start-vibing-stacks 2.29.0 → 2.30.0

package/stacks/_shared/ads-agents/references/bidding-strategies.md

@@ -0,0 +1,333 @@
+# Bidding Strategy Decision Trees
+
+<!-- Updated: 2026-05-14 | v1.6: corrected AI Max conversion lift + DSA migration timing -->
+<!-- Sources: Google Research PDF 2, Claude Research, Gemini Research, 2026 Verification Pass -->
+
+## Google Ads Bidding Decision Engine
+
+### Strategy Selection Flow
+
+```
+START → How many conversions in last 30 days?
+
+┌─ < 15 conversions ──────────────────────────┐
+│  → Maximize Clicks (cold start)             │
+│  → Set Max CPC = Target_CPA / (CVR × 1.5)  │
+│  → Learning: 3-5 days                       │
+│  → Monitor until 15+ conversions achieved    │
+│  → OR use Manual CPC if full control needed  │
+│  → NOTE: eCPC fully removed March 2025       │
+└──────────────────────────────────────────────┘
+
+┌─ 15-29 conversions ─────────────────────────┐
+│  → Maximize Conversions (uncapped)          │
+│  → Learning: 7-14 days                      │
+│  → Transition when CPA SD < 20% over 14d   │
+│  → THEN switch to Target CPA               │
+└──────────────────────────────────────────────┘
+
+┌─ 30+ conversions, NO dynamic values ────────┐
+│  → Target CPA                               │
+│  → Strict: 30+ (Google says 15, but 30+     │
+│    for reliable performance, 50+ ideal)      │
+│  → Set at: 1.1x-1.2x historical CPA        │
+│  → Adjust: Max 10% change every 14 days     │
+│  → Never lower by more than 15% at once     │
+└──────────────────────────────────────────────┘
+
+┌─ 50+ conversions WITH dynamic values ───────┐
+│  → Target ROAS                              │
+│  → Requires dynamic conversion values        │
+│  → Set at: EXACT historical ROAS            │
+│  → Formula: Bid = P(conv) × Value × 1/tROAS│
+│  → Adjust: Same rules as tCPA              │
+└──────────────────────────────────────────────┘
+
+### Smart Bidding Exploration (Google, 2025+)
+
+Allows flexible ROAS targets to discover new traffic. Delivers 18% more
+unique search query categories + 19% more conversions.
+
+- **How it works**: Algorithm temporarily relaxes ROAS constraints to
+  enter auctions it would normally skip, testing new user segments
+- **When to enable**: Stable tROAS campaigns with 50+ conversions/month
+  seeking incremental growth beyond current query coverage
+- **When to avoid**: Tight-margin accounts or campaigns already spending
+  full budget (exploration adds volume, not just efficiency)
+- **Available on**: Target ROAS bid strategies only
+
+### AI Max for Search (Google, 2025+)
+
+AI Max for Search layers broad match + keywordless targeting on existing
+Search campaigns. Google reports 7% more conversions or conversion value
+at comparable CPA/ROAS when the full feature suite is enabled (search
+term matching + text customization + final URL expansion). DSA officially
+being consolidated into AI Max with automatic upgrades starting September
+2026; new DSA campaign creation ends September 2026. Requires strong
+negative keyword lists.
