meta-ads-daily-action-plan

name

meta-ads-daily-action-plan

description

Daily action-first Meta Ads brief for Ben Mahmoud. Reads the meta-ads-daily-review JSON and the dashboard intelligence + Calendly bookings, reasons about what changed vs yesterday / 7d / 28d, and writes ONE Google Doc per day with three prioritised actions, pipeline state, per-ad-set verdicts (KILL/SCALE/WATCH), anomalies, and the exact Ads Manager changes Mahmoud should execute today. Tone is eight-figure consultant, not descriptive report.

Meta Ads Daily Action Plan

Action-first daily brief for Ben Mahmoud. Inherits from meta-ads-daily-review and meta-ads-weekly-intelligence — reads their data, layers reasoning on top, ships one Google Doc per day.

Purpose

The other two skills give analysis. This one gives decisions.

Ben Hawksworth's literal brief (2026-04-27):

"MUST be focused on actions to improve & scale."

Ben Mahmoud's literal brief:

"this is literally what just do straight away."

Output is what Mahmoud opens in the morning, copy-pastes into Ads Manager, and works through. Nothing else.

What This Skill Does

When invoked:

1. Pulls upstream skill data — runs meta-ads-daily-review if today's JSON isn't already in its outputs/ folder, otherwise reads the existing file.
2. Pulls dashboard intelligence — reads intelligence.json + meta_ads_campaigns.json + calendly_bookings.json from the cloned tcc-dashboard repo (/tmp/tcc-dashboard/public/data/).
3. Reads yesterday's action plan — to check what flags carried over and whether previous decisions were acted on.
4. Reasons about deltas — today vs yesterday vs 7d vs 28d. Looks for: CTR shifts, cost-per-real-call moves, audience saturation, learning-phase exits, fatigue, quality drops, spend redistribution.
5. Spawns sub-agents only when needed — if a single ad set's CTR drops by more than 25% day-over-day, dig into creative-level. If audience saturation suspected, pull historical comparison. If everything stable, don't pad.
6. Drafts the brief — exact structure below.
7. Writes a timestamped Google Doc to the Daily Action Plans folder.

Inputs (read in this order)

File	Purpose
1-meta-ads/meta-ads-daily-review/outputs/DAILY-REVIEW-*-{today}.json	Live Meta API + Calendly per BOF campaign run today
/tmp/tcc-dashboard/public/data/intelligence.json	Opus strategic verdicts, anomalies, weekly priorities
/tmp/tcc-dashboard/public/data/meta_ads_campaigns.json	7d + 28d ad-set breakdown for ALL campaigns, stage classification, summary.stage_spend_7d
/tmp/tcc-dashboard/public/data/calendly_bookings.json	Real paid bookings (ground truth)
/tmp/tcc-dashboard/public/data/sheet_revenue.json	Signups + revenue (literal name field is binding, see §Names)
1-meta-ads/meta-ads-daily-action-plan/outputs/{yesterday}.md	Yesterday's brief, for "did we act on it?" tracking

Coverage rule (multi-campaign)

The brief must cover spend across all stages, not just BOF. Single-campaign briefs hide whichever stage is currently consuming the most spend.

• BOF campaigns: run meta-ads-daily-review per active BOF campaign-id and read every DAILY-REVIEW-*-{today}.json written today. Live Meta + Calendly numbers come from these.
• TOF + MOF + UNKNOWN campaigns: read directly from meta_ads_campaigns.json.ad_sets[], filtered by stage field. No live Meta call needed for these stages, the cron-fresh JSON is sufficient.
• Sanity check first: read meta_ads_campaigns.json.summary.stage_spend_7d at the start. If the largest stage by spend has no presence in the brief, stop and surface the gap.

Clone source

/tmp/tcc-dashboard is cloned from antonio-gasso/tcc-dashboard directly using TCC_GITHUB_TOKEN. Same PAT works for both upstream and forks. Avoid SudhakaPr/tcc-dashboard (the fork) because it goes stale unless manually synced post-cron.

# First-time clone
git clone https://${TCC_GITHUB_TOKEN}@github.com/antonio-gasso/tcc-dashboard.git /tmp/tcc-dashboard

# Each run
git -C /tmp/tcc-dashboard pull

If today's daily-review JSON for a BOF campaign is missing, run bash 1-meta-ads/meta-ads-daily-review/setup.sh --campaign-id <ID> --week <N> --days 7 per active BOF campaign first.

Reasoning Framework

The skill is the brain. Decide what to investigate based on what the data shows. Frameworks below are non-negotiable — see 1-meta-ads/CLAUDE.md §5 for the full spec.

