Ad performance review
Daily read-only ad-performance runbook — budget pacing, CPA/ROAS drift, underperforming ads and keywords, and anomaly detection across Google Ads and Meta Ads, plus how to rank and word optimization recommendations for {{alert_channel}}.
Overview
Turn raw daily spend and performance numbers from Google Ads and Meta Ads into a ranked list of findings and optimization recommendations. A daily cron spawns a fresh session with read-only access to both platforms; this skill turns those numbers into pacing checks, drift signals, underperformer flags, and anomaly detection, then a specific recommended action for each. The agent never acts on its own findings — every recommendation is for a human to execute.
When to load
- The daily cron fires the ad-performance scan.
- A human asks for the current state of a campaign, or why a recommendation was made.
Workflow
Step 1 — Pull spend and performance from Google Ads (read-only)
For every active campaign, ad group, ad, and keyword, pull: daily spend, monthly budget, clicks, impressions, conversions, cost per acquisition (CPA), and return on ad spend (ROAS) where conversion value is tracked. Pull the trailing 30-day baseline for the same metrics.
Step 2 — Pull spend and performance from Meta Ads (read-only)
For every active campaign, ad set, and ad across Facebook and Instagram placements, pull the same fields: daily spend, budget, clicks, impressions, conversions, CPA, and ROAS, plus the trailing 30-day baseline.
Step 3 — Check budget pacing
For each campaign, compare spend-to-date against the elapsed fraction of its budget period.
| Pacing signal | Criteria |
|---|---|
| On pace | Spend-to-date within ±15% of the expected linear pace |
| Overpacing | Spend-to-date more than 15% ahead of expected pace — will exhaust budget early |
| Underpacing | Spend-to-date more than 20% behind expected pace — budget is going unused |
Step 4 — Check CPA/ROAS drift
Compare each campaign's/ad's trailing 7-day CPA and ROAS against its 30-day baseline.
| Drift signal | Criteria |
|---|---|
| Stable | CPA within ±15% and ROAS within ±15% of baseline |
| CPA drift | CPA up more than 15% vs. baseline with flat or falling conversions |
| ROAS drift | ROAS down more than 15% vs. baseline |
| Compounding drift | CPA up AND ROAS down in the same window — flag first |
Step 5 — Flag underperforming ads and keywords
| Underperformer signal | Criteria |
|---|---|
| Low-CTR ad | Click-through rate less than half the ad set's/ad group's average, with 1,000+ impressions |
| High-cost, no-convert keyword/ad | Spend above the account's median per-unit spend with zero conversions over the trailing window |
| Wasted-spend keyword | Search term with 3+ clicks and zero conversions over 14+ days (Google Ads search terms report) |
Step 6 — Detect anomalies
Flag anything that breaks the normal pattern rather than drifting gradually:
- A single day's spend more than 2x the campaign's trailing 7-day daily average.
- A sudden CTR drop of more than 50% day-over-day on a previously stable ad.
- A campaign that stopped delivering (near-zero impressions) while still marked active with budget remaining.
Step 7 — Draft the recommendation for each finding
Every finding gets exactly one concrete, human-actionable recommendation:
- Overpacing → suggest lowering the daily budget or adding a spend cap so it doesn't exhaust the period early.
- Underpacing → suggest raising bids, widening targeting, or reallocating the unused budget toward a better-performing campaign.
- CPA/ROAS drift → suggest pausing the specific ad or keyword driving it, or shifting budget toward the campaign's best-performing ad.
- Underperforming ad → suggest pausing that ad and shifting its budget to the top performer in the same ad set/ad group.
- Wasted-spend keyword/search term → suggest adding it as a negative keyword.
- Anomaly → surface it plainly with the numbers; recommend investigating before assuming it's a pacing or drift issue.
Step 8 — Rank and post
Sort findings highest-impact first (compounding drift and overpacing that risks exhausting budget rank above a single underperforming keyword). Post exactly one message to {{alert_channel}}: for each finding, the campaign/ad/ keyword, the specific numbers, and the recommended action. Post exactly once per run — this is a fresh session, so there is no prior list to update or diff against; the whole scan is recomputed and reposted every day.
Guardrails
- Read-only, always. Google Ads and Meta Ads are read-only connectors. Never write to either platform, even if the credential would technically allow it.
- Recommend, never act. Every finding carries a suggested action — pause, shift budget, add a negative keyword — but the agent never pauses a campaign, changes a budget, or edits an ad itself. That action stays with the marketing team.
- One output. The Slack post to {{alert_channel}} is the only thing that leaves the sandbox. No campaign changes, no bid changes, no budget edits.
- No memory between runs. Each run is a fresh session; recompute pacing, drift, and underperformers from the current state of both platforms rather than assuming anything from the prior day's list.
- Scoped secrets. Google Ads and Meta Ads access is brokered server-side through connectors. No raw credential is ever pasted into chat.
- People decide, not the agent. The list flags risk and suggests a next step; a human on the marketing team decides whether and how to act.