+
+- **How it works**: Automatically expands keyword coverage using AI to
+  match user intent beyond exact keyword matches
+- **When to enable**: Established Search campaigns with healthy conversion
+  volume and comprehensive negative keyword lists
+- **When to avoid**: Low-budget campaigns without negative keyword hygiene;
+  accounts with strict brand safety requirements
+- **Migration note**: Migrate DSA campaigns to AI Max before September 2026 auto-upgrade
+
+┌─ Brand Protection campaigns ────────────────┐
+│  → Target Impression Share                  │
+│  → Target: 95-100% on brand keywords        │
+│  → No conversion data requirement           │
+│  → Search only (not available elsewhere)     │
+└──────────────────────────────────────────────┘
+
+SPECIAL CASES:
+- PMax: Always uses Maximize Conversions or Maximize Conv Value
+- Demand Gen: Supports Target CPC (new), tCPA, tROAS, Max Clicks
+- Display Pay per Conversion: Eligibility required
+```
+
+### Attribution Model (Google)
+- **DDA (Data-Driven Attribution) is now MANDATORY default** (September 2025)
+- Only two models remain: DDA and Last Click
+- All rule-based models deprecated (first-click, linear, time decay, position-based)
+- No minimum data threshold for DDA
+- Impact on bidding: DDA distributes conversion credit across touchpoints, affecting Smart Bidding signals
+
+### Portfolio Bid Strategies
+
+When to use portfolios:
+- Multiple campaigns each with <15 conv, but combined >30
+- Need cross-campaign budget optimization
+- **CPC Cap Hack**: Only way to set max CPC bid on tCPA/tROAS
+
+Rules:
+- Minimum 3 campaigns per portfolio for meaningful data
+- Group campaigns with similar target CPAs/ROAS
+- Never mix brand and non-brand in same portfolio
+
+### Transition Triggers (Google)
+
+| From | To | Trigger |
+|------|----|---------|
+| Maximize Clicks | Maximize Conversions | 15+ conversions in 30 days |
+| Maximize Conversions | Target CPA | CPA SD <20% over 14 days + 30+ conv |
+| Target CPA | Target ROAS | 50+ conv + dynamic values available |
+| Manual CPC | Maximize Clicks | Ready to test automation |
+| Any | Target Impression Share | Brand protection need identified |
+
+---
+
+## Meta Ads Bidding Decision Engine
+
+### Auction Formula
+
+```
+Total Value = Bid × Estimated Action Rate + User Value
+```
+
+### Strategy Selection
+
+```
+DEFAULT (90% of campaigns):
+  → Lowest Cost (no cap)
+  → Best for: Most campaigns, volume maximization
+  → Risk: CPA can spike during high-competition (Q4)
+
+IF need cost predictability:
+  → Cost Cap
+  → Set at: 1.2x - 1.5x target CPA
+  → Best for: Scaling with margin protection
+  → Risk: Under-delivery if cap too aggressive
+
+IF need strict cost control:
+  → Bid Cap
+  → Set at: 2x - 3x target CPA
+  → Best for: Experienced advertisers with clear unit economics
+  → Risk: Significant under-delivery if too low
+
+IF e-commerce with revenue tracking:
+  → ROAS Goal
+  → Best for: Advantage+ Sales Campaigns
+  → Requires: Purchase event + dynamic values
+
+IF high-value diverse products:
+  → Highest Value
+  → Prioritizes high-value conversions
+  → Best for: Wide AOV range
+```
+
+### Andromeda Context
+
+Under Andromeda, creative diversity matters more than bid strategy.
+Focus budget on creative testing, not micro-optimizing bids.
