Pilot Mode + GA4 + Platform-tagged Recommendations #2

New Issue

2026-05-10T09:05:31Z

colorlink commented

2026-05-10 09:05:31 +00:00

Manar — Pilot Mode & Advanced Features

Phase 1 — Platform tags on recommendations ✅ In progress

Each recommendation card shows platform badge (Google Ads, Meta, etc.)
Campaign name clearly visible
Filter bar: All | Google Ads | Meta | TikTok | etc.

Phase 2 — Pilot Mode settings

Per-brand permission matrix stored in brand_context:

{
  "pilot_mode": {
    "enabled": false,
    "trust_level": "L1",
    "permissions": {
      "can_pause_campaigns": true,
      "can_enable_campaigns": true,
      "can_adjust_bids": true,
      "bid_change_max_pct": 20,
      "can_add_negatives": true,
      "can_add_keywords": false,
      "can_change_budget": false,
      "budget_increase_requires_approval": true,
      "can_create_ads": false,
      "can_delete_anything": false
    }
  }
}

UI: Toggle to enable Pilot Mode + permission matrix checkboxes

Phase 3 — Execution engine

When recommendation is approved (or auto-approved in L2/L3):
- Route to platform-specific executor
- Google Ads: use existing OAuth token + gRPC API
- Log every action to audit_log with before/after values
- Send notification to admin after execution
Execution functions:
- pause_campaign(customer_id, campaign_id)
- set_campaign_budget(customer_id, campaign_id, new_budget)
- set_target_cpa(customer_id, campaign_id, new_cpa)
- add_negative_keyword(customer_id, campaign_id, keyword)
- adjust_bidding_strategy(customer_id, campaign_id, strategy)

Phase 4 — Google Analytics 4

OAuth flow same pattern as Google Ads (same Google account, different scope)
Scopes: analytics.readonly
Fetch: sessions, bounce rate, avg session duration, goal completions, top landing pages
Inject into AI prompt alongside ad data
AI can then correlate: "Clicks are high but bounce rate is 90% — landing page issue, not bid issue"

Phase 5 — Multi-platform execution

Meta: Marketing API — pause/enable ad sets, adjust budgets
TikTok: same pattern
Each platform needs its own executor class

Trust levels (already in DB)

L0: Monitor only
L1: Suggest + wait for approval
L2: Execute with veto window (24h to cancel)
L3: Full autopilot — execute silently, audit log only

# Manar — Pilot Mode & Advanced Features ## Phase 1 — Platform tags on recommendations ✅ In progress - Each recommendation card shows platform badge (Google Ads, Meta, etc.) - Campaign name clearly visible - Filter bar: All | Google Ads | Meta | TikTok | etc. ## Phase 2 — Pilot Mode settings Per-brand permission matrix stored in `brand_context`: ```json { "pilot_mode": { "enabled": false, "trust_level": "L1", "permissions": { "can_pause_campaigns": true, "can_enable_campaigns": true, "can_adjust_bids": true, "bid_change_max_pct": 20, "can_add_negatives": true, "can_add_keywords": false, "can_change_budget": false, "budget_increase_requires_approval": true, "can_create_ads": false, "can_delete_anything": false } } } ``` UI: Toggle to enable Pilot Mode + permission matrix checkboxes ## Phase 3 — Execution engine - When recommendation is approved (or auto-approved in L2/L3): - Route to platform-specific executor - Google Ads: use existing OAuth token + gRPC API - Log every action to `audit_log` with before/after values - Send notification to admin after execution - Execution functions: - `pause_campaign(customer_id, campaign_id)` - `set_campaign_budget(customer_id, campaign_id, new_budget)` - `set_target_cpa(customer_id, campaign_id, new_cpa)` - `add_negative_keyword(customer_id, campaign_id, keyword)` - `adjust_bidding_strategy(customer_id, campaign_id, strategy)` ## Phase 4 — Google Analytics 4 - OAuth flow same pattern as Google Ads (same Google account, different scope) - Scopes: `analytics.readonly` - Fetch: sessions, bounce rate, avg session duration, goal completions, top landing pages - Inject into AI prompt alongside ad data - AI can then correlate: "Clicks are high but bounce rate is 90% — landing page issue, not bid issue" ## Phase 5 — Multi-platform execution - Meta: Marketing API — pause/enable ad sets, adjust budgets - TikTok: same pattern - Each platform needs its own executor class ## Trust levels (already in DB) - L0: Monitor only - L1: Suggest + wait for approval - L2: Execute with veto window (24h to cancel) - L3: Full autopilot — execute silently, audit log only

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: colorlink/manar#2