keitaro-optimize

name	keitaro-optimize
description	Auto-optimize Keitaro campaigns: kill losing flows, scale winners, rebalance weights, suggest actions. Use when user says "optimize", "auto-optimize", "kill losers", "scale winners", "rebalance", "what should I pause", or "what should I scale".

Keitaro Auto-Optimization

Process

1. Read keitaro/references/optimization-rules.md for decision logic
2. Read keitaro/references/verticals.md for vertical-specific thresholds
3. Pull performance data via report API
4. Apply rules to identify actions
5. Present action plan with estimated impact
6. Execute only after user confirmation

Optimization Pipeline

Step 1: Collect Data

Build report grouped by stream_id with metrics: clicks, conversions, revenue, cost, profit, roi, cr, epc

Step 2: Classify Each Flow

For each flow, determine status based on vertical thresholds:

Status	Condition	Action
KILL	ROI < kill threshold AND clicks >= min	Recommend pause
WARNING	ROI in warning range	Monitor, flag
STABLE	ROI in good range	Keep running
SCALE	ROI > scale threshold AND clicks >= scale min	Recommend weight increase
TESTING	clicks < min threshold	Wait for data

Step 3: Generate Action Plan

Present as prioritized list:

Optimization Plan — Campaign #12
═══════════════════════════════════════

KILL (save $X/day):
  ✗ Flow "Test Landing C" — ROI: -45%, 180 clicks, 0 conv
    → Action: Pause flow (saves ~$15/day)

SCALE (earn +$X/day):
  ↑ Flow "DE Mobile" — ROI: +78%, 3,259 clicks, 62 conv
    → Action: Increase weight 60% → 70% (est. +$12/day)

TESTING (wait):
  ? Flow "New Offer X" — ROI: +15%, 45 clicks, 2 conv
    → Action: Need 55 more clicks before decision

TOTAL ESTIMATED IMPACT: +$27/day

Step 4: Execute (After Confirmation)

Only execute after user says "yes", "do it", "apply", etc.

Actions:

• Kill → POST /streams/{id}/disable
• Scale → PUT /streams/{id} with new weight
• Rebalance → adjust all weights proportionally

Bulk Optimization

When user says "optimize all campaigns":

1. Get all active campaigns
2. Build report grouped by campaign_id
3. For each campaign with sufficient data:
   • Run flow-level optimization
   • Aggregate actions
4. Present summary:

Bulk Optimization Summary
═════════════════════════
Campaigns analyzed: 8
Flows to kill: 5 (saving ~$85/day)
Flows to scale: 3 (est. +$45/day)
Flows still testing: 12

Net estimated impact: +$130/day

Safety Rules

• ALWAYS run with --dry-run first — show plan to user, get confirmation, then execute
• Never auto-execute without explicit user confirmation ("yes", "do it", "apply")
• Never kill all flows in a campaign (API helper blocks this automatically)
• Never set a single flow above 80% weight
• Don't optimize campaigns with < 24h of data
• Don't optimize during known traffic spikes (user must confirm)
• Show "undo" instructions after every action
• Snapshots are automatic — every stream/campaign is saved before modification
• For bulk operations: execute one at a time, verify result before next
• If ANY API call fails during bulk optimization: STOP immediately, report what was done

Undo Instructions

After each optimization action, show how to reverse:

Applied: Paused flow #45 in campaign #12
Undo: Run `/keitaro flows` → "enable flow 45 in campaign 12"