Workflow - Meta Ads Kit

Meta Ads Kit orchestrates a closed-loop workflow that replaces your entire daily ad management routine. From monitoring performance to detecting fatigue to generating fresh copy to uploading new ads — all without opening Ads Manager.

The system is read-only by default. All reporting runs automatically. Any action that affects spend (pause/resume/budget changes/ad uploads) requires your explicit approval.

The Full Loop

Here’s how the five skills chain together:

Monitor (meta-ads)
    ↓
Detect fatigue (ad-creative-monitor)  
    ↓
Shift budget (budget-optimizer)
    ↓
Generate new copy (ad-copy-generator)
    ↓
Upload to Meta (ad-upload)
    ↓
Monitor again (meta-ads)

No Ads Manager required at any step.

Morning Briefing (Automated)

This is the daily routine that replaces 20 minutes of Ads Manager clicking.

How It Works

Scheduled: Every morning at 8am (configurable) What happens:

Cron triggers daily check at configured time
Agent pulls insights via the meta-ads skill
Analyzes performance:
- Spend pacing (on track? over/under?)
- Active campaigns (anything that shouldn’t be running?)
- 7-day trends (improving or declining?)
Identifies issues:
- Bleeders 🩸 — CTR < 1% or frequency > 3.5 + high spend
- Winners 🏆 — Top CTR, low CPC, scaling headroom
- Fatigue signals 😴 — CTR dropping day-over-day
Generates summary with recommendations
Delivers to you via Telegram/Slack/configured channel
Waits for approval on any actions

Example Briefing

☀️ Meta Ads Briefing — March 4, 2026

💰 SPEND PACING
🟡 18% above target
Current: $287 (as of 8:00am)
Expected: $243  
Projected EOD: $524 (budget: $450)

📊 ACTIVE CAMPAIGNS (4)
✅ Summer Launch - Traffic ($150/day)
✅ Retargeting - Conversions ($200/day)
⚠️ Q1 Test Campaign ($50/day) — scheduled to end 3/1, still running
✅ Brand Awareness ($75/day)

🏆 WINNERS (Ready to scale)
1. Ad #23847111 "Notes App Hero"
   CTR: 3.2% | CPC: $0.87 | Spend: $124
   
2. Ad #23847112 "Kitchen Before/After"
   CTR: 2.9% | CPC: $0.95 | Spend: $98

🩸 BLEEDERS (Pause candidates)  
1. Ad #23847120 "Generic Product Shot"
   CTR: 0.6% | CPC: $3.20 | Spend: $187
   Frequency: 4.2 — audience exhausted

😴 FATIGUE DETECTED (2 ads)
Ad #23847111 "Notes App Hero"
CTR trend (7d): 3.2% → 2.2% (31% decline)
Frequency: 3.9 🟡
→ Needs creative refresh this week

📋 RECOMMENDATIONS
1. Pause Q1 Test Campaign (was scheduled to end 3/1)
2. Pause Ad #23847120 "Generic Product Shot" (bleeding $187 at 0.6% CTR)
3. Refresh copy for Ad #23847111 (fatigue detected)
4. Scale Ad #23847112 (+20% budget if performance holds)

Reply to approve, or ask questions.

You (2 minutes over coffee):

Read the summary
Reply: “Approve 1, 2, and 3. Hold on 4 for now.”
Done.

The agent executes approved actions and logs them.

Setting Up Automated Briefings

Via OpenClaw

openclaw start
# Then:
"Run my daily ads check every morning at 8am and send me the summary"

The agent sets up the cron job automatically.

Via Cron (Manual)

Add to your crontab:

# Daily at 8am
0 8 * * * cd /path/to/meta-ads-kit && ./skills/meta-ads/scripts/meta-ads.sh daily-check | mail -s "Meta Ads Briefing" [email protected]

Or pipe to Slack/Telegram:

0 8 * * * cd /path/to/meta-ads-kit && ./skills/meta-ads/scripts/meta-ads.sh daily-check | curl -X POST -H 'Content-Type: application/json' -d '{"text": "'$(cat)'"}' YOUR_SLACK_WEBHOOK_URL

On-Demand Workflow (Interactive)

You can also run the workflow interactively throughout the day.

1. Monitoring

You ask:

"How are my ads doing?"
"Any bleeders I should pause?"
"Check for creative fatigue"

Agent does:

Runs appropriate report via meta-ads skill
Pulls data via social-cli
Interprets results in context of benchmarks from ad-config.json
Presents findings with actionable recommendations

2. Budget Optimization

You ask:

"Where's my money working best?"
"How should I shift my budget?"

Agent does:

Runs efficiency analysis via budget-optimizer skill
Ranks campaigns/adsets by CPA, ROAS, or CPC
Identifies top and bottom performers
Calculates recommended budget shifts
Presents recommendations with reasoning
Waits for approval before adjusting any budgets

Example:

Efficiency Analysis (Last 7 Days):

🏆 Best Performer:
Retargeting - Conversions  
Spend: $1,400 | CPA: $18.50 (target: $25)
→ Recommend +30% budget ($1,820/week)

🩸 Underperformer:
Q1 Test Campaign
Spend: $350 | CPA: $42.00 (target: $25)
→ Recommend -50% budget ($175/week) or pause

Recommended Shift:
Move $175/week from Q1 Test → Retargeting
Result: Estimated 8 more conversions/week at target CPA

Approve?

You: “Approve” Agent: Adjusts budgets, logs the change.

3. Creative Refresh (When Fatigue Hits)

You ask:

"Ad #23847111 is fatigued. Refresh the copy."

Agent does:

Analyzes current ad (via meta-ads)
- Pulls creative details from Graph API
- Confirms fatigue signal (CTR decline, frequency spike)
Generates fresh copy (via ad-copy-generator)
- Analyzes the image creative
- Pulls account’s top-performing copy patterns from last 30 days
- Writes 3-5 new headline + body variants
- Outputs asset_feed_spec JSON ready for upload
Shows you the copy
- Presents variants with psychology labels (Money, Status, Fear, Time)
- Explains how copy reinforces the image
Waits for approval

You: “Looks good. Upload it.” Agent does:

Uploads to Meta (via ad-upload)
- Creates new creative with fresh copy
- Attaches to existing ad ID (preserves metric history)
- Sets status to PAUSED for final review
Confirms completion
- Returns ad ID and Ads Manager review URL

Example output:

✓ Copy generated (3 headlines, 3 bodies)
✓ Creative created → ID: 23847293847293999
✓ Ad #23847111 updated with fresh creative (PAUSED)

Review at:
https://www.facebook.com/adsmanager/manage/ads?act=123456789

Activate when ready.

You: Review in Ads Manager, activate when satisfied.

4. Launching New Ads

You ask:

"Write copy for this image" (attach creative)

Agent does:

Analyzes the image (via ad-copy-generator)
- Identifies format (notes app, receipt, tweet, chart, etc.)
- Extracts on-image text
- Determines psychology and funnel stage
Pulls winning patterns from your account
Writes matched copy (3-5 headlines, 3-5 bodies)
Outputs asset_feed_spec JSON

You: “Upload it to ad set 23847000000001” Agent does:

Uploads (via ad-upload)
- Validates copy + image
- Uploads image → gets hash
- Creates creative
- Creates ad in specified ad set (PAUSED)
Returns ad ID

You: Review and activate.

Approval Gates

The system has two operational modes:

Read-Only (No Approval Needed)

These actions run automatically:

Running reports and insights
Analyzing data
Generating recommendations
Logging learnings
Pulling account performance data

Write Actions (Approval Required)

These actions always ask first:

Pausing any ad, adset, or campaign
Resuming any ad, adset, or campaign
Changing any budget
Uploading new ads or creatives
Updating existing ad copy

The agent will:

Present the action with full context
Show the data justification
Explain the expected outcome
Ask: “Proceed?”
Wait for explicit “yes” / “approve” / confirmation
Execute on approval
Log the action with timestamp and reasoning

Example:

Recommendation: Pause Ad #23847120 "Generic Product Shot"

Data:
- CTR: 0.6% (target: 1.0%)
- CPC: $3.20 (target: $2.50)
- Spend: $187 (last 7 days)
- Frequency: 4.2 (threshold: 3.5)

Reasoning: High spend + low CTR + audience exhausted = bleeding budget

Estimated savings: ~$27/day if reallocated to winners

Proceed?