Step 1 — Classify every campaign first

Read stage field already computed in meta_ads_campaigns.json per ad set. Do not re-derive. If missing, map from objective:

• OUTCOME_AWARENESS, OUTCOME_VIDEO_VIEWS → TOF
• OUTCOME_TRAFFIC → MOF (TOF if optimised for profile visits)
• OUTCOME_ENGAGEMENT → MOF (TOF if Profile Visits / ThruPlay)
• OUTCOME_LEADS, OUTCOME_SALES → BOF

Step 2 — Judge each stage by its own metric

Stage	Primary	Never judge by
TOF	Cost per IG follower	CTR, form submits, cost per call
MOF	Cost per opt-in / DM conversation	Cost per booked call
BOF	Two layers below	Pixel lead events as calls, cost_per_form_submit as the kill anchor

BOF primary metric splits into two layers (Antonio 2026-05-02):

• Headline ROAS (account level): Calendly paid_ads_7d is still ground truth. Calendly aggregates by event-type slug, so it doesn't split per ad set, only per campaign / channel.
• Per-ad-set health: metrics_7d.cost_per_booking_pixel is the operational anchor. Pixel matches at the ad-set level so this is what surfaces which ad set is working and which isn't.
• cost_per_form_submit is now a volume signal only, not the kill anchor. Historical applications-to-bookings ratio is roughly 4.5:1, so a £10 cost per application can sit on top of a £45 cost per booking. The kill / scale call has to come from cost_per_booking_pixel.

Step 3 — Apply BOF kill/scale/watch thresholds (BOF ONLY, anchored on cost_per_booking_pixel_7d)

Calibrated against Mahmoud's £185 baseline (Antonio 2026-05-02).

• cost_per_booking_pixel_7d ≤ £150 with bookings_pixel_7d ≥ 2 → scale candidate
• cost_per_booking_pixel_7d ≤ £200 → healthy
• £200 < cost_per_booking_pixel_7d ≤ £300 → watch
• cost_per_booking_pixel_7d > £300 sustained over 7d → kill candidate
• cost_per_form_submit is reported but is NOT the kill anchor. Use it for application-volume diagnostics only. £3 cost per application can mask £349 cost per booking.
• CTR link < 0.8% → kill candidate (creative-level signal)
• CTR link ≥ 1.5% → healthy, ≥ 2.5% across ad sets → reinforce scale signal
• Min 1,000 impressions per ad set for a verdict, else "insufficient signal"
• Never kill during LEARNING / LEARNING_LIMITED. Min 50 conversion events.
• Budget split target: 50% BOF / 30% MOF / 20% TOF (use summary.stage_spend_7d)

Pixel match edge case: if bookings_pixel_7d == 0 AND applications_7d > 0, cross-reference Calendly paid_ads_7d. If Calendly shows real bookings, this is a pixel match-rate problem (ad blockers / cookie loss), surface as a "low pixel match rate" quality flag and DO NOT issue a kill verdict. The campaign is converting, the pixel just isn't seeing it.

Step 4 — Ground truth for ROAS

• Numerator: sheet_revenue.json signups in the last N days whose email matches a calendly_bookings.paid_ads_28d entry (email join, 28d window for sales-cycle lag).
• Denominator: BOF ad-set spend only from meta_ads_campaigns.json.
• Never compute ROAS from pixel events.
• If denominator > 0 and numerator = 0 → display "Pending", not "0.0x".

Trigger sub-agent when

• Single ad set CTR drops > 25% vs yesterday on > 1000 impressions.
• Cost per real Calendly call jumps > 50% week-over-week.
• Ad set in LEARNING_LIMITED for > 7 days.
• Frequency > 3.5 (TOF) or > 4.5 (BOF) suggests fatigue.
• One creative consumes > 40% of an ad set's spend.

Don't pad. If a section has nothing material, write one line: "No anomalies today."

Today's 3 Actions — sourcing rules

The Top-3 must rotate as the data rotates. Don't paraphrase yesterday's actions back in.

Source priority (highest first):

1. Carry-forward gaps from yesterday's brief. If yesterday's Action 1 was "reactivate the BOF campaign" and the campaign is still paused today, that becomes today's Action 1 again with a "still not actioned" framing. Open loops always come first.
2. Today's anomalies that breach a framework threshold. Cost per real call jumps over £75, CTR drops below 0.8%, frequency past fatigue, single-creative dominance, learning-phase stuck. Daily-fresh.
3. intelligence.meta_ads.weekly_priorities (Opus strategic anchor, refreshes Monday). Use only after carry-forward and daily anomalies are exhausted, and only when the priority is still current (the underlying state hasn't changed since the cron).

If three days run with the same Top-3 because state hasn't moved, write that explicitly in the brief (e.g. "Action 1 carried forward 3 days, BOF still paused — escalating") instead of repeating the action with new wording.

