ceo-orchestration 1.0.0

package/.claude/skills/domains/paid-media/skills/paid-social-strategist/SKILL.md

@@ -0,0 +1,493 @@
+---
+name: paid-social-strategist
+description: >
+  Paid social discipline covering campaign-objective mapping, platform
+  selection scoring, Advantage-Plus versus manual structure, post-iOS-14.5
+  measurement (SKAdNetwork conversion-value mapping, aggregated event
+  measurement), creative-volume strategy, bid and budget mechanics, and
+  attribution methodology. Operates across Meta, TikTok, LinkedIn, Twitter/X,
+  Reddit, Pinterest, and Snapchat. Enforces funnel-stage alignment before any
+  platform or budget decision, treats creative as the primary algorithmic
+  lever, and rejects single-source attribution by default. Use when designing
+  or restructuring a paid social campaign; selecting or rebalancing a platform
+  mix; architecting Advantage-Plus versus manual structures; building a
+  post-iOS-14.5 measurement stack; designing a creative-volume strategy;
+  constructing a bid or budget framework; or triangulating attribution across
+  platform-reported, lift-study, and modelled sources.
+owner: Valentina Reyes (Paid Social Strategist, domain persona)
+tier: domain:paid-media
+scope_tags: [paid-social, meta-ads, tiktok-ads, linkedin-ads, advantage-plus, skadnetwork, attribution]
+inspired_by:
+  - source: msitarzewski/agency-agents/paid-media/paid-media-paid-social-strategist.md@783f6a72bfd7f3135700ac273c619d92821b419a
+    license: MIT
+    relationship: structural_inspiration
+    authored_by: ceo-orchestration framework
+    authored_at: 2026-05-07
+# --- smart-loading fields (PLAN-083 Wave 0b sub-agent 0.7c) ---
+domain: paid-media
+priority: 8
+risk_class: low
+stack: []
+context_budget_tokens: 500
+inactive_but_retained: true
+repo_profile_binding:
+  frontend: {active: false, priority: 10}
+  engine: {active: false, priority: 10}
+  fintech: {active: false, priority: 10}
+  trading-readonly: {active: false, priority: 10}
+  generic: {active: false, priority: 10}
+activation_triggers: []
+# --- K1 paths: native file-touch activation (PLAN-135 W3 unit k1a) ---
+paths:
+  - "**/paid-social/**"
+  - "**/meta-ads/**"
+  - "**/skadnetwork/**"
+  - "**/campaigns/**"
+---
+
+# Paid Social Strategist
+
+## Cardinal Rule
+
+Paid social spend earns return by matching campaign objective to funnel
+stage, platform to audience density, and creative to platform-native
+consumption behaviour — all three simultaneously. A technically correct
+Advantage-Plus structure pointed at the wrong funnel stage wastes the
+machine-learning budget on audience segments that cannot convert at the
+bid price. A correct objective on the wrong platform wastes impressions
+on users whose consumption context does not support the required action.
+Correct objective, correct platform, and wrong creative produces high
+CPMs against an algorithm that learned the audience but could not drive
+the desired action. All three axes must be aligned before any campaign
+goes live. Platform selection, objective mapping, and creative strategy
+are not sequential decisions — they are a single joint decision made at
+brief stage, not at optimisation stage.
+
+---
+
+## Fail-Fast Rule
+
+A campaign MUST NOT enter active spend without three gates confirmed:
+a defined funnel stage with a corresponding campaign objective, a
+platform selection scored against the audience-platform fit matrix
+(not selected by habit or budget precedent), and a minimum creative
+inventory of four variants per ad set. The following conditions MUST
+hold before any campaign activates:
+
+1. The funnel stage is named (Awareness, Consideration, Conversion,
+   Retention, or Winback) and the campaign objective is confirmed as
+   the correct platform-native objective for that stage — not the
+   default objective selected by the platform wizard.
+2. The platform was selected by scoring audience density, content fit,
+   commercial intent, and cost efficiency against the target segment —
+   not by historical precedent or channel convenience.
+3. The ad set launches with a minimum of four creative variants. Single-
+   creative ad sets are blocked; the algorithm requires creative signal
+   diversity to exit the learning phase efficiently.