You: “Yes” or “No” or “Not yet, let it run another day” If “Yes”, agent pauses the ad and logs:

workspace/brand/learnings.md:

2026-03-04 08:15 — Paused Ad #23847120 "Generic Product Shot"
Reason: CTR 0.6%, frequency 4.2, $187 spend in 7 days
Approved by: user (via morning briefing)
Outcome: Saved $187 budget over 7 days by stopping underperformer

Memory & Learning

The system builds institutional knowledge over time.

What Gets Logged

Every day, the agent logs to memory/YYYY-MM-DD.md:

Reports run and key findings
Actions taken (paused/resumed/budget changes/uploads)
Performance trends noted
Recommendations made and outcomes

What Gets Learned

Findings append to workspace/brand/learnings.md:

Creative fatigue patterns (how long do your creatives last?)
Winning copy traits (what hooks convert?)
Audience performance (which demos respond best?)
Budget efficiency (which campaigns deliver best ROI?)

Example learnings.md:

# Meta Ads Learnings

## Creative Lifespan
- Notes app format: ~14 days before fatigue (avg CTR decay 2% per day after day 10)
- Before/after images: ~21 days (more durable)
- Generic product shots: ~7 days (fatigue fast)

## Copy Patterns That Win
- Headlines with specific numbers outperform questions by 18% CTR
- Social proof ("X+ customers") in body increases conversions 12%
- Opening with pain point > opening with benefit (22% higher CTR)

## Audience Insights  
- Women 25-34 respond best to Status psychology (1.8% CTR avg)
- Men 35-44 respond best to Money psychology (2.1% CTR avg)
- Frequency sweet spot: 2.5-3.0 (CTR peaks, then declines sharply)

## Budget Efficiency
- Retargeting campaigns consistently deliver 40% lower CPA than cold traffic
- Scaling winners works best at +20-30% increments (larger jumps kill performance)

Using Learnings

The agent references learnings.md when:

Generating new copy (“Your top performers use question headlines — I’ll match that”)
Detecting fatigue (“Notes app creatives typically last 14 days. This one is on day 12.”)
Recommending budget shifts (“Retargeting has delivered 40% lower CPA historically — good scaling candidate”)

Over time, recommendations get smarter and more brand-specific.

Example: Full Day in the Loop

8:00am — Morning Briefing Arrives

Your phone buzzes. Telegram message from Meta Ads Copilot:

☀️ Meta Ads Briefing — March 4, 2026

💰 Spend on track ($243 / $450 budget)
🏆 2 winners ready to scale
🩸 1 bleeder burning $27/day
😴 1 ad showing fatigue (refresh needed)

Recommendations:
1. Pause Ad #23847120 (0.6% CTR, bleeding budget)
2. Refresh copy for Ad #23847111 (CTR down 31% over 7 days)
3. Scale Ad #23847112 (+20% budget)

Approve all, some, or reply for details.

You’re at the coffee shop. You reply: “Approve 1 and 2. Hold on 3.”

8:02am — Agent Executes

Pauses Ad #23847120
Starts creative refresh workflow for Ad #23847111:
- Analyzes the image
- Pulls top-performing copy from your account
- Generates 3 new headline + body variants
- Sends you the copy for review

8:05am — Copy Review

You get another message:

✍️ Fresh copy for Ad #23847111 "Notes App Hero"

Image: iPhone notes showing "$17K/month"
Format: Notes app / organic
Psychology: Money + Status

3 new headlines:
1. "We hit $17K/mo in 90 days" (29 chars)
2. "This wasn't supposed to work" (30 chars)  
3. "3 months. $17K. Here's how." (28 chars)

3 new body variants (Money, Status, Fear psychology)

Review: [link to full copy doc]

Upload to Meta?

You click the link, scan the copy. Looks good. You reply: “Upload it.”

8:07am — Upload Complete

✓ Creative created → ID: 23847293847293999
✓ Ad #23847111 updated with fresh copy (PAUSED)

Review at: [Ads Manager link]

Activate when ready.

You make a note to review in Ads Manager later.

11:30am — On-Demand Check

Between meetings, you wonder how the morning changes are performing. You message: “How’s Ad #23847111 doing with the new copy?” Agent:

Ad #23847111 "Notes App Hero" (refreshed copy)
Status: PAUSED (uploaded 8:07am, not yet activated)

Activate to start collecting data.

You: “Activate it.” Agent:

✓ Ad #23847111 activated

I'll check performance in 24 hours and compare to pre-refresh metrics.