+
+### CBO vs ABO Decision
+
+```
+IF daily_budget < $100:
+  → ABO (manual control for testing)
+
+IF daily_budget $100-$500:
+  → Test both; CBO if similar audiences, ABO if diverse
+
+IF daily_budget > $500:
+  → CBO (Advantage+ Campaign Budget)
+  → Let Meta optimize distribution
+
+NOTE: Advantage+ Sales/Leads auto-enable CBO
+```
+
+### Learning Phase Rules
+
+**Exit criteria:** 50 conversions per week per ad set
+
+**Reset triggers (avoid during learning):**
+- Budget change >20%
+- Any targeting change
+- Creative edit (even text)
+- Bid strategy change
+- Pausing >7 days
+
+**If "Learning Limited":**
+1. Broaden audience
+2. Increase budget
+3. Switch to higher-funnel event (AddToCart instead of Purchase)
+4. Consolidate ad sets
+5. Ensure daily budget ≥5× target CPA
+
+---
+
+## LinkedIn Ads Bidding
+
+### Strategy Selection
+
+```
+Sponsored Content:
+  → Start with Manual CPC for cost control (recommended default)
+  → Test automated bidding only with established data
+  → Maximum Delivery is the most expensive option
+  → Minimum daily budget: $10/day (recommend $50+)
+
+Message Ads:
+  → CPS (Cost Per Send): bid aggressively
+  → Frequency cap: 1 per 30-45 days per user
+  → Win the auction for limited inbox slot
+
+Lead Gen Forms:
+  → Lowest Cost or Manual CPC
+  → 13% CVR benchmark (3.25× landing pages)
+
+Conversation Ads:
+  → CPS: similar to Message Ads
+  → 10-12% CTR benchmark
+
+TLA (Thought Leader Ads):
+  → Allocate ≥30% budget
+  → CPC $2.29-$4.14 (vs $13.23 standard)
+
+Accelerate Campaigns:
+  → AI-powered end-to-end
+  → 42% lower CPA, 21% lower CPL
+```
+
+---
+
+## TikTok Ads Bidding
+
+### Strategy Selection
+
+```
+DEFAULT:
+  → Lowest Cost (volume maximization)
+  → Most common, good starting point
+
+IF margin protection needed:
+  → Cost Cap
+  → Set target CPA, system tries to stay near
+
+IF strict cost control needed:
+  → Bid Cap
+  → Controls maximum bid per auction (not average CPA)
+  → Set at 2-3x target CPA initially
+  → Best for: retargeting, competitive niches (Q4/BFCM)
+  → Risk: significant under-delivery if set too aggressively
+
+IF Smart+ campaign:
+  → Modular automation (choose per module)
+  → Up to 30 ad groups, 50 creatives per asset group
+  → Smart+ now supports modular control (2025). Lock
+    targeting/creative/budget/placement independently
+    while automating others
+
+GMV Max (TikTok Shop):
+  → Default/only for Shop campaigns (July 2025)
+  → Optimizes for Gross Merchandise Value
+```
+
+### Budget Rules
+- Campaign minimum: $50/day
+- Ad group minimum: $20/day
+- Budget should be ≥50× target CPA per ad group (provides sufficient learning room)
+- Learning phase: ~50 conversions in 7 days
+
+---
+
+## Microsoft Ads Bidding
+
+### Strategy Selection
+
+```
+IF importing from Google:
+  → Match Google bid strategy
+  → Validate settings transferred correctly
+  → Adjust targets (typically 20-35% lower CPC on Bing)
+
+IF native campaign:
+  → Same strategy framework as Google
+  → Lower CPC targets (expect $1.20-$1.55 avg)
+  → Enable Copilot placement for PMax
+
+PMax on Microsoft:
+  → Enable "Target New Customers" (Beta 2026)
+  → Copilot chat placement: 73% higher CTR
+  → LinkedIn profile targeting (unique)
+```
+
+---
+
+## Apple Ads Bidding
+
+### Maximize Conversions (GA Feb 26, 2026)
+
+AI-powered auto-bidder using Search Match. Currently only optimizes for installs.
