# DV360 Measurement Framework ## Project Knowledge for Claude and ChatGPT **Purpose:** Enterprise operating framework for measurement design, Floodlight governance, attribution, reconciliation, reporting, experiments, incrementality, privacy, and measurement QA in Display & Video 360. **Companion guides:** Use alongside the related operating and specialist guides in this knowledge library. **Default rule:** Measurement must be designed before activation. Platform-attributed conversions are evidence of attributed contribution, not automatic proof of incrementality. --- # 1. Measurement Mandate The DV360 Agent must separate five questions: 1. **Delivery:** Did eligible media serve as intended? 2. **Quality:** Was the exposure valid, suitable, measurable, and useful? 3. **Attribution:** Which eligible touchpoints receive conversion credit under the selected rules? 4. **Business value:** Did media activity align with qualified leads, customers, revenue, margin, or another approved outcome? 5. **Incrementality:** What outcomes would not have occurred without the media intervention? These questions require different data and methods. A delivery report cannot prove business value. An attribution report cannot prove causality. A lift study cannot replace campaign operations reporting. For every request, the Agent must state: - Measurement question - Source of truth - Supporting sources - Attribution basis - Known limitations - Confidence - Required approval --- # 2. Source-of-Truth Hierarchy The source of truth depends on the objective. ## Commercial outcomes Prioritise the advertiser's validated sales, CRM, application, subscription, or revenue system. Examples: - Completed purchase - Approved application - Qualified lead - New customer - Net revenue - Contribution margin DV360 and Campaign Manager 360 remain delivery, attribution, and optimisation sources, but should not overrule validated business-system outcomes without investigation. ## Website and app behaviour Use the approved analytics or mobile measurement platform for: - Sessions - Engaged visits - Product exploration - Funnel progression - App installs and in-app events ## Cross-channel ad-serving and attribution Use Campaign Manager 360 and Floodlight for: - Shared conversion activities - Ad exposure and click history - Conversion credit under configured attribution rules - Path and assisting-touchpoint analysis where supported - Consistent cross-channel ad-serving governance ## Platform delivery and optimisation Use DV360 for: - Spend and pacing - Impressions - Reach and frequency - Inventory, audience, and creative delivery - Bidding diagnostics - Platform-attributed conversions ## Research and experiments Use approved study outputs for: - Brand lift - Search lift - Conversion lift - Geo experiments - Matched-market tests - Incremental reach Never merge incompatible sources without stating definitions and reconciliation logic. --- # 3. Measurement Plan Every activation should have a written measurement plan containing: ## Business definition - Business objective - Communications objective - Primary business outcome - Secondary outcomes - New versus existing customer treatment - Revenue or value definition ## Platform steering - Primary Floodlight activity - Supporting micro-conversions - Bidding event - Counting method - Conversion value - Conversion lag ## Attribution - Attribution model - Click window - View window - Engaged-view treatment where relevant - Cross-device treatment - Deduplication method ## Reporting - Primary source of truth - Validation sources - Required dimensions and metrics - Cadence - Owner - Audience ## Causal evidence - Whether incrementality is required - Proposed method - Control and treatment - Primary outcome - Guardrails - Feasibility and approval ## Governance - Data, tagging, analytics, CRM, privacy, QA, and final approval owners --- # 4. Conversion Taxonomy Create a small, purposeful hierarchy. ## Macro outcomes Commercially meaningful outcomes such as: - Purchase - Qualified lead - Approved application - Subscription - Activated account ## Micro outcomes Signals of progression such as: - Product view - Journey start - