 specific, prioritized tasks for next week. Each item:

• Action: what to do (specific, not "review the account")
• Expected impact: estimated dollar/conversion result
• Owner: who should do it (PPC manager, designer, dev, client-side, etc.)

Examples:

• "Add 12 negative keywords to Generic Search campaign (saves est. $400/week)"
• "Investigate landing page conversion rate drop on Brand campaign - possible site change last week"
• "Increase Long Tail campaign budget from $500/day to $700/day - IS lost to budget = 35%, est. +18 conversions/week"

Output format

Read references/digest-template.md for the full template.

A markdown response structured as:

# Google Ads Weekly Digest - [Date Range]

## Executive Summary
[3 sentences: spend, conversions, CPA, ROAS this week + biggest win + biggest concern]

## Key Metrics
[Table: Metric | This Week | Last Week | WoW Δ | Last Month Same Week | MoM Δ]

## Campaign Highlights

### Top performers
- [Campaign A]: [what changed]
- [Campaign B]: [what changed]
- [Campaign C]: [what changed]

### Bottom performers / concerns
- [Campaign X]: [what changed]
- [Campaign Y]: [what changed]
- [Campaign Z]: [what changed]

## Anomalies
[Bulleted list of campaigns + metric movements meeting flag thresholds]

## Action Items for Next Week
1. [Action] - Impact: [estimate] - Owner: [role]
2. ...

If the user says this is for a stakeholder presentation, generate a Word doc using the docx skill. If they say "for my dashboard," generate a markdown response and offer to also create an Excel workbook with the data tables.

Output style requirements

This is the most reader-facing skill in the set. Style matters:

• Use actual numbers, not vague descriptions. Not "spend was up significantly" - say "$42,300, up 18% WoW."
• Plain language, not PPC jargon. Replace "CPA" → "cost per conversion" on first use, can use the abbreviation after.
• Frame metrics in business impact. "CPA dropped $8" → "We're acquiring customers $8 cheaper this week, saving ~$1,600 at current volume."
• Don't editorialize beyond what the data supports. "Strong week" is fine if numbers warrant it; "incredible breakthrough" is not.
• Active voice. "We launched a new campaign" beats "A new campaign was launched."

What this skill must NOT do

• Don't use PPC jargon without translation. Recipients may not be PPC experts.
• Don't speculate about causes you can't substantiate. "CPA might have spiked because of seasonality" - only say this if you have data showing it. Otherwise, "CPA spiked; cause unclear, recommend investigation."
• Don't pad. If the week was uneventful, say so in 3 sentences and move on.
• Don't recommend more than 5 action items. Stakeholders won't read more than that.
• Don't invent metrics. If the user didn't include ROAS data, don't compute or report ROAS.
• Don't write headers or sections that have nothing to say. Better to skip a section than say "no anomalies this week" with bullet padding.

Reference files

• references/digest-template.md - full reference template with example
• references/style-guide.md - language and framing rules with do/don't examples