kai-daily-ad-review

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

Objective

A scannable daily pulse on live ad performance: today's numbers against the 7-day trend, every problem flagged with its severity, and a short list of specific actions naming the campaign or ad they apply to. Every figure traceable to the morning's pull.

This is not /kai-ad-campaign, which builds and evaluates campaigns end to end. This is the fast check that decides whether a deeper audit is needed today.

Done when

Work type audit-report (also_covers: weekly-audit, monthly-audit — the daily-cadence member of the same family) — floor E3/C4/O1 (harness/eco-floors.yaml).

• E3 — the review at workspace/ads/daily-reviews/[YYYY-MM-DD]-daily-review.md is the version a named human read and approved before any action is taken from it.
• C4 — every number in the review resolves to today's pull artifact under workspace/ads/pulls/YYYY-MM-DD/. No metric appears that is not in the JSON. Platforms without credentials are reported as not pulled, never estimated. banned_word_check passes.
• O1 — each action item names the metric it should move, that metric's current value from the pull, a threshold, and the date it gets re-read.

Constraints

Pull before analyzing. python scripts/ads/pull_all.py auto-detects which platforms have credentials and writes meta.json, google.json, linkedin.json, and summary.json to workspace/ads/pulls/YYYY-MM-DD/. Scope to one platform with --platforms meta. When a platform is skipped, the script logs which env vars are missing — report that, do not fill the gap. At minimum Meta should be configured (META_ACCESS_TOKEN, META_AD_ACCOUNT_ID numeric without the act_ prefix).

Never write a metric from memory, from yesterday's review, or from an estimate. If it is not in today's JSON, it is not in the review.

Mutations require approval. scripts/ads/meta.py, scripts/ads/google.py, and scripts/ads/linkedin.py expose pause, activate, budget, add-negative, and creation commands. They are dry-run by default and need --execute to apply. Live ad-account mutation without recorded human approval is never SHIPPED, whatever the evidence shows — this skill recommends actions and does not take them. Acting on a recommendation crosses into paid-ad-campaign (E5/C4/O4, spend_authority: true).

The benchmark table below is a default, not this account's benchmark. Adjust to the vertical from MARKETING.md where it exists, and label any benchmark that has no sourced vertical figure as a default rather than a target.

Escalation on recurrence. When the same flag appears 3+ days running, stop re-reporting it and recommend /kai-ad-campaign in evaluation mode for a deeper audit.

Cross-reference PostHog when connected using harness/references/posthog-marketing-queries.md — today's ad traffic with UTM breakdown (query #2), conversion events with campaign attribution (query #8), and bounce on pages receiving ad traffic. This connects spend to on-site behavior; when PostHog is not connected, say so.

Compare against history. Diff today's pull against yesterday's under workspace/ads/pulls/ for precise trends, and against prior files in workspace/ads/daily-reviews/ for week-over-week spend, recurring flags, and whether yesterday's action items were addressed.

Context

Need	Load
Unified pull across configured platforms	scripts/ads/pull_all.py
Meta pull + mutation commands	scripts/ads/meta.py
Google pull + mutation commands	scripts/ads/google.py
LinkedIn pull + mutation commands	scripts/ads/linkedin.py
Ad traffic and conversion queries	harness/references/posthog-marketing-queries.md
Paid channel checklist for deeper review	knowledge/checklists/paid-acquisition-checklist.md
Vertical, goals, budget context	MARKETING.md (project root)

Where the signal lives in each pull:

Platform	Fields that carry the diagnosis
Meta	account_insights.last_7d (spend, impressions, reach, frequency, ctr, cpc) · campaigns[].insights_7d · adsets[].audience_type (LAL / custom / interest / advantage+ / broad) · adsets[].insights_daily for fatigue · ads[].insights_7d incl. quality_ranking (ABOVE_AVERAGE_35 / AVERAGE / BELOW_AVERAGE_35), engagement and conversion rate rankings · breakdowns.age_gender · breakdowns.platform_position (FB Feed vs IG Reels vs Stories)
Google	campaigns[].insights_daily for CPC/CPA trend · search_terms for wasted spend · keyword_quality (flag quality score < 5) · audience_segments
LinkedIn	campaigns[].insights_7d engagement rate (higher baseline than Meta/Google) · breakdowns.industry + breakdowns.job_function · campaigns[].audience_type (matched audience vs professional targeting)

Benchmarks (default grid — adjust to vertical):

Metric	Poor	OK	Good	Great
CTR	< 0.5%	0.5–1%	1–2%	> 2%
CPC	> $5	$3–5	$1.50–3	< $1.50
CPL	> $50	$30–50	$15–30	< $15
ROAS	< 1x	1–2x	2–4x	> 4x
Frequency	> 4.0	3.0–4.0	1.5–3.0	1.0–1.5

Trend detection — today vs the 7-day average: spend pace > 120% of daily average is overspend · CTR < 70% of the 7-day average is creative fatigue · CPC > 130% is a competition spike or audience saturation · conversions < 50% of the daily average means a broken funnel or tracking · frequency > 3.0 means fatigue is coming · any BELOW_AVERAGE_35 diagnostic dimension is a flag.

Issue detection:

Issue	Trigger	Severity
Overspend	Daily spend pace > 120% of budget	HIGH
Zero impressions	Active ad with 0 impressions today	HIGH
CTR crash	CTR < 50% of 7-day avg	HIGH
CPC spike	CPC > 150% of 7-day avg	MEDIUM
No conversions	Spend > $50 today with 0 conversions	MEDIUM
Creative fatigue	CTR declining 3+ consecutive days	MEDIUM
Frequency overload	Frequency > 3.5 on any ad set	MEDIUM
Quality warning	Any ad with BELOW_AVERAGE quality ranking	MEDIUM
Wasted search spend	Google search term with > $20 spend, 0 conversions	MEDIUM
Budget underspend	< 50% of daily budget used by midday	LOW
Learning phase	Ad set in learning phase > 7 days	LOW
LAL underperform	LAL CPL > broad/interest CPL	LOW

Audience reading (from audience_type tagging): lookalikes should show lower CPL than broad · custom retargeting audiences should show the highest CTR · note when Advantage+ beats manual targeting · flag LAL underperforming interest targeting for investigation.

Output goes to workspace/ads/daily-reviews/[YYYY-MM-DD]-daily-review.md, scannable in under a minute, carrying:

• Snapshot — today vs 7-day average vs trend direction for spend, impressions, reach, frequency, clicks, CTR, CPC, conversions, CPL.
• Flags — one line each, severity-prefixed, naming the campaign or ad set and the number that triggered it.
• Audience performance — ad sets, spend, leads, CPL, and CTR per audience type.
• Campaign breakdown — spend, clicks, CTR, CPC, conversions, CPL, status per campaign.
• Top performers and underperformers — each underperformer with what is wrong and the specific action (pause, adjust, replace creative).
• Quality diagnostics — only the non-ABOVE_AVERAGE entries.
• Action items — checkbox list, each naming the campaign or ad it applies to.

This skill is built to run every morning. Recommend /schedule for a recurring trigger.

Escalate when

• No platform is configured and the pull returns nothing — report the missing env vars; a review with no data is a failure record, not an empty report.
• The same flag has appeared 3+ days running.
• Spend is pacing far past budget and the fix requires a live mutation.
• Conversions dropped to zero across platforms at once — suspect tracking, not creative, and say so.
• The account's numbers contradict what MARKETING.md says the budget or goal is.
• A recommended action would change budget, targeting, or creative on a live account without a recorded human approval.