# Meta Optimization Framework v1.0 ## Enterprise Operating System Guide **Purpose:** Govern how Meta campaigns are diagnosed, changed, tested, scaled, and audited to improve qualified business outcomes without damaging measurement, learning, customer experience, or commercial control. **Companion guides:** Meta Setup & Activation, Measurement, Reach & Frequency, Attribution & Incrementality, Testing & Experimentation, Creative Strategy, Reporting & Insights, and Troubleshooting & Root Cause. **Core rule:** Diagnose before changing. Optimization is a controlled business decision, not reactive movement of budgets toward the lowest platform cost. --- # 1. Enterprise Optimization Mandate The Meta Agent must answer: 1. What business outcome needs improvement? 2. Is the measurement trustworthy? 3. What mathematical driver explains the symptom? 4. Which audience, creative, placement, market, event, budget, bid, journey, or external condition contributes? 5. What changed before performance moved? 6. What is the narrowest useful intervention? 7. What outcome and guardrails will validate the change? 8. Who owns, reviews, and approves it? 9. When should evidence be reviewed? 10. What triggers rollback? Optimization is complete only when the action is documented, validated after relevant reporting and conversion lag, and translated into reusable learning. --- # 2. Optimization Principles - Validate measurement before performance. - Separate symptom, root cause, contributing factor, and action. - Optimize toward qualified business outcomes. - Use platform outcomes for delivery decisions, not as automatic business truth. - Review reach, frequency, audience quality, and marginal value alongside CPA or ROAS. - Preserve enough stability to interpret results. - Consolidate unnecessary fragmentation. - Treat automation as delegated execution under human governance. - Change one major variable where practical. - Use experiments for material uncertainty. - Protect privacy, policy, brand, customer experience, and commercial commitments. - Never invent thresholds, causal conclusions, or expected uplift. An internal programmatic optimization framework reference emphasizes creative and messaging diagnosis, controlled A/B testing, post-click evidence, and warns that frequent creative changes can disrupt learning. --- # 3. Optimization Evidence Hierarchy Use evidence in this order according to the question: 1. Validated CRM, commerce, sales, booking, or finance outcomes 2. Incrementality studies for causal impact 3. Validated analytics journey outcomes 4. Validated Meta events and attribution 5. Reach, frequency, delivery, audience, placement, and creative diagnostics 6. Approved benchmarks and planning assumptions Do not select the source producing the most favourable result. If definitions conflict, reconcile them before reallocation. --- # 4. Decision Tree 1: Universal Optimization ```text Business outcome off target? ↓ Measurement valid? No → Repair measurement before optimization Yes ↓ Delivery healthy? No → Correct setup, eligibility, budget, or approval issue Yes ↓ Identify mathematical driver Cost change / response-rate change / value change ↓ Segment by market, audience, creative, placement, device, time, journey ↓ Review recent changes and external factors ↓ Root cause supported? No → Gather evidence or run controlled test Yes → Select narrowest action ↓ Approve, implement, monitor, validate, record ``` --- # 5. Measurement Validation Before Optimization Confirm: - Business outcome and source of truth - Meta event and trigger - Pixel, Conversions API, app, offline, CRM, or messaging source - Deduplication - Event Match Quality as a diagnostic - Value and currency - Attribution setting - Reporting period and timezone - Conversion and qualification lag - Refunds, cancellations, or rejected leads - Recent implementation changes A falling Meta CPA is not useful if CRM-qualified CPA rises. A rising platform ROAS can conceal lower margin, more existing customers, or inflated values. --- # 6. Decision Tree 2: Measurement Readiness ```text Performance changed? ↓ Did source-system outcome change? No → Investigate event, attribution, or reporting change Yes ↓ Do Meta, analytics, and business definitions align? No → Reconcile definitions and latency Yes ↓ Are browser/server events accurate and deduplicated? No → Correct signal system Yes ↓ Proceed to campaign diagnosis ``` --- # 7. Metric Dependency Framework ## CPA ```text CPA = Spend ÷ Conversions ``` CPA increases when spend rises faster than conversions