Ad Campaigns
Category: Marketing · Difficulty: High · Tags: campaigns · ads · metrics · anomalies
What it does
Launch paid ad campaigns from a prepared marketing plan, monitor daily metrics, detect anomalies, and apply gated remediations. Two-phase design:
- Phase A — Build & Launch: validate prerequisites → pick channels → get budget approval → build creatives → build audiences → launch → verify.
- Phase B — Monitor & Iterate: pull daily metrics → detect anomalies → recommend actions → apply remediations (with approval gates) → weekly synthesis.
The agent switches phases automatically based on whether a launched sheet already exists in the workbook.
Steps
| # | Step | What happens |
|---|---|---|
| 1 | Validate prerequisites | Confirms marketing-plan exists, positioning is locked, budget is set, creative brief is ready. |
| 2 | Select + budget channels | Ranks channels (Google, Meta, LinkedIn, TikTok) against ICP + budget. |
| 3 | Get budget approval | Posts approval request to the user/boss node before spending. |
| 4 | Build ad creatives | Drafts copy variants, sources images, writes headlines + CTAs. |
| 5 | Build target audiences | Defines look-alike / interest / retargeting audiences per channel. |
| 6 | Launch campaigns | Submits campaigns via platform dashboards (Firefox inside the agent’s desktop VM). |
| 7 | Verify launch + handoff | Confirms campaigns are live, tracking URLs fire, no setup errors. |
| 8 | Pull daily metrics | Scrapes platform dashboards daily — CTR, CPC, CPA, ROAS, impressions, spend pace. |
| 9 | Detect anomalies | Flags pacing off >20%, CPA 1.5× target, CTR decay, creative fatigue, audience exhaustion. |
| 10 | Recommend actions | Proposes pauses, budget shifts, creative rotations, audience refreshes. |
| 11 | Apply remediations | Executes approved actions on the platforms. |
| 12 | Weekly synthesis + reflection | Rolls up the week, logs learnings, proposes next week’s plan. |
Guardrails built in
- Real-money guardrail — every spend decision above the node’s starting-context threshold triggers
request_approval. The skill blocks until approved. - Positioning lock — if an anomaly suggests positioning is broken (e.g. CTR below 0.5% across all audiences), the skill routes the failure back upstream to
market-companyinstead of throwing more money at broken creatives. - Never source metrics from training data — all dashboards are scraped live via Firefox.
How to invoke from Claude Desktop
do("Launch this week's paid campaigns from our marketing-plan.
Target budget $10k/wk split Google + LinkedIn. Monitor daily and
flag anything over 1.5x our $85 CPA target.")Or conversationally:
“Kick off the ad-campaigns skill for this week. Prioritize LinkedIn.”
Artifacts produced
ad-campaigns.xlsx— workbook withplanned,launched,metrics,anomalies,actionssheets.- Screenshots of each platform’s launch confirmation in
screenshots/ad-campaigns/. - Daily metric snapshots attached to the node’s activity timeline.
- Weekly synthesis reports with recommended next-week actions.
Full source
The authoritative workbook lives at cto-gui-libvirt-backend/app/services/agent_skills/ad-campaigns.md in the OpenFactory monorepo — the agent loads this directly into context when you invoke the skill.
Last updated on