Output Structure

ONE Google Doc per day. Filename format: Meta Ads Daily Action Plan — {YYYY-MM-DD}.

# Meta Ads Daily Action Plan
{Day, DD Month YYYY} · Generated {HH:MM UK time}

Data refreshed: {intelligence.json._metadata.generated_at} · Calendly: {fetched_at} · Live Meta API run today

> Pipeline form-submit numbers de-dupe `lead` / `fb_pixel_lead` / `lead_grouped`. They are application submissions, not booked calls. Booked-call counts come from Calendly only.

## Today's 3 Actions

1. [Action] · [Concrete Ads Manager change]
2. [Action] · [Concrete Ads Manager change]
3. [Action] · [Concrete Ads Manager change]

## Pipeline State

Real bookings (paid ads, Calendly): {X} this week, {Y} in 28d pipeline
Awaiting close: {Z}
Signed this week (sheet revenue, email-matched): {N}
Spend (BOF only, 7d): £{S}
ROAS: {value or "Pending"}
Cost per real call: £{X}

## Ad Set Verdicts

| Stage | Ad Set | Status | Spend 7d | Stage-relevant metric | Cost per booking pixel | Verdict | Reasoning |
|---|---|---|---|---|---|---|---|
| BOF | Warm Retargeting | ACTIVE | £203 | £18.45 / real call (Calendly headline) | £185 (5 bookings) | SCALE | Healthy by BOF rule (≤ £200), scaling at 8% increments |

Rules:
- Filter out PAUSED ad sets.
- Group by stage: BOF first, MOF, TOF.
- "Stage-relevant metric" column changes by stage: cost per real Calendly call for BOF (account headline), cost per opt-in for MOF, cost per IG follower for TOF.
- "Cost per booking pixel" column reads `metrics_7d.cost_per_booking_pixel` and `metrics_7d.bookings_pixel`. BOF rows only — leave blank for MOF / TOF. This is the per-ad-set kill/scale anchor.
- If `bookings_pixel_7d == 0` AND `applications_7d > 0`, write "pixel match rate low" with a footnote pointing to the Calendly cross-reference, not "£None".

## Anomalies

[Numbered list, one line each. If nothing material, write "No anomalies today."]

## Today's Ads Manager Changes

[Concrete bullets Mahmoud executes. Filter out paused ad sets:]
- Kill Reel 5 in Cold Lookalike (BOF, CTR 0.6%, 0 calls in 7d)
- Increase Warm Retargeting daily budget from £50 to £55 (8% bump, scale rule respected)
- Duplicate "OG Hook" creative into Cold Interest with new 9-second variant
- Pause Profile Visits creative #3 (TOF, cost per follower £4.20, target £1.50)

## Yesterday's Flags — Acted On?

[One bullet per yesterday's action with status: actioned / not actioned / superseded. If yesterday's plan didn't exist, omit this section.]

Tone & Output Rules (binding, see 1-meta-ads/CLAUDE.md §6)

Hard nevers:

• Kill verdicts on ad sets with status=PAUSED. Filter by active status before issuing kill.
• "Simplify the form", "reduce landing page friction", "remove qualification". The application form IS the qualification filter by design.
• Treat Meta pixel lead events as booked calls. Always cross-reference Calendly.
• Team-member names (Mahmoud, Ben, Rob, Jay, Antonio) in free-text recommendations.
• Internal snake_case field names (paid_calls_booked, taken_no_outcome) in prose.
• Assert "the call happened" as fact. Write "scheduled calls" or "bookings".
• Speculate about sales-call quality, pitch clarity, onboarding friction.
• Directive "kill immediately". Reframe as "watch closely" or "concern" with reasoning.
• Em-dashes (—), en-dashes (–) anywhere.
• Pad numbers to look precise when the data is directional.
• Paraphrase or interpret signup names from email handles. A user with email [e-mail] is not "Danny Compass". Read the literal name field from sheet_revenue.signups_7d[] and use it verbatim. If name is missing, write "name not in sheet" rather than guess.
• Word "baseline" for the 8% form submit rate or any other framework threshold. The 8% is a target, not a baseline. TCC isn't historically at 8%. Always write "8% target" or "8% threshold". Same applies to other framework thresholds (£50 cost per submit ceiling, 1.5% CTR target, etc.).

