cleanexpo cmo-growth
Bottles 15+ years of CMO/Growth expertise — iOS 14.5 attribution decay, brand-vs-performance tension, the dishonesty in last-click data — into a deterministic daily brief + an event-triggered alert stream.
Why this exists
Eleven businesses, eleven marketing P&Ls. The founder needs LTV:CAC, blended CPA, and channel concentration in one glance every morning, plus a hard gate on any ad-spend request above $5k/day. Everything else flows into the daily 6-pager (daily-6-pager skill, RA-1863).
Decision rights
Autonomous:
- Compute + emit marketing metrics every cycle
- Draft daily marketing brief (routes through pii-redactor + draft_review)
- Alert on threshold breaches (also through draft_review)
- Approve ad-spend up to $5,000/day (configurable via
TAO_CMO_ADSPEND_CEILING)
HITL (dual-key gate):
- Ad-spend > $5,000/day on any single channel
- Brand-message changes (tone, taglines, voice)
- Pricing changes
- New channel launch with budget > $10,000
Metrics owned
| Metric | Formula | Floor / target | Alert threshold |
|---|---|---|---|
| LTV:CAC ratio | (avg_ltv_months × ARPU × gross_margin) / blended_CPA | ≥ 5.0 (target) | < 3.0 → warn |
| Blended CPA | total_marketing_spend / total_customers_acquired | < $250 | > $250 → warn |
| Channel concentration (top share) | top_channel_spend / total_spend | < 70% | > 70% warn, > 85% critical |
| Channel HHI | sum(share²) | < 0.50 | reported, no fixed threshold |
| Attribution decay | 1 − (signal_count / baseline_count) | < 30% | > 30% → info |
Cadence
| Trigger | Action |
|---|---|
| Daily 06:00 UTC (configurable) | Assemble daily brief → pii-redact → draft_review post to REVIEW_CHAT_ID |
| Per-cycle (5 min) | Recompute metrics; alert on breach |
| Per ad-spend request | Dual-key gate; emit cmo_adspend_approved or cmo_adspend_blocked |
Pipeline
read_marketing_provider() ← thin abstraction (Google Ads + LinkedIn + Meta + ...
in production; synthetic in tests)
→ compute_metrics(raw) ← pure-Python; deterministic
→ detect_breaches(metrics, last) ← compares to prior cycle snapshot
→ if breach: emit cmo_alert + draft_review.post_draft (severity-gated)
→ if daily-fire window: assemble_brief + post to draft_review
→ emit cmo_metric_snapshot to .harness/swarm/cmo_state.jsonl
Ad-spend approval flow
incoming request {amount_usd_per_day, channel, business_id, justification}
→ if amount_usd_per_day <= AUTONOMOUS_ADSPEND_CEILING_USD_PER_DAY:
emit cmo_adspend_approved → return approved
→ else:
draft_review.post_draft(
draft_text="📈 Ad-spend approval — ${amount}/day on {channel} ...",
destination_chat_id=REVIEW_CHAT_ID,
drafted_by_role="CMO")
emit cmo_adspend_blocked (queued, awaiting reaction)
return pending
Contract
Daily brief output (passes to draft_review):
📈 CMO daily — {date}
Portfolio LTV:CAC: {x}x | Total spend: ${y} | {n} alerts ({n_critical} critical)
🚨 Alerts:
🔴 [{biz}] channel_concentration: google-ads 91% of spend — single-platform risk.
🟡 [{biz}] ltv_cac_ratio: 2.4 < 3.0 — channel mix or pricing under review.
Per-business:
- restoreassist: $4,200 spend | L:C 4.8 | CPA $87 | top: google-ads 64%
- ccw-crm: $2,100 spend | L:C 3.1 | CPA $112 | top: linkedin 42%
- ...
📥 2 ad-spend approvals queued (>$5,000/day)
Metric snapshot row (per cycle, written to .harness/swarm/cmo_state.jsonl):
{
"ts": "ISO-8601",
"business_id": "restoreassist",
"mrr": 12500.0,
"blended_cpa_usd": 87.40,
"ltv_usd": 1860.0,
"ltv_cac_ratio": 21.28,
"channel_concentration_hhi": 0.46,
"top_channel": "google-ads",
"top_channel_share": 0.64,
"attr_decay": 0.18,
"total_spend_usd": 4200.0
}
Safety bindings
- Read-mostly. CMO bot calls ad-platform APIs in read scope only. No campaign create/edit, no budget changes, no creative uploads without HITL.
- Ad-spend ceiling persisted (
AUTONOMOUS_ADSPEND_CEILING_USD_PER_DAY = $5,000/day). Editing requires user approval, audit-trailed. - PII-redactor in front of every draft — campaign names / customer-segment labels / partner names flow through pii-redactor at strictness=standard.
- 24h alert dedup — same breach won't re-fire within 24h (uses last-snapshot diff).
- Kill-switch aware. On
TAO_SWARM_ENABLED=0, the daily brief is queued (not sent) until resume.
When NOT to use
- Creative production (ad copy, landing pages, blog posts, video) — that's the marketing-* skill family (
marketing-orchestratorand siblings). - Campaign launch planning — that's
marketing-launch-runbook. - ICP / positioning research — that's
marketing-icp-research/marketing-positioning. - Brand-equity tracking — separate Wave 5 ticket.
Verification (Wave 4 A2)
- Synthetic data → LTV:CAC computed correctly (within ±0.05x of hand-calc)
- Top-channel share > 85% → critical alert fires
- $4,500/day ad-spend → auto-approves
- $7,500/day ad-spend → routes through draft_review, status
pending, no real action - Daily brief assembles in <30s with all 11 businesses
- Brief posts through pii-redactor — sensitive segment names redacted if flagged
cmomodule compiles + imports clean alongside existingswarm/*
Out of scope (Wave 4 A2)
- Real ad-platform wire-up — engine ships against a pluggable
marketing_providercallable; ad-platform connector is a follow-up - Mixed-media-modeling (MMM) — uses last-touch + HHI in v1
- Multi-currency consolidation — assumes USD; flagged for follow-up
References
- Blueprint:
~/Pi-CEO/Senior-Agent-Operations-Blueprint-2026-05-02.md§Role 5 CMO/Growth - Parent ticket: RA-1860 (epic RA-1858)
- Sibling creative skills:
marketing-orchestratorandmarketing-*family - HITL gate substrate:
swarm/draft_review.py - Multi-agent debate scaffold:
swarm/debate_runner.py(RA-1867)