Kai Daily Ad Review

Daily ad performance check-in. Pull live data via scripts, compare against benchmarks, flag problems, output a summary with action items.

This is NOT the same as /kai-ad-campaign (which creates/evaluates campaigns end-to-end). This is a fast daily pulse check — meant to run every morning or on-demand.

Phase 0: Pull Ad Data

Run the unified pull script. It auto-detects which platforms have credentials and pulls everything.

cd E:/Dev2/kai-cmo-harness-work && python scripts/ads/pull_all.py

This writes structured JSON to workspace/ads/pulls/YYYY-MM-DD/:

• meta.json — Full Meta/Facebook/Instagram data (if META_ACCESS_TOKEN configured)
• google.json — Full Google Ads data (if GOOGLE_ADS_* configured)
• linkedin.json — Full LinkedIn Ads data (if LINKEDINADS_* configured)
• summary.json — Cross-platform totals

To pull a specific platform only: python scripts/ads/pull_all.py --platforms meta

What the scripts pull (comprehensive)

Meta (scripts/ads/meta.py pull):

• Account insights: today, yesterday, 7d, 14d, 28d
• Campaign insights: 7d aggregate + 14d daily breakdown
• Ad set insights: 7d aggregate + 14d daily + targeting spec with LAL/audience classification
• Ad insights: 7d with creative details
• Breakdowns: age/gender, platform/position, device
• Fields include: impressions, reach, frequency, spend, clicks, unique_clicks, ctr, unique_ctr, cpc, cpm, actions, cost_per_action_type, conversions, quality_ranking, engagement_rate_ranking, conversion_rate_ranking, video_p25/50/75/100, video_thruplay
• Mutations: meta.py pause/activate/budget/create-campaign/create-adset/create-ad/upload-image/upload-video/duplicate-adset (dry-run by default, --execute to apply)

Google (scripts/ads/google.py pull):

• Campaign performance: 30d daily breakdown
• Ad group performance: 14d daily
• Ad performance: 14d aggregate with RSA headline/description details
• Search terms report: top 100 by spend (7d)
• Audience segments: campaign-level
• Keyword quality scores: quality_score, creative_quality, post_click_quality, predicted_ctr
• Mutations: google.py pause/activate/budget/add-negative (dry-run by default, --execute to apply)

LinkedIn (scripts/ads/linkedin.py pull):

• Campaign group and campaign insights: 7d, 14d daily, 28d
• Creative-level insights: 7d
• Audience breakdowns: company size, industry, job function, seniority, country
• Audience classification from targeting criteria
• Mutations: linkedin.py pause/activate/budget (dry-run by default, --execute to apply)

If a platform isn't configured

The script will skip it and log which env vars are missing. At minimum, Meta should be configured:

META_ACCESS_TOKEN=<long-lived token>
META_AD_ACCOUNT_ID=<numeric, without act_ prefix>

Phase 1: Read and Analyze Pull Data

Read the JSON files from today's pull:

import json
from pathlib import Path

date = "YYYY-MM-DD"  # today's date
pull_dir = Path(f"workspace/ads/pulls/{date}")

# Read whatever's available
for f in pull_dir.glob("*.json"):
    with open(f) as fh:
        data = json.load(fh)
    # Analyze...

Key metrics to extract per platform

From Meta (meta.json):

• account_insights.last_7d → spend, impressions, reach, frequency, ctr, cpc
• campaigns[].insights_7d → per-campaign performance
• adsets[].audience_type → LAL vs custom vs interest vs advantage+ vs broad
• adsets[].insights_daily → day-by-day trends for fatigue detection
• ads[].insights_7d → per-ad creative performance
• ads[].insights_7d.quality_ranking → Meta's diagnostic: ABOVE_AVERAGE_35, AVERAGE, BELOW_AVERAGE_35
• breakdowns.age_gender → demographic performance
• breakdowns.platform_position → FB Feed vs IG Reels vs Stories etc.