+
+```
+Strategy Selection:
+  → Maximize Conversions (recommended for most campaigns)
+  → Target CPA (weekly average target) replaces CPA Cap (being deprecated)
+  → Daily budget >= 5x target CPA
+  → Two-week learning period before evaluating performance
+
+Key Rules:
+  → Set realistic target CPA based on historical install cost
+  → Avoid budget or target changes during the 2-week learning period
+  → Currently only optimizes for installs (not post-install events)
+  → Pair with strong Search Match negative keywords to control relevance
+```
+
+---
+
+## Cross-Platform Bidding Red Flags
+
+| Red Flag | Severity | Platform | Action |
+|----------|----------|----------|--------|
+| Broad Match + Manual CPC | Critical | Google | Switch to Smart Bidding or Exact Match |
+| tCPA <50% of actual CPA | Critical | Google | Unrealistic target; set at 1.1-1.2× historical |
+| Smart Bidding with <15 conv/month | High | Google | Use Manual CPC or Maximize Clicks |
+| >50% ad sets "Learning Limited" | Critical | Meta | Consolidate, broaden audience, increase budget |
+| Cost Cap below historical CPA | High | Meta | Set at 1.2-1.5× target, not below |
+| Daily budget <5× CPA | High | Meta | Increase budget or switch to higher-funnel event |
+| Budget < $50/day campaign | High | TikTok | Increase to minimum or consolidate |
+| No negative keywords with Broad Match | Critical | Google | Add themed negative lists immediately |
+| No Portfolio strategy for low-vol campaigns | Medium | Google | Group <15 conv campaigns into portfolio |

package/stacks/_shared/ads-agents/references/brand-dna-template.md

@@ -0,0 +1,190 @@
+# Brand DNA Template
+
+<!-- Updated: 2026-03-12 -->
+<!-- Used by: ads-dna (extraction), ads-create (consumption), ads-generate (style injection) -->
+
+## Purpose
+
+Defines the canonical schema for `brand-profile.json`. All agents that produce or
+consume brand data must use these exact field names and value types.
+
+---
+
+## JSON Schema
+
+```json
+{
+  "schema_version": "1.0",
+  "brand_name": "string",
+  "website_url": "string (full URL including https://)",
+  "extracted_at": "ISO-8601 timestamp (e.g. 2026-03-12T14:30:00Z)",
+
+  "voice": {
+    "formal_casual":          7,
+    "rational_emotional":     4,
+    "playful_serious":        5,
+    "bold_subtle":            8,
+    "traditional_innovative": 6,
+    "expert_accessible":      5,
+    "descriptors": ["authoritative", "warm", "direct"]
+  },
+
+  "colors": {
+    "primary":    "#1A2E4A",
+    "secondary":  ["#F4A623", "#FFFFFF"],
+    "forbidden":  ["#FF0000"],
+    "background": "#F9F9F9",
+    "text":       "#1A1A1A"
+  },
+
+  "typography": {
+    "heading_font":       "Inter",
+    "body_font":          "Source Sans Pro",
+    "pairing_descriptor": "modern sans-serif, clean and readable"
+  },
+
+  "imagery": {
+    "style":       "professional photography",
+    "subjects":    ["people using product", "clean product shots", "diverse workforce"],
+    "composition": "clean backgrounds, good negative space",
+    "forbidden":   ["stock photo clichés", "corporate handshakes", "cheesy smiles"]
+  },
+
+  "aesthetic": {
+    "mood_keywords": ["trustworthy", "modern", "approachable"],
+    "texture":        "minimal, flat",
+    "negative_space": "generous"
+  },
+
+  "brand_values":   ["transparency", "innovation", "customer-first"],
+
+  "target_audience": {
+    "age_range":   "28-45",
+    "profession":  "marketing managers and small business owners",
+    "pain_points": ["time-consuming reporting", "unclear ROI"],
+    "aspirations": ["grow efficiently", "look professional"]
+  },
+
+  "screenshots": {
+    "homepage":  "./brand-screenshots/example_com_desktop.png",
+    "secondary": ["./brand-screenshots/example_com_pricing_desktop.png"]
+  }
+}
+```
+
+---
+
+## Field Reference
+
+### Voice Axes (1-10 scale)
+
+Score interpretation: 1 = extreme left pole, 10 = extreme right pole, 5 = neutral.
+
+| Field | 1 (Left) | 10 (Right) | Ad Implication |
+|-------|----------|------------|----------------|
+| `formal_casual` | Very formal, corporate | Very casual, conversational | Headlines tone |
+| `rational_emotional` | Data-driven, logical | Emotionally evocative | Story vs stats |
+| `playful_serious` | Fun, humorous | Serious, no-nonsense | CTA phrasing |
+| `bold_subtle` | Big claims, loud | Understated, nuanced | Visual hierarchy |
+| `traditional_innovative` | Classic, established | Cutting-edge, disruptive | Imagery style |
+| `expert_accessible` | Deep expertise, jargon | Everyone can understand | Copy complexity |
+
+`descriptors` (array of 3-5 strings): Free-form adjectives capturing tone not covered by axes.