+
+If any gate is unresolved, the campaign does not activate until it is
+closed.
+
+---
+
+## When to Apply
+
+Activate this skill when the work involves any of the following:
+
+- Designing or auditing campaign architecture for Meta, TikTok, LinkedIn,
+  Twitter/X, Reddit, Pinterest, or Snapchat.
+- Selecting a platform mix for a new product launch, seasonal push, or
+  budget reallocation cycle.
+- Deciding between Advantage-Plus and manual campaign structure for a
+  specific campaign type or data-density state.
+- Building a post-iOS-14.5 measurement stack: Conversions API, aggregated
+  event measurement configuration, SKAdNetwork conversion-value schema.
+- Designing a creative-volume strategy and refresh cadence.
+- Constructing a bid strategy or budget framework (CBO versus ABO, bid
+  caps versus cost caps, daily versus lifetime budgets).
+- Triangulating attribution across platform-reported metrics, lift studies,
+  marketing mix modelling, and GA4.
+
+Skip when: the task is search advertising — route to
+`domains/paid-media/skills/ppc-strategist`; the task is programmatic
+display or video — route to `domains/paid-media/skills/programmatic-buyer`;
+the task is tracking implementation (pixel, server-side events, GA4
+schema) — route to `domains/paid-media/skills/tracking-specialist`; or
+the task is creative production and concepting — route to
+`domains/paid-media/skills/creative-strategist`.
+
+---
+
+## Platform Selection
+
+Platform selection is a scoring decision, not a channel preference. The
+question is never "should we be on platform X?" — the question is "which
+platforms score highest against the target segment across the four
+evaluation dimensions?" Score before committing budget. Committing to a
+platform before scoring produces spend inertia: budgets persist on
+platforms that no longer match the audience or objective because historical
+presence substitutes for current evidence.
+
+### Evaluation Dimensions
+
+Score each candidate platform on a 1-5 scale across four dimensions. Do
+not weight dimensions equally — weight them by the campaign objective:
+commercial intent is the heaviest weight for conversion campaigns; audience
+density is the heaviest weight for awareness campaigns.
+
+| Dimension | Definition | Conversion weight | Awareness weight |
+|---|---|---|---|
+| Audience density | Verified presence and active engagement of target segment on platform | High | High |
+| Content fit | Platform's native content format matches the creative type the brand can produce | Medium | High |
+| Commercial intent | Platform's usage context supports the required action (click, lead form, purchase) | Highest | Low |
+| Cost efficiency | CPM and CPA benchmarks relative to vertical norms for this platform | Medium | Medium |
+
+### Platform Characteristics (reference — verify against current benchmarks)
+
+| Platform | Audience density strength | Commercial intent | Content fit | Cost profile |
+|---|---|---|---|---|
+| Meta (Facebook/Instagram) | Broadest reach; all demographics | High for ecommerce and lead gen | Image, video, carousel, DPA | Moderate CPM; rising YoY |
+| TikTok | 18-34 dominant; rapid 35+ growth | Moderate; purchase intent improving | Short-form native video only | Lower CPM; rising |
+| LinkedIn | B2B professional; job-title targeting | High for enterprise B2B and recruitment | Sponsored content, document ads | Highest CPM in category |
+| Twitter/X | News-aware, tech-adjacent audiences | Low to moderate | Short text + image | Variable; inventory compressed |
+| Reddit | Interest-graph communities; high intent niche | Moderate; community trust required | Promoted posts in community context | Lower CPM; niche targeting strong |
+| Pinterest | High purchase intent in home, fashion, food | High for visual product categories | Static and video pins | Moderate; strong ROAS in verticals |
+| Snapchat | 13-34 dominant; mobile-first | Moderate; strong for app installs | Vertical video; AR lenses | Lower CPM; young-skew limitation |
+
+A platform that scores below 3 on audience density is not a candidate
+regardless of its other scores. An audience that is not present cannot be
+reached at any cost efficiency.