5:00pm — End of Day

You message: “Daily summary?” Agent:

📊 End of Day — March 4, 2026

Total spend: $448 (budget: $450) ✅

Actions taken today:
- Paused Ad #23847120 (saved est. $27)
- Refreshed + activated Ad #23847111 (new copy live as of 11:31am)

Early signals:
Ad #23847111 (new copy): 2.8% CTR in first 5 hours (92 impressions)
Previous 7-day avg: 2.2% CTR
→ Trending positive, check again tomorrow

No new issues detected.

You’re done. Total time spent managing ads today: ~5 minutes.

Customizing the Workflow

Adjust Benchmarks

Edit ad-config.json to change what triggers alerts:

{
  "benchmarks": {
    "target_cpa": 25.00,
    "target_roas": 3.0,
    "max_frequency": 3.5,
    "min_ctr": 1.0,
    "max_cpc": 2.50
  },
  "alerts": {
    "bleeder_ctr_threshold": 1.0,
    "bleeder_frequency_threshold": 3.5,
    "fatigue_ctr_drop_pct": 20,
    "spend_pace_alert_pct": 15
  }
}

Change Briefing Time

Tell the agent:

"Change daily briefing to 7am"
"Send briefing at 9am on weekdays only"

Or edit cron directly.

Adjust Approval Requirements

By default, all write actions require approval. If you want the agent to auto-execute certain actions:

"Auto-pause any ad with CTR < 0.5% and spend > $50"
"Auto-approve budget shifts under $50"

The agent updates its rules and logs the policy change.

Use auto-execution carefully. Start with full approval mode until you trust the system’s judgment.

What You Never Touch

With this workflow fully operational:

Ads Manager — Only for final creative review before activating new ads
Manual reporting — All insights delivered automatically
Budget micromanagement — Agent recommends, you approve in seconds
Creative performance tracking — Fatigue detected before you’d notice manually
Copy generation — AI writes matched to your image creatives and account data
Uploading ads — Push straight to Meta via API

Your job becomes strategic decision-making, not tactical clicking.

Next Steps

Daily Questions

Deep dive on the 5 questions that power the briefing

Skills Overview

Learn what each skill does and when to use it

Setup Guide

Get your ad account connected and run your first briefing

Configuration

Customize benchmarks, thresholds, and automation rules

Get Started

Core Concepts

Skills

Configuration

Reports

Actions

Advanced

​The Full Loop

​Morning Briefing (Automated)

​How It Works

​Example Briefing

​You (2 minutes over coffee):

​Setting Up Automated Briefings

​Via OpenClaw

​Via Cron (Manual)

​On-Demand Workflow (Interactive)

​1. Monitoring

​2. Budget Optimization

​3. Creative Refresh (When Fatigue Hits)

​4. Launching New Ads

​Approval Gates

​Read-Only (No Approval Needed)

​Write Actions (Approval Required)

​Memory & Learning

​What Gets Logged

​What Gets Learned

​Using Learnings

​Example: Full Day in the Loop

​8:00am — Morning Briefing Arrives

​8:02am — Agent Executes

​8:05am — Copy Review

​8:07am — Upload Complete

​11:30am — On-Demand Check

​5:00pm — End of Day

​Customizing the Workflow

​Adjust Benchmarks

​Change Briefing Time

​Adjust Approval Requirements

​What You Never Touch

​Next Steps

Daily Questions

Skills Overview

Setup Guide

Configuration

Build docs developers (and LLMs) love

The Full Loop

Morning Briefing (Automated)

How It Works

Example Briefing

You (2 minutes over coffee):

Setting Up Automated Briefings

Via OpenClaw

Via Cron (Manual)

On-Demand Workflow (Interactive)

1. Monitoring

2. Budget Optimization

3. Creative Refresh (When Fatigue Hits)

4. Launching New Ads

Approval Gates

Read-Only (No Approval Needed)

Write Actions (Approval Required)

Memory & Learning

What Gets Logged

What Gets Learned

Using Learnings

Example: Full Day in the Loop

8:00am — Morning Briefing Arrives

8:02am — Agent Executes

8:05am — Copy Review

8:07am — Upload Complete

11:30am — On-Demand Check

5:00pm — End of Day

Customizing the Workflow

Adjust Benchmarks

Change Briefing Time

Adjust Approval Requirements

What You Never Touch

Next Steps