+
+### Colors
+
+- `primary`: Main brand color. **Always inject into image generation prompts.**
+- `secondary`: Supporting palette. Use for accents in generation prompts.
+- `forbidden`: Colors to explicitly exclude from prompts (e.g., competitor brand colors).
+- `background` / `text`: Digital UI colors (less relevant for ad image generation).
+
+**Null handling:** If extraction fails for a color, set to `null`. Agents must skip null
+color injection rather than defaulting to arbitrary colors.
+
+### Typography
+
+Used by `copy-writer` agent for tone calibration. Not injected into image prompts
+(typography is handled by ad platform copy fields, not image generation).
+
+### Imagery
+
+- `style`: Main descriptor passed to image generation (e.g., "professional photography",
+  "illustration", "flat design").
+- `subjects`: Array injected as subject guidance in prompts.
+- `composition`: Passed verbatim as a composition constraint.
+- `forbidden`: Converted to negative prompt modifiers (e.g., "no corporate handshakes").
+
+### Aesthetic
+
+- `mood_keywords`: Injected directly into generation prompts as atmosphere descriptors.
+- `texture`: Passed as texture preference.
+- `negative_space`: "generous" → "plenty of white space, uncluttered composition".
+
+### Screenshots (optional, populated by `/ads dna`)
+
+- `homepage`: Desktop screenshot of homepage. **Primary visual style reference** for
+  `visual-designer` agent. When present, generation auto-uses `gemini-3.1-flash-image-preview`
+  (Nano Banana 2) for better brand style-transfer.
+- `secondary`: Additional pages (pricing, about, product). Supplemental context.
+- **If field absent**: `visual-designer` falls back to text-only generation (`gemini-2.5-flash-image`).
+- **If field present but file missing on disk**: `visual-designer` logs a warning and falls back gracefully.
+- Set by `ads-dna` Step 2b. Omitted entirely if screenshot capture failed or `--quick` was used.
+
+---
+
+## Extraction Guide
+
+### Pages to scan (in order)
+
+1. **Homepage**: Primary brand impression, dominant colors, hero headline tone
+2. **About / Our Story**: Brand values, voice descriptors, team photography style
+3. **Product / Service page**: Imagery style, composition, subject matter
+4. **Existing ads** (if accessible via Meta Ad Library or Google Ads Transparency): Override
+
+### CSS extraction targets
+
+```
+background-color, color         → colors.primary / secondary
+font-family                     → typography.heading_font / body_font
+@import url (Google Fonts)      → typography (font name from URL path)
+og:image meta tag               → imagery.style (analyze dominant visual)
+```
+
+### Voice scoring heuristics
+
+| Signal | Scoring |
+|--------|---------|
+| Uses "you" and "your" frequently | formal_casual +2 (casual) |
+| Industry jargon in hero text | expert_accessible -2 (expert) |
+| Short sentences (≤10 words) | bold_subtle +1 (bold) |
+| Testimonials lead with emotion | rational_emotional +2 (emotional) |
+| "Trusted by 10,000+ companies" | traditional_innovative -1 (traditional) |
+
+### Fallback heuristics
+
+- **No Google Fonts detected**: Set `typography.heading_font` to "system-ui"
+- **Sparse content (< 200 words)**: Mark `extracted_at` note: "low confidence extraction"
+- **Dark mode site**: If background is dark (#333 or darker), swap `background` and `text`