+
+---
+
+## Campaign Objective Mapping
+
+Campaign objective must match funnel stage. Platform algorithms optimise
+delivery toward the selected objective. Selecting Traffic when the goal is
+Conversions trains the algorithm to find people who click links — a different
+population than people who purchase. Misaligned objectives are not recoverable
+through bid adjustments; the learning phase encodes the wrong population and
+the campaign must be rebuilt, not optimised.
+
+### Funnel-Stage to Objective Mapping
+
+| Funnel stage | Goal | Correct objective | Common misalignment |
+|---|---|---|---|
+| Awareness | Reach and frequency against target segment | Reach, Brand Awareness, Video Views | Traffic (optimises for clickers, not viewers) |
+| Consideration | Engagement, site visits, content consumption | Traffic, Engagement, Video Views | Conversions (insufficient conversion signal; learning fails) |
+| Conversion — Lead | Qualified lead capture | Lead Generation, Leads (Meta) | Traffic (sends to landing page; higher drop-off than native form) |
+| Conversion — Purchase | Direct revenue | Sales, Conversions (Purchase event) | Traffic or Add-to-Cart (underpowers purchase signal) |
+| Retention | Re-engagement and repeat purchase | Sales with existing-customer audience, Engagement | Reach (wastes impressions on cold audiences) |
+| Winback | Lapsed-customer reactivation | Sales or Lead Gen with suppressed recent purchasers | Broad audience (dilutes lapsed-customer signal) |
+
+Objective selection is final at campaign creation on most platforms. Changing
+objective mid-flight resets the learning phase. Build the correct objective
+into the campaign brief before any structure is created.
+
+---
+
+## Advantage-Plus vs Manual
+
+Advantage-Plus (and its equivalents on other platforms — Performance Max
+on Google, Reach and Frequency on TikTok Smart+) shifts targeting, placement,
+and creative selection decisions from the advertiser to the platform
+algorithm. The decision between ML-driven and manual structure is a function
+of data density, not platform preference or account maturity.
+
+### When ML-Driven Structure Outperforms
+
+Advantage-Plus outperforms manual when the following conditions are met:
+
+- **Conversion event volume:** the campaign or account generates a minimum
+  of 50 conversion events per week at the target conversion event level.
+  Below this threshold the algorithm cannot model efficiently; manual
+  targeting preserves a narrower, higher-signal delivery audience.
+- **Creative diversity:** a minimum of six to eight creative variants are
+  available for algorithmic selection. Advantage-Plus with fewer than four
+  variants degrades to random rotation, which manual structure already
+  provides without the ML overhead.
+- **Audience breadth acceptable:** Advantage-Plus will expand targeting
+  beyond defined audiences when it detects conversion signal. This expansion
+  is a feature for scale-stage campaigns and a liability for campaigns with
+  strict audience constraints (geography, compliance, ABM target lists).
+- **No segment-isolation requirement:** A/B tests, incrementality holds,
+  and sequential-messaging programs require audience isolation that
+  Advantage-Plus cannot guarantee.
+
+### Manual Override Scenarios
+
+Manual campaign structure is required when:
+
+1. The conversion event fires fewer than 50 times per week — algorithm
+   cannot exit the learning phase reliably.
+2. The campaign requires strict audience isolation (hold groups, exclusion
+   cells, ABM lists).
+3. The creative inventory is fewer than four variants — ML selection has
+   no meaningful choice surface.
+4. Regulatory or compliance constraints prohibit audience expansion (financial
+   products, health categories, age-gated products).
+
+### Learning-Phase Respect
+
+The learning phase (typically 50 conversion events on Meta) is not a
+period to optimise — it is a period to observe. Editing budgets, bids,
+audiences, or creatives during the learning phase resets it. Allow the
+learning phase to complete before evaluating performance; early exit or
+edit is the most common cause of false-negative performance attribution
+in ML-driven structures.
+
+---
+
+## Audience Signals Post-iOS-14.5
+
+iOS-14.5 App Tracking Transparency (ATT) eliminated deterministic
+cross-site signal for opted-out iOS users. The practical effect: platform-
+reported conversions are a model, not a measurement, for a substantial
+fraction of iOS traffic. Post-iOS-14.5 audience strategy requires a
+different signal architecture.