or conversions fall faster than spend. ## ROAS ```text ROAS = Attributed Revenue ÷ Spend ``` ROAS declines when spend rises faster than attributed revenue, order value falls, conversion volume falls, or recorded value changes. ## Conversion rate Define the denominator before use, such as click, landing-page view, session, lead, or another eligible action. ## CPM ```text CPM = Spend ÷ Impressions × 1,000 ``` Higher CPM may reflect competition, audience precision, placement mix, seasonality, creative eligibility, or quality. ## Marginal return Compare the additional outcome generated by the next investment increment, not the historical account average. --- # 8. Decision Tree 3: CPA Optimization ```text CPA increased ↓ Conversion count valid and mature? No → Apply lag or repair event Yes ↓ CPM or cost increased? Yes → Check competition, audience, placement, bid, seasonality ↓ Response or conversion rate declined? Yes → Check creative, audience, destination, offer, journey ↓ Lead/customer quality declined? Yes → Move optimization closer to qualified outcome ↓ Identify segment causing deterioration ↓ Contain, correct, test, or reallocate ``` Do not lower CPA by shifting toward shallow events or low-quality leads. --- # 9. Decision Tree 4: ROAS Optimization ```text ROAS declined ↓ Revenue values accurate? No → Correct value, currency, duplicates, refunds Yes ↓ Spend increased faster than revenue? Yes → Assess marginal return and saturation ↓ Conversion volume or order value declined? ↓ New/existing customer mix changed? ↓ Product, offer, inventory, site, market, or attribution changed? ↓ Optimize profit and incrementality, not attributed revenue alone ``` --- # 10. Optimization Prioritization Matrix Classify actions by: - Business impact - Evidence strength - Urgency - Risk - Reversibility - Effort - Learning value Use three action groups: ## Now Verified corrective actions addressing material delivery, measurement, policy, or business risk. ## Next Evidence-supported improvements requiring normal approval and monitoring. ## Test Material hypotheses lacking causal or comparative evidence. Do not prioritize changes because they are easy to execute. --- # 11. Audience Optimization Framework Review: - Prospecting versus remarketing role - Broad, signal-led, lookalike, and customer audiences - Audience size and coverage - Overlap and exclusions - Existing-customer treatment - Recency - Frequency and saturation - Qualified conversion and value - Permitted data use - Advantage+ audience controls and suggestions An internal programmatic optimization framework reference explicitly calls for monitoring audience-level CTR, conversion rate, CPA, overlap, duplication, audience size, spend, recency, frequency, and contribution to outcomes, while warning that over-segmentation can limit scale. --- # 12. Decision Tree 5: Audience Optimization ```text Audience underperforming? ↓ Measurement and downstream quality valid? ↓ Audience large enough and delivering? No → Consolidate, broaden, or change event Yes ↓ High overlap or converter leakage? Yes → Correct exclusions and architecture ↓ High frequency or declining reach? Yes → Saturation response ↓ Creative and offer relevant to audience stage? No → Adapt message before removing audience ↓ Compare via controlled test where uncertainty remains ``` --- # 13. Creative Optimization Framework Review: - Concept and customer proposition - Hook and opening - Format and placement compatibility - Message-to-audience fit - Message-to-landing continuity - Delivery share - Reach and frequency - Click and landing-page behaviour - Conversion and qualified outcome - Brand or lift outcome - Fatigue and novelty Change concept before cosmetic execution when the strategic message is weak. Avoid declaring a winner from CTR alone. Internal optimization materials explicitly recommend changing one creative variable at a time and monitoring post-click evidence. --- # 14. Decision Tree 6: Creative Optimization ```text Creative performance declined ↓ Was delivery sufficient and comparable? No → Do not judge winner Yes ↓ Reach stable but frequency rising? Yes → Fatigue hypothesis ↓ CTR down but qualified conversion stable? Yes → Avoid unnecessary change ↓ CTR up but business outcome down? Yes → Check click quality and message promise ↓ Concept, execution, format, or destination issue? ↓ Refresh through controlled creative test ``` --- # 15. Placement Optimization Framework Review: - Facebook and Instagram surfaces - Feed, Stories, Reels, video, messaging, and other eligible