+- **Cannot extract primary color**: Use most prominent color from og:image analysis
+
+---
+
+## Usage in Generation Prompts
+
+When building image generation prompts, inject brand DNA as:
+
+```
+"[subject], [imagery.style], [imagery.composition],
+brand colors [colors.primary] and [colors.secondary[0]],
+[aesthetic.mood_keywords joined by comma] atmosphere,
+[aesthetic.texture] texture, [aesthetic.negative_space] composition,
+no [imagery.forbidden joined by comma]"
+```
+
+Example output:
+```
+"person using laptop, professional photography, clean backgrounds good negative space,
+brand colors #1A2E4A and #F4A623, trustworthy modern approachable atmosphere,
+minimal flat texture, generous white space composition,
+no stock photo clichés, no corporate handshakes"
+```

package/stacks/_shared/ads-agents/references/budget-allocation.md

@@ -0,0 +1,150 @@
+# Budget Allocation & Scaling Framework
+
+<!-- Updated: 2026-02-10 -->
+<!-- Sources: Google Research PDF 2, Claude Research, Gemini Research -->
+
+## Platform Selection Matrix by Business Type
+
+| Business Type | Platform 1 | Platform 2 | Platform 3 | Other | Min Monthly | Primary KPI | Time to Profit |
+|--------------|------------|------------|------------|-------|-------------|-------------|---------------|
+| SaaS B2B | Google 35-45% | LinkedIn 30-40% | Meta 15-25% | YouTube 10% | $5,000 | Pipeline ROI, LTV:CAC | 3-6 months |
+| E-commerce DTC | Meta 50-68% | Google PMax 23-30% | TikTok 5-15% | Email 5% | $3,000 | ROAS, MER, POAS | 0-2 months |
+| Local Service | Google LSA/Search 60% | Meta 30% | Bing 10% | N/A | $1,500 | Cost Per Lead/Booking | 1 month |
+| B2B Enterprise | LinkedIn 39-60% | Google 20-35% | ABM Display 15-20% | Programmatic 11% | $10,000 | Pipeline, SQLs | 6-12 months |
+| Info Products | YouTube 40% | Meta 40% | Email 20% | N/A | $2,000 | ROAS, Webinar CPL | 1-3 months |
+| Mobile App | Apple Search 30% | Google App 30% | Meta/TikTok 40% | N/A | $5,000 | CPI, LTV, D7 Retention | 3-6 months |
+| Real Estate | Meta Lead Forms 50% | Google Search 40% | LinkedIn 10% | N/A | $2,500 | Cost Per Lead | 2-4 months |
+| Healthcare | Google Search 55% | Meta 20% | Microsoft 10% | YouTube/Display 15% | $4,000 | Cost Per Patient | 2-5 months |
+| Finance/Fintech | Google Search 40-50% | LinkedIn 20-30% | Meta 15% | YouTube/Display 10% | $8,000 | CAC, LTV:CAC | 4-8 months |
+| Agency (Own) | LinkedIn 50% | Meta 30% | Google 20% | N/A | $1,500 | Cost Per Lead | 1-3 months |
+| B2C / DTC | Meta 50% | Google 20% | TikTok 30% | N/A | $2,000 | ROAS, CAC | 1-3 months |
+
+## Budget Distribution Rule: 70/20/10
+
+```
+70% → Proven performers (campaigns with established positive ROAS/CPA)
+20% → Promising growth (campaigns showing traction, need scale)
+10% → Experiments (new platforms, audiences, formats)
+```
+
+## Scaling Decision Tree
+
+### 20% Rule (Scale Up)
+```
+IF actual_CPA < target_CPA by > 10%
+AND conversions_last_7d >= learning_threshold
+AND campaign NOT in learning phase
+THEN increase budget by 20%
+→ Wait 3-5 days before next increase
+→ Never increase more than 20% at once on Meta (triggers learning reset)
+```
+
+### 3× Kill Rule (Pause)
+```
+IF spend > 3× target_CPA
+AND conversions == 0
+THEN pause ad/ad set/campaign immediately
+→ Review: creative, targeting, landing page, tracking
+→ Don't restart without changes
+```
+
+### Decreasing Returns Detection
+```
+IF CPA increased > 15% after last budget increase