+
+### SKAdNetwork Conversion-Value Mapping
+
+SKAdNetwork (SKAN) is Apple's privacy-preserving attribution framework.
+It reports conversion events through a 6-bit conversion value (0-63) with
+a 24-72 hour postback timer. The conversion value schema must be designed
+before campaign activation — SKAN does not retroactively map events.
+
+| Conversion value range | Recommended mapping | Signal quality |
+|---|---|---|
+| 0 | No conversion event fired within timer window | No signal |
+| 1-15 | Low-value engagement (add-to-cart, registration) | Weak |
+| 16-31 | Mid-value conversion (trial start, first purchase below AOV) | Medium |
+| 32-63 | High-value conversion (repeat purchase, high-AOV, subscription activate) | Strong |
+
+Map conversion values to revenue buckets, not event types, to enable
+bid optimisation against value. Event-only mapping (purchase = value 1,
+regardless of purchase size) discards the revenue-signal needed for
+value-based bidding.
+
+### Aggregated Event Measurement
+
+Meta's Aggregated Event Measurement (AEM) protocol limits each domain to
+eight conversion events, ranked by priority. Events ranked below the active
+priority are not reported when ATT consent is absent. Configure the eight
+events in priority order matching the campaign objective hierarchy: Purchase
+first, then Add-to-Cart, then Initiate Checkout, then Lead, then lower-
+funnel events. Misconfigured event priority is the most common cause of
+under-reported conversions on Meta for iOS audiences.
+
+### Modelled vs Reported Conversions
+
+Following iOS-14.5, Meta and other platforms supplement deterministic
+attribution with statistical modelling. Modelled conversions appear in
+reporting alongside observed conversions. Do not subtract modelled
+conversions from reported totals as a trust adjustment — modelled conversions
+are directionally accurate at the campaign level and should be treated as
+valid signals for budget allocation decisions, not as noise to be discounted.
+
+### Consent Compliance
+
+Conversions API (CAPI) implementation must route only consented user events
+to platform servers in jurisdictions with consent requirements (GDPR, LGPD).
+Server-side events sent without consent eligibility checking create compliance
+exposure even when the platform receives and processes them. Implement
+consent-state filtering in the CAPI middleware layer before the API call,
+not as a post-processing step.
+
+---
+
+## Creative-Volume Strategy
+
+Creative is the primary algorithmic lever in paid social. Platform algorithms
+allocate delivery budget to the creative variants that generate the best
+signal against the objective. A campaign with a single creative forces the
+algorithm into a no-choice state; the algorithm delivers that creative
+regardless of performance signal because there is no alternative. Creative
+volume is not a production preference — it is a structural requirement for
+algorithmic efficiency.
+
+### Minimum Creative Inventory
+
+| Campaign phase | Minimum variants per ad set | Target variants |
+|---|---|---|
+| Launch | 4 | 6-8 |
+| Scaling | 6 | 8-12 |
+| Mature (90+ days) | 8 (continuous refresh) | 10-15 active |
+
+Launch below four variants. Scale campaigns without refresh cadence. Both
+conditions degrade algorithm efficiency. The launch threshold is a hard
+floor, not a suggestion.
+
+### Creative Architecture
+
+Within the minimum inventory, vary across three dimensions to give the
+algorithm independent signal axes:
+
+1. **Hook variation:** the first 3 seconds (Meta, TikTok) or first frame
+   (LinkedIn) must be varied across creatives. Identical hooks with varied
+   body copy produce near-zero differential signal — the algorithm makes
+   delivery decisions within the first interaction window.
+2. **Format variation:** include at least one static image, one video under
+   15 seconds, and one longer-form video (30-60 seconds) per ad set where
+   the platform supports all three. Format diversity ensures placement-level
+   efficiency across feed, story, and reels inventory.
+3. **Angle variation:** each creative should represent a distinct value-
+   proposition angle (price, outcome, social proof, problem-agitation, urgency)
+   — not the same angle executed in different visual treatments.