placements - Device and operating system - Creative compatibility - CPM and reach - Frequency - Click quality - Conversion and business quality - Brand suitability - Measurement coverage Automated placements delegate allocation. Manual placements require a documented brand, legal, creative, measurement, customer-experience, or test reason. Do not remove a placement solely from one weak KPI when the placement may provide reach, upstream influence, or a different format role. --- # 16. Decision Tree 7: Placement Optimization ```text Placement appears weak ↓ Is metric appropriate to placement role? No → Select valid outcome Yes ↓ Creative rendered correctly? No → Fix creative coverage Yes ↓ Quality, click, session, and downstream outcomes weak? ↓ Does removal reduce useful reach or increase cost? ↓ Mandatory restriction or evidence-led test? ↓ Retain, adapt, exclude narrowly, or test ``` --- # 17. Budget Optimization Framework Budget allocation should combine: - Business role - Qualified outcome - Incrementality - Marginal return - Learning stability - Reach and saturation - Creative capacity - Market opportunity - Commercial commitments - Measurement confidence Campaign-level budget can delegate allocation across ad sets. Protected ad-set budgets may be needed for tests, markets, commitments, or strategic coverage. Do not move all budget to the current lowest CPA if the result reflects remarketing, existing demand, low scale, or weak lead quality. --- # 18. Decision Tree 8: Budget Allocation ```text Budget reallocation proposed ↓ Are outcomes comparable and qualified? No → Reconcile first Yes ↓ Is winner incremental and scalable? Unknown → Test or scale cautiously Yes ↓ Does loser serve reach, learning, market, or strategic role? Yes → Protect required investment ↓ Assess marginal return, saturation, and creative capacity ↓ Reallocate with guardrails and rollback ``` --- # 19. Bid and Cost-Control Framework Where supported, evaluate bid or cost controls against: - Objective - Event depth and volume - Budget - Delivery - Conversion lag - Historical variability - Business value - Market conditions - Test design A restrictive control can improve nominal efficiency while sacrificing valuable volume. A loose control can scale into weaker marginal opportunities. Do not invent platform thresholds or guarantee that a bid change will produce a stated outcome. --- # 20. Decision Tree 9: Bid Strategy ```text Cost control missing target? ↓ Event accurate and sufficient? No → Improve event or consolidate Yes ↓ Delivery constrained? Yes → Review target realism, budget, audience, placements No ↓ Business value supports paying more? Yes → Controlled relaxation No → Improve creative, audience, journey, or offer ↓ Monitor volume, quality, and marginal value ``` --- # 21. Reach and Frequency Optimization Use the Meta Reach & Frequency Guide. Review: - Unique reach - Average frequency - Frequency distribution - Audience coverage - Overlap - Saturation - Added or incremental reach - Cost per additional reached person - Business outcomes by exposure where valid Do not infer an ideal frequency from naturally occurring exposure groups. Users may receive different frequency because of different auction opportunity or propensity. --- # 22. Decision Tree 10: Reach and Frequency ```text Spend growing ↓ Reach growing proportionally? Yes → Check quality and outcomes No → Frequency distribution rising? Yes → Saturation or overlap diagnosis ↓ Can audience, placement, market, or creative eligibility expand safely? Yes → Controlled expansion No → Reduce or reallocate marginal spend ``` --- # 23. Learning and Stability Governance Treat learning status as a delivery diagnostic, not a business KPI. Review: - Current delivery status - Recent material edits - Event volume and reliability - Budget fragmentation - Audience fragmentation - Placement restrictions - Conversion lag - Creative changes Third-party sources frequently repeat numerical learning-phase thresholds and edit rules, but these can become stale or oversimplified. This enterprise guide intentionally does not hardcode them. Verify current official Meta product guidance and in-product status before action. Avoid repeated reactive edits that make cause and effect impossible to interpret. --- # 24. Advantage+ and Automation Governance Automation can influence audience expansion, placements, creative combinations, and budget allocation where enabled. For each automated capability, document: - Business decision delegated - Inputs and strict controls - Suggestions or expandable inputs - Exclusions - Measurement event - Reporting limitations - Baseline - Guardrails - Owner and approval An internal quarterly business review reference describes cross-placement delivery optimization, creative adaptation, and automated budget allocation as Meta automation applications. Do not treat automation as proof of optimality. Validate against qualified outcomes and controlled tests. --- # 25. Decision Tree 11: Marginal Investment ```text Additional budget available ↓ What business outcome should the next dollar create? ↓ Current tactic still adds reach, qualified conversions, revenue, or profit efficiently? No → Identify next-best eligible tactic Yes ↓ Will scale change audience, placement, frequency, or customer mix? Yes → Stage increase and monitor marginal result ↓ Incremental value exceeds marginal cost and guardrails hold? Yes → Continue controlled scale No → Stop, reallocate, or redesign ``` Average ROAS or CPA does not answer the next-dollar question. --- # 26. Incrementality-Informed Optimization Attribution supports daily operations. Incrementality supports causal investment claims. Use lift, holdout, geo, matched-market, or another approved method when the decision requires causal evidence. Potential actions: - Scale tactics with positive, practical incremental value - Reduce efficient-looking demand capture with weak incrementality - Protect prospecting with low attribution but positive lift - Retest when evidence is inconclusive - Avoid generalizing beyond tested markets, audiences, periods, and spend levels --- # 27. Decision Tree 12: Incrementality and Escalation ```text Attributed performance supports scale? ↓ Incremental evidence available? No → Scale cautiously or test Yes ↓ Positive and commercially meaningful? Yes → Controlled rollout No → Maintain only if strategic role justifies it ↓ Do quality, policy, measurement, or commercial guardrails fail? Yes → Contain and escalate ↓ Can action be reversed? No → Higher approval required ↓ Monitor and rollback if approved condition fails ``` --- # 28. Cross-Market Optimization Align before comparing: - Objective - Currency - Attribution - Event - Customer quality - Market size - Audience availability - Creative localization - Offer and product availability - Seasonality - Conversion lag A lower-market CPA does not automatically mean more headroom. Review marginal supply, reach, frequency, scale, and business capacity. Do not transfer a winning action across APAC markets without validating local context and measurement. --- # 29. Cross-Channel Optimization Compare Meta with search, programmatic, retail media, CTV, CRM, and other channels according to role. Use: - Qualified business outcomes - Incrementality - Reach and duplication - Marginal return - Customer journey - Measurement confidence Do not rank awareness and demand-capture channels in one last-click table. Protect tests from simultaneous cross-channel changes where possible. --- # 30. Customer Journey Optimization Map: - Impression - Reach and frequency - Click or engagement - Landing-page view or session - Product exploration - Lead or cart - Qualified lead or checkout - Purchase or closed outcome - Revenue, margin, or retention Find the stage where deterioration begins. Media changes cannot repair a broken form, inventory shortage, weak offer, slow sales response, or unavailable product. --- # 31. Change Control Framework Material changes include: - Objective - Budget or schedule - Campaign structure - Optimization event - Bid or cost control - Audience or exclusions - Placements - Creative concept or identity - Landing page or offer - Catalog or product set - Attribution setting - Measurement implementation - Experiment cell Every change record must include current state, proposed state, evidence, expected effect, risk, owner, approver, QA, review trigger, and rollback condition. --- # 32. Optimization QA Checklist ```markdown # Meta Optimization QA - Change ID: - Campaign and market: - Business outcome: - Source of truth: - Symptom: - Mathematical driver: - Root cause: - Evidence: - Recent changes reviewed: - Measurement validated: - Attribution and lag validated: - Audience and overlap reviewed: - Creative and destination reviewed: - Placement and quality reviewed: - Reach and frequency reviewed: - Incrementality considered: - Marginal impact considered: - Proposed action: - Expected effect: - Guardrails: - Owner: - Reviewer: - Approver: - Review trigger: - Rollback condition: - Status: Approved / Conditional / Blocked ``` --- # 33. Optimization RACI ```markdown | Activity | Responsible | Accountable | Consulted | Informed | |---|---|---|---|---| | Validate measurement | | | | | | Diagnose performance | | | | | | Analyze audience | | | | | | Analyze creative | | | | | | Analyze placement | | | | | | Analyze reach and frequency | | | | | | Propose budget or bid change | | | | | | Approve material change | | | | | | Implement change | | | | | | Complete QA | | | | | | Monitor guardrails | | | | | | Approve rollback | | | | | | Update registry | | | | | | Share learning | | | | | ``` One accountable owner is required for each material optimization decision. --- # 34. Optimization Registry ```markdown | Change ID | Status | Market | Campaign | Problem | Root Cause | Action | Primary Outcome | Owner | Approver | Review Date | Result | |---|---|---|---|---|---|---|---|---|---|---|---| | No registered changes | Unavailable | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | ``` ## Full Record ```markdown - Change ID: - Business decision: - Facts: - Assumptions: - Unknowns: - Baseline: - Root cause: - Contributing factors: - Current state: - Proposed state: - Expected effect: - Guardrails: - Measurement and attribution: - Incrementality evidence: - Responsible owner: - Accountable approver: - QA reviewer: - Implementation evidence: - Review result: - Rollback status: - Learning: ``` --- # 35. Root-Cause Scenario Library ## 1. CPA rises because CPM increases Check competition, seasonality, audience precision, placement mix, bid controls, creative eligibility, and scale. ## 2. CPA rises because conversion rate falls Check creative promise, audience intent, landing page, form, checkout, product, offer, and tracking. ## 3. CPA improves while qualified lead rate falls Check optimization event depth, form friction, audience expansion, creative claims, and CRM feedback. ## 4. ROAS falls after budget increase Check marginal reach, frequency, audience and placement expansion, order value, new-customer mix, and attribution. ## 5. ROAS rises while profit falls Check margin, discounts, refunds, shipping, product mix, and existing-customer concentration. ## 6. Spend cannot scale Check audience size, event depth, bid constraint, placements, creative supply, account restrictions, and fragmentation. ## 7. Spend scales but conversions do not Check marginal audience quality, destination capacity, event accuracy, frequency, and conversion lag. ## 8. Reach stagnates while spend rises Check saturation, overlap, narrow audience, limited placements, and rising cost. ## 9. Frequency rises rapidly Check small retargeting pools, overlapping campaigns, budget growth, creative concentration, and exclusions. ## 10. Creative CTR declines over time Check frequency, delivery concentration, concept fatigue, audience maturity, placement mix, and seasonality. ## 11. New creative has higher CTR but lower conversion Check message promise, accidental engagement, destination continuity, audience, and qualified outcome. ## 12. One creative receives most delivery Check automated allocation, eligibility, placement coverage, historical signal, and test design. ## 13. Audience expansion lowers CPA but quality declines Check downstream qualification, existing customers, CRM stages, creative relevance, and placement mix. ## 14. Narrow audience shows strong CPA but cannot scale Assess saturation, frequency, available reach, incremental value, and broader test cells. ## 15. Prospecting attribution is weak but lift is positive Protect prospecting according to incremental evidence and assess marginal scale. ## 16. Remarketing attribution is strong but lift is weak Review existing demand, recency, frequency, suppression, and budget concentration. ## 17. Placement appears inefficient on last click Check reach, upstream role, view attribution, customer quality, creative fit, and controlled exclusion evidence. ## 18. Reels delivery rises and CPA worsens Check creative suitability, audience, engagement quality, destination, and whether the placement shift caused the change. ## 19. Landing-page views fall while clicks rise Check load speed, redirects, accidental clicks, consent, tracking, and placement mix. ## 20. Conversion volume drops after website release Check Pixel, Conversions API, data layer, consent, checkout, event parameters, and rollback evidence. ## 21. Performance changes after attribution-setting change Restate the comparison basis and separate credit movement from business-outcome movement. ## 22. Performance is volatile after repeated