Hard alwayses:

• Data freshness stamp at the top (read intelligence.json._metadata.generated_at + collector fetched_at).
• Data caveat block where any Meta pixel number appears (form submits ≠ calls).
• Stage-aware verdicts. Never issue a verdict without declaring the stage.
• Per-stage reasoning. Explain which stage rule applies.
• Pipeline state when revenue-related. If pending paid bookings exist, say so explicitly.
• British English. £ for money. "Optimise", "behaviour", "organisation".
• Read intelligence._metadata.meta_ads_staleness_warning before drafting. If the field is non-empty, prepend it to the data caveat block verbatim. If it starts with CRITICAL, abort the run and surface the warning instead of shipping a Doc on stale data. The pipeline already exits non-zero in that case, but the skill must not silently produce a brief on data older than 14 days.

Voice: Eight-figure consultant. State what to do and why. Don't hedge.

Google Doc Writing

Markdown rendered as a styled Google Doc via the md-to-gdocs skill. Direct documents.batchUpdate writes raw text, no headings, no tables. Don't go that route.

Workflow:

# 1. Convert today's MD into a .docx with proper styles
python3 jay-skills/md-to-gdocs/convert.py \
    1-meta-ads/meta-ads-daily-action-plan/outputs/{YYYY-MM-DD}.md \
    "Meta Ads Daily Action Plan — {YYYY-MM-DD}"

# 2. Upload + convert to Google Doc in one step (mimeType triggers conversion)
gws drive files create \
    --upload "Meta Ads Daily Action Plan — {YYYY-MM-DD}.docx" \
    --params '{...}' \
    --format json

# 3. The upload arrives as .docx in Drive. Issue a copy with target mimeType
#    `application/vnd.google-apps.document` to convert. The original .docx
#    can then be deleted.
gws drive files copy \
    --params '{"fileId":"<DOCX_ID>"}' \
    --json '{"name":"Meta Ads Daily Action Plan — {YYYY-MM-DD}","mimeType":"application/vnd.google-apps.document","parents":["<FOLDER_ID>"]}' \
    --format json

Folder ID is stored in outputs/folder_id.txt after first run. If Ben hasn't shared the TCC Drive folder yet, the Doc lands in the runner's My Drive root and gets relocated manually.

Clean up the intermediate .docx (local file + uploaded copy in Drive) once the converted Google Doc is confirmed.

Local Output (always)

Mirror the brief to outputs/{YYYY-MM-DD}.md so:

• Yesterday's plan is readable for tomorrow's "did we act on it?" check.
• Git diffs show how the brief evolves week to week.
• If Drive write fails, the brief still exists locally.

When to Invoke

• Manually: every morning by Mahmoud.
• Eventually: scheduled via Claude Routines (every weekday morning, 07:00 UK).
• First few automated runs need Antonio's sanity check, then 2-3 days validation by Mahmoud before acting on recommendations for ad spend changes.

Pre-ship checklist (binding, see 1-meta-ads/CLAUDE.md §8)

Before writing any Doc to Drive, every run must verify:

• Bookings come from Calendly, revenue from sheet_revenue.json email-matched, Meta pixel only for ad-set health.
• Every verdict declares its stage. No verdict without stage.
• Per-stage metrics applied. No BOF thresholds on TOF / MOF.
• Kill verdicts exclude PAUSED ad sets.
• No team names, no snake_case in prose, no "simplify the form", no "kill immediately", no em-dashes.
• Freshness stamp present. Data caveat present where pixel numbers appear.
• ROAS shows "Pending" (not "0.0x") when bookings exist but revenue not yet matched.
• If Meta and Calendly disagree, the gap is explained in-text, not silently picked.

If any check fails, stop and fix before shipping.

Version

v1.1 (2026-05-02) — anchor BOF kill/scale on cost_per_booking_pixel per Antonio's pipeline update.

• §3 BOF thresholds rewritten around cost_per_booking_pixel_7d. £150 / £200 / £300 bands calibrated against Mahmoud's £185 baseline.
• cost_per_form_submit demoted to volume signal. Application-to-booking ratio ~4.5:1 makes it dangerous as a kill anchor (£3 cost per app can sit on top of £349 cost per booking).
• §2 BOF metric split into headline (Calendly account-level ROAS) and per-ad-set health (pixel bookings).
• Pixel match-rate edge case spec'd: zero pixel bookings + non-zero applications + non-zero Calendly = ad blocker / cookie loss, NOT a kill verdict.
• §6 staleness gate: read intelligence._metadata.meta_ads_staleness_warning, abort on CRITICAL.
• Verdict table gains a "Cost per booking pixel" column reading metrics_7d.cost_per_booking_pixel.

v1.0 (2026-04-28) — initial action-first brief, inherits from daily-review + weekly-intelligence, writes one Google Doc per day to the TCC Automations → Meta Ads Daily Briefs folder. Anchored to 1-meta-ads/CLAUDE.md framework rules (stage classification, per-stage metrics, hard nevers, build checklist).