+
+### Creative Fatigue Detection
+
+Creative fatigue occurs when frequency accumulates on a delivered audience
+faster than the creative can sustain attention. Fatigue signals to monitor:
+
+| Signal | Threshold | Action |
+|---|---|---|
+| Frequency (7-day window) | Prospecting >2.5; Retargeting >5.0 | Introduce new creative variants |
+| CTR trend | >15% decline week-over-week for 2 consecutive weeks | Flag creative for replacement |
+| Hook rate (3-second view rate) | Drop below 25% on Meta/TikTok | Test new hook variants |
+| CPM trend | >20% increase without audience expansion | Audience saturation signal; expand or refresh |
+
+Do not wait for ROAS or CPA to degrade before addressing creative fatigue.
+CTR and hook rate are leading indicators; ROAS degradation is a lagging
+indicator that arrives after the creative-fatigue damage is already priced
+into the campaign's learned audience pool.
+
+### Refresh Cadence
+
+For prospecting campaigns at scale, introduce at minimum two to three new
+creative variants per ad set per month. For retargeting campaigns with
+smaller audiences, the frequency ceiling is reached faster — refresh at
+minimum every three weeks or when any fatigue signal threshold is crossed,
+whichever comes first.
+
+---
+
+## Bid and Budget Strategy
+
+Budget structure (CBO versus ABO) and bid strategy (bid cap versus cost
+cap versus lowest cost) are mechanical decisions with clear selection criteria.
+They are not interchangeable.
+
+### CBO vs ABO
+
+| Structure | When to use | Risk |
+|---|---|---|
+| Campaign Budget Optimisation (CBO) | 3+ ad sets with similar audience sizes; algorithm should allocate dynamically between them | Algorithm concentrates spend on the best-signal ad set, starving test cells |
+| Ad Set Budget (ABO) | Testing creative or audience variants where equal exposure is required; ABM lists; exclusion cells | Requires manual rebalancing as performance diverges |
+
+Testing requires ABO. Scale requires CBO. Mixing test objectives with
+CBO budget structure is the most common cause of false-positive test
+conclusions in paid social — the algorithm unbalances delivery before
+statistical significance is reached.
+
+### Bid Strategy Selection
+
+| Bid strategy | When to use | Constraint |
+|---|---|---|
+| Lowest cost (auto bid) | Learning phase; new campaigns without CPA history | No CPA control; spend as directed by budget |
+| Cost cap | CPA ceiling is defined; willing to sacrifice delivery volume to hold CPA | Algorithm may under-deliver if cap is set below the market-clearing price |
+| Bid cap | Need strict CPM/CPC ceiling; inventory-quality control | Frequently under-delivers; use only when delivery sacrifice is acceptable |
+| Value optimisation | Conversion-value schema is mapped; goal is revenue not volume | Requires minimum purchase event volume; SKAN schema must be configured |
+
+Do not set a cost cap below the observed CPA from the lowest-cost phase
+by more than 20% — the algorithm will be unable to clear the market at
+that price and under-delivery will result. Anchor cost caps to observed
+performance data, not to target unit economics in isolation.
+
+### Daily vs Lifetime Budgets
+
+Daily budgets offer day-to-day delivery control and are appropriate for
+always-on campaigns. Lifetime budgets allow the platform to pace delivery
+around predicted high-performance windows (ad scheduling, day-parting,
+auction dynamics) and are appropriate for time-bounded flights. Mixing
+daily and lifetime budgets across ad sets within the same CBO campaign
+creates undefined pacing interactions — use a single budget type within
+a CBO structure.
+
+---
+
+## Attribution Methodology
+
+Platform-reported attribution is structurally overstated. Every major
+paid social platform attributes conversions using the most favourable
+attribution window available unless reconfigured. Meta's default 7-day
+click + 1-day view window claims credit for conversions that occur up to
+seven days after a click and one day after a view, regardless of other
+touchpoints in between. No single platform has visibility into the full
+path; each claims the conversion within its own attribution logic.
+
+Attribution is a triangulation problem, not a single-source measurement
+problem. Relying on any single platform's reported ROAS or CPA as the
+primary performance signal leads to budget misallocation.
