Module 2: Creative Performance Review Agent

Detect creative fatigue, identify winning angles and hooks, get systematic recommendations

What This Agent Does

This agent will:

Analyze creative performance data across campaigns
Detect fatigue patterns in angles, hooks, and visuals
Identify which creative elements are winning
Suggest new directions based on gaps in current creative
Prioritize creative testing roadmap

This removes guesswork from creative iteration.

Prerequisites

Completed Module 1 (Weekly Performance Review Agent)
Folder structure in place (/inputs, /workflows, /outputs)
Claude Code operational

Step 1: Define Creative Review Standard

Create: `workflows/creative_review_standard.md`

Objective: Identify creative fatigue and winning patterns before performance drops.

Fatigue Detection Thresholds

CTR decline of 20%+ over 7 days
Impression frequency above 3.5
CPA increase while volume stays flat

Angle Taxonomy (Customize Yours)

Pain-first
Outcome-first
Social proof
Authority/expert
Contrarian

Hook Patterns

Question hooks
Stat hooks
Story hooks
Challenge hooks

What to Track

CTR by creative
Conversion rate by creative
Frequency
Days live

Step 2: Prepare Creative Inputs

Export and place in `/inputs`:

bash

creative_performance_current.csv (CTR, impressions, conversions by creative)

bash

creative_metadata.csv (angles, hooks, visual themes)

Include creative ID, angle, hook, and visual type.

Step 3: Create Creative Analysis Workflow

Inputs Required

creative_performance_current.csv
creative_metadata.csv
creative_review_standard.md

Analysis Steps

Group creatives by angle and hook
Calculate fatigue score per creative
Identify top performers by angle/hook combo
Flag underutilized angles
Suggest 3 new creative directions

Rules

Do not suggest more creative without data justification
Prioritize angles with proven CTR but low volume
Flag over-rotation on single angle

Output Format

Fatigued creatives list
Winning patterns
Suggested new tests (ranked)

Step 4: Run the Creative Agent

bash

claude "Execute workflows/creative_review_workflow.md using latest creative data."

Step 5: Save Output

bash

outputs/creative_review_YYYY_MM_DD.md

Step 6: Human Override

Remove suggestions that don't align with brand
Add creative constraints Claude doesn't know
Prioritize based on production capacity

Step 7: Iterate Monthly

Review accuracy of fatigue predictions
Refine angle taxonomy based on what works
Update thresholds

Common Failure Modes

Suggestions feel generic

Cause: Angle taxonomy too broad

Fix: Get more specific with angle definitions

Missing emerging patterns

Cause: Not tracking enough creative metadata

Fix: Add visual theme and copy structure tags

Over-rotation warnings ignored

Cause: Output not actionable

Fix: Add "what to pause" recommendations

Why This Works

Fatigue is measurable, not guessed
Winning patterns are extracted from data
Creative testing follows gaps, not hunches
Brand voice stays intact via human override

Creative iteration becomes systematic, not reactive.

← Previous: Module 1: Weekly Performance Review Agent Next: Module 3: Budget Reallocation Agent →