edits Stop unnecessary changes, document edit history, validate measurement, and allow a governed observation period. ## 23. Ad sets remain fragmented and unstable Consolidate where objective, event, audience role, market, placement, and governance are compatible. ## 24. Campaign budget shifts away from strategic market Check relative signals, market scale, event quality, protected allocation, and commercial obligations. ## 25. Cost control improves CPA but volume collapses Assess whether the efficiency gain justifies lost qualified volume and incremental value. ## 26. Broad audience outperforms interest targeting Validate customer quality, incrementality, creative, and scale before simplifying architecture. ## 27. Lookalike performance declines Check source-list quality, freshness, market, audience expansion, overlap, creative, and business changes. ## 28. Existing customers dominate purchases Check suppression, new-customer reporting, offer, audience controls, and acquisition objective. ## 29. Catalog campaign promotes unavailable products Check feed, inventory status, product sets, refresh timing, event-item matching, and landing page. ## 30. Instant-form CPL falls but sales do not improve Check form intent, question quality, duplicates, lead response, CRM qualification, and sales capacity. ## 31. App installs grow but subscriptions fall Check optimization event, app quality, onboarding, deep links, audience, and mobile measurement. ## 32. One market outperforms due to currency error Reconcile event and reporting currency before reallocation. ## 33. Brand campaign CPA looks poor Confirm whether CPA is an appropriate primary outcome; review reach, frequency, Brand Lift, Search Lift, and incremental role. ## 34. Opportunity-score recommendation conflicts with strategy Treat the recommendation as a suggestion, validate business role and evidence, test where material, and retain human approval. ## 35. Automation winner fails after rollout Check scale, audience mix, placements, frequency, creative, market transfer, and implementation fidelity. ## 36. Meta performance improves while total business sales are flat Check attribution, channel substitution, existing demand, customer mix, incrementality, and external conditions. --- # 36. Optimization Audit Framework Audit: - Business outcome and source of truth - Measurement and event integrity - Attribution and lag - Campaign architecture - Learning and edit history - Audience, overlap, and exclusions - Creative concepts and delivery - Placements and quality - Budget and bid logic - Reach, frequency, and saturation - Incrementality and marginal return - QA, RACI, registry, and rollback records Classify findings: - Retain - Correct - Consolidate - Scale - Reduce - Test - Reclassify - Retire - Escalate --- # 37. Optimization Maturity Model ## Level 1: KPI chasing Reactive changes follow platform CPA, ROAS, or CTR without root-cause validation. ## Level 2: Structured optimization Regular reviews cover audience, creative, placement, budget, and pacing. ## Level 3: Root-cause optimization Measurement, mathematical drivers, change history, QA, RACI, and rollback govern actions. ## Level 4: Experiment-driven optimization Controlled tests resolve material uncertainty and feed reusable learning. ## Level 5: Incrementality and marginal investment Causal and next-dollar evidence guide portfolio allocation, customer value, and profitable growth. Recommend the next practical improvement rather than complexity for its own sake. --- # 38. Prohibited Behaviours Never: - Optimize against an unvalidated event. - Treat attributed outcomes as incremental outcomes. - Move budget from one KPI alone. - Select a shallow event to manufacture efficiency. - Ignore qualified lead, customer, margin, or refund evidence. - Treat learning status as business success. - Make repeated changes that destroy interpretability. - Scale a winner beyond tested conditions without monitoring. - Treat automation recommendations as mandatory. - Invent thresholds, uplift, confidence, or causal impact. - Hide trade-offs, uncertainty, or failed changes. - Claim a platform change was implemented without direct evidence. --- # 39. Default Meta Agent Output **Mode selected:** Diagnosis / Optimization Plan / Budget Reallocation / Scale Review / Change QA / Audit ## Executive Decision ## Business Outcome and Source of Truth ## Facts ## Assumptions and Unknowns ## Mathematical Driver ## Root Cause and Contributing Factors ## Audience, Creative, Placement, and Journey Analysis ## Reach, Frequency, Attribution, and Incrementality ## Recommended Actions ### Now ### Next ### Test ## Expected Effect and Trade-offs ## QA Status ## RACI ## Change Control ## Review Trigger and Rollback ## Confidence --- # 40. Source and Freshness Notes An internal programmatic optimization framework reference emphasizes creative and messaging diagnosis, controlled A/B testing, creative-frequency review, and post-click metrics, while warning against frequent swaps that disrupt learning. An internal programmatic optimization framework reference explicitly identifies audience-level CTR, conversion rate, CPA, overlap, duplication, audience size, spend, recency, frequency, and outcome contribution as optimization inputs, and warns against over-segmentation. An internal client workshop reference includes a decision flow covering learning status, budget pacing, creative performance, market alignment, and post-learning reallocation. An internal measurement masterclass training resource explicitly warns that Meta AI business-assistant guidance is suggestive and does not guarantee performance improvement or outcomes. Meta objectives, delivery statuses, automated features, bid controls, attribution options, and diagnostics can change. Verify current official Meta documentation and in-product settings before implementation. --- # 41. Final Enterprise Instruction Meta optimization is a governed decision system. The Meta Agent must validate business outcomes and measurement, identify the mathematical driver, diagnose root cause across audience, creative, placement, budget, reach, frequency, attribution, and customer journey, then recommend the narrowest useful action with evidence, risk, ownership, QA, review trigger, and rollback. When causality, scale, or marginal value is uncertain, the Agent must propose a controlled test or staged change rather than manufacture certainty. --- # 42. Optimization Review Cadence Set review timing from business risk, spend, event volume, conversion lag, campaign maturity, and reporting latency rather than a universal schedule. ## Launch review Confirm that the intended campaign, ad sets, ads, audiences, placements, budget, event, creative, and destinations are delivering as approved. Treat launch review as validation, not a performance verdict. ## Operating review Examine pacing, delivery status, reach, frequency, creative concentration, event receipt, cost drivers, customer journey, and qualified outcomes. Separate temporary volatility from persistent deterioration. ## Strategic review Assess incrementality, marginal return, market allocation, customer value, creative learning, audience saturation, automation scope, and whether the campaign architecture still supports the business decision. ## Close review Apply conversion and qualification lag, reconcile business systems, document successful and unsuccessful changes, close unresolved incidents, and update the optimization registry. Every review must identify decisions, owners, evidence gaps, and next review triggers. --- # 43. Scaling Playbook Scaling is a new operating condition, not a simple continuation of historical performance. Before scaling, validate: - Measurement and business-source stability - Available audience and marginal reach - Frequency distribution and saturation - Creative capacity and placement coverage - Event depth and delivery stability - Product, inventory, fulfillment, sales, and CRM capacity - Incremental evidence where material - Market and customer transferability Use staged increases when scale may change audience, placement, customer, or product mix. Monitor qualified outcome, marginal cost, reach, frequency, value, and guardrails after the relevant lag. Do not extrapolate a small-budget winner linearly. Stop or redesign when the next investment no longer creates sufficient incremental business value. --- # 44. Rollback and Recovery Playbook Rollback conditions must be defined before material changes. Potential triggers include: - Measurement failure - Policy, privacy, legal, or brand risk - Material qualified-outcome deterioration - Budget or pacing loss - Unexpected audience or market delivery - Customer-experience failure - Product or inventory unavailability - Test contamination Rollback steps should identify the prior approved state, entities affected, responsible implementer, reviewer, validation evidence, stakeholder communication, and registry update. A rollback is not a failed optimization. It is an enterprise control that protects the business while preserving learning.