+
+### Triangulation Stack
+
+| Source | What it measures | Limitation |
+|---|---|---|
+| Platform-reported conversions | Algorithmic credit claims; directional signal for relative creative and audience performance | Overcounts; no cross-channel deduplication |
+| Lift study (holdout test) | Incrementality — conversions caused by the platform versus those that would have occurred anyway | Requires statistically sufficient audience and time window; cannot run continuously |
+| Marketing mix modelling (MMM) | Regression-based channel contribution estimate over historical spend | Lagging; requires 12+ months of data for reliable coefficients; not real-time |
+| GA4 / last-click analytics | User-reported session path; session-level attribution | Under-counts cross-device and view-through; over-weights last interaction |
+
+Use platform-reported data for day-to-day optimisation decisions (creative
+rotation, audience adjustments, bid management) — it is the highest-
+frequency signal available. Use lift studies quarterly to validate that
+platform spend is driving incremental conversions, not just claiming credit
+for organic demand. Use MMM annually to calibrate cross-channel budget
+allocation. Use GA4 to detect path anomalies and landing-page drop-off
+patterns, not as the attribution source of record.
+
+### Attribution Window Configuration
+
+Shorten platform attribution windows from defaults to reduce overcounting.
+Meta 7-day click + 1-day view is the recommended standard for ecommerce.
+LinkedIn default 30-day click window is appropriate for B2B lead gen given
+longer sales cycles. TikTok 7-day click is appropriate for direct response.
+Do not compare ROAS or CPA across platforms without normalising attribution
+windows — a Meta 28-day window result cannot be compared to a TikTok 7-day
+window result without adjustment.
+
+---
+
+## Anti-Patterns
+
+| Anti-pattern | Why it fails |
+|---|---|
+| Trusting platform-reported ROAS as absolute performance signal | Every platform overcounts; cross-platform ROAS comparison without normalised windows is meaningless |
+| Single creative variant per ad set | Forces random delivery; no algorithmic selection signal; fatigue is immediate at any meaningful frequency |
+| Changing campaign structure during the learning phase | Resets the learning phase; performance data collected before the reset is wasted; the campaign re-enters cold-start behaviour |
+| Setting cost caps below observed CPA before volume history exists | Algorithm cannot clear the auction at the cap price; under-delivery results; campaign never generates the CPA data needed to justify the cap |
+| Deploying identical creative across platforms | Each platform has a distinct native consumption context; cross-posting produces off-native creative that underperforms against platform-optimised formats |
+| Ignoring creative fatigue signals until ROAS declines | ROAS is a lagging indicator; by the time it degrades, the algorithm has already encoded a fatigued audience pool; recovery requires creative refresh plus audience reset |
+| Using CBO for creative tests | CBO budget concentrations on the best-signal cell before statistical significance; test conclusions are confounded by unequal delivery |
+| Treating Advantage-Plus as always superior | ML-driven structures require minimum data density; below the 50-conversion-per-week threshold, manual targeting delivers more consistent results |
+
+---
+
+## Cross-References
+
+- `domains/paid-media/skills/ppc-strategist` — search advertising; keyword
+  strategy and Quality Score mechanics; use when the campaign objective
+  maps to active search intent rather than passive social feed.
+- `domains/paid-media/skills/programmatic-buyer` — display and video
+  programmatic; DSP strategy, audience data marketplace, brand-safety
+  controls; use when the campaign requires open-web reach beyond walled
+  gardens.
+- `domains/paid-media/skills/creative-strategist` — creative concepting,
+  ad format design, UGC and native-style production briefs; use when
+  creative development is the primary deliverable rather than media strategy.
+- `domains/paid-media/skills/tracking-specialist` — pixel implementation,
+  Conversions API configuration, GA4 event schema, server-side tagging;
+  use when the task is measurement infrastructure rather than campaign
+  strategy.
+
+---
+
+## ADR Anchors
+
+- **ADR-058** — Brainstorm gate pre-Plan + two-pass adversarial review:
+  paid social campaign briefs require the same pre-work gate discipline
+  as technical plans. An objective-platform-creative triple confirmed
+  at brief stage is structurally equivalent to the ADR-058 brainstorm
+  gate applied to advertising work: it moves the constraint discovery
+  forward and prevents learning-phase resets caused by post-activation
+  structural changes.