From Google (google.json):

• campaigns[].insights_daily → day-by-day CPC/CPA trends
• search_terms → wasted spend on irrelevant queries
• keyword_quality → quality score distribution (flag anything < 5)
• audience_segments → which audiences are converting

From LinkedIn (linkedin.json):

• campaigns[].insights_7d → engagement rate (higher baseline than Meta/Google)
• breakdowns.industry + breakdowns.job_function → who's engaging
• campaigns[].audience_type → matched audience vs professional targeting

Phase 2: Cross-Reference with PostHog

Load: harness/references/posthog-marketing-queries.md

If PostHog is connected, pull:

• Today's ad traffic — pageviews with UTM breakdown (query #2)
• Conversion events — campaign attribution (query #8)
• Landing page bounce — for pages receiving ad traffic

This connects ad spend to actual on-site behavior.

Phase 3: Benchmark Comparison

Performance Benchmarks

Adjust benchmarks to the vertical (from MARKETING.md if available).

Trend Detection

Compare today vs 7-day average:

• Spend: flag if today's pace > 120% of daily average (overspend)
• CTR: flag if today < 70% of 7-day avg (creative fatigue)
• CPC: flag if today > 130% of 7-day avg (competition spike or audience saturation)
• Conversions: flag if today < 50% of daily avg (broken funnel or tracking issue)
• Frequency: flag if > 3.0 (audience seeing ads too often — fatigue incoming)
• Quality ranking: flag if BELOW_AVERAGE_35 on any diagnostic dimension

Audience Performance (from LAL tagging)

Compare ad set performance by audience_type:

• LAL audiences should have lower CPL than broad
• Custom retarget audiences should have highest CTR
• If Advantage+ is outperforming manual targeting, note it
• If LAL is underperforming interest targeting, flag for investigation

Phase 4: Issue Detection

Flag these automatically:

Phase 5: Daily Summary Output

Output format — keep it scannable:

# Daily Ad Review — [Date]

## Snapshot
| Metric | Today | 7-Day Avg | Trend |
|--------|-------|-----------|-------|
| Spend | $X | $X/day | up/down/flat |
| Impressions | X | X/day | |
| Reach | X | X/day | |
| Frequency | X.X | X.X | |
| Clicks | X | X/day | |
| CTR | X% | X% | |
| CPC | $X | $X | |
| Conversions | X | X/day | |
| CPL | $X | $X | |

## Flags
- [HIGH] Overspend: Campaign "X" pacing 140% of daily budget
- [MEDIUM] Frequency: Ad set "Y" at 3.8 — rotate creative or expand audience
- [MEDIUM] Quality: Ad "Z" has BELOW_AVERAGE engagement ranking

## Audience Performance
| Audience Type | Ad Sets | Spend | Leads | CPL | CTR |
|---------------|---------|-------|-------|-----|-----|
| Lookalike 1% | 2 | $X | X | $X | X% |
| Custom Retarget | 1 | $X | X | $X | X% |
| Interest | 1 | $X | X | $X | X% |
| Advantage+ | 1 | $X | X | $X | X% |

## Campaign Breakdown
| Campaign | Spend | Clicks | CTR | CPC | Conversions | CPL | Status |
|----------|-------|--------|-----|-----|-------------|-----|--------|

## Top Performers
1. [Ad name] — [metric that makes it stand out]
2. [Ad name] — [metric]

## Underperformers
1. [Ad name] — [what's wrong] — **Action:** [pause/adjust/replace creative]
2. [Ad name] — [what's wrong] — **Action:** [specific fix]

## Quality Diagnostics
| Ad | Quality | Engagement | Conversion |
|----|---------|------------|------------|
(only show non-ABOVE_AVERAGE entries)

## Action Items
- [ ] [Specific action with campaign/ad name]
- [ ] [Specific action]
- [ ] [Specific action]

Write output to workspace/ads/daily-reviews/[YYYY-MM-DD]-daily-review.md.

Phase 6: Historical Tracking

If previous daily reviews exist in workspace/ads/daily-reviews/, compare:

• Week-over-week spend trend
• Which flags are recurring (persistent issues need /kai-ad-campaign evaluation)
• Which action items from yesterday were addressed

If previous pull data exists in workspace/ads/pulls/, diff today's metrics against yesterday's for precise trend detection.

If the same flag appears 3+ days in a row, escalate: recommend running /kai-ad-campaign in evaluation mode for a deeper audit.

Scheduling

This skill is designed to run daily. Recommend the user set up a schedule:

• /schedule to create a recurring morning trigger
• Or run manually with /kai-daily-ad-review