+THEN roll back to previous budget level
+→ Wait 7 days
+→ Try horizontal scaling instead (new audiences, platforms)
+```
+
+### Saturation Signals
+| Platform | Signal | Threshold | Action |
+|----------|--------|-----------|--------|
+| Google | Impression Share | >80% | Diminishing returns; diversify |
+| Meta | Frequency (7-day) | >4.0 | Audience exhausted; refresh creative or expand |
+| TikTok | Frequency | >3.0 | Creative fatigue; replace assets |
+| LinkedIn | Audience penetration | >50% | Audience exhausted; expand targeting |
+
+## Marketing Efficiency Ratio (MER)
+
+```
+MER = Total Revenue / Total Ad Spend (all platforms combined)
+```
+
+| Business Type | Healthy MER | Excellent MER | Danger Zone |
+|--------------|-------------|---------------|-------------|
+| E-commerce | 3.0-5.0 | >5.0 | <2.0 |
+| SaaS | Use LTV:CAC (3:1 target) | >4:1 | <2:1 |
+| Lead Gen | Revenue/Lead × CVR / CPL | N/A | <1.5 |
+
+**Why MER over ROAS:**
+- ROAS is platform-reported (overclaims by 20-40%)
+- MER captures true business economics
+- Cross-platform; not inflated by attribution overlap
+
+## Seasonality Adjustments
+
+### Q4 (October-December)
+- CPMs increase 30-50%
+- Reduce ROAS targets by 20%
+- Front-load creative testing in September
+- Increase budgets 40-60% for Black Friday / Cyber Monday
+- Meta: Advantage+ Sales performs best during peaks
+
+### Q1 (January-March)
+- CPMs decrease 20-30% post-holiday
+- Best time for new campaign testing
+- Aggressive customer acquisition (lower competition)
+- Meta Jan 2026 CPC: $0.85 (24% below prior January)
+
+### Q2 (April-June)
+- Steady CPMs, good testing window
+- Mother's Day / Father's Day spikes for retail
+
+### Q3 (July-September)
+- LinkedIn CPC peaks in Q3 ($15.72 average)
+- Back-to-school surge for education
+- Good time to build retargeting pools for Q4
+
+## Incrementality Testing (Geo-Lift)
+
+### Setup
+- Duration: 2-4 weeks minimum
+- Holdout: 10% of geographic regions (matched pairs)
+- Method: Compare conversion rates exposed vs holdout
+- Statistical significance: p < 0.05
+
+### Tools
+- Meta: Robyn (R-based, open source), best for ~80% of orgs
+- Google: Meridian (Python, Bayesian, geo-level)
+- Advanced: PyMC-Marketing (fully customizable)
+- Budget: Significant; allocate 10% of monthly spend for testing
+
+### Post-Purchase Surveys
+- Fill ~30% of attribution gap that digital tracking misses
+- Question: "How did you first hear about us?"
+- Implement on order confirmation page
+- Weight responses against platform-reported data
+
+## Minimum Viable Budgets
+
+| Platform | Minimum Monthly | Rationale |
+|----------|----------------|-----------|
+| Google Ads | $1,000 | 15+ conv/month minimum for Smart Bidding |
+| Meta Ads | $600-$800 | 50 conv/week/ad set × $CPA |
+| LinkedIn Ads | $3,000 | High CPCs; $10/day minimum × multiple campaigns |
+| TikTok Ads | $300 | $50/day campaign min × creative testing needs |
+| Microsoft Ads | $200-$300 | 20-30% of Google budget |
+
+## Cross-Platform Attribution Hierarchy
+
+```
+Source of Truth (best → worst):
+1. CRM Data (actual revenue, closed deals, LTV)
+2. MER (total revenue / total ad spend)
+3. Post-Purchase Surveys ("How did you hear about us?")
+4. Google Analytics 4 (cross-channel, multi-touch)
+5. Platform-Reported Data (always overclaims 20-40%)
+
+RULE: Never trust platform-reported ROAS alone.
+Always cross-reference with ≥2 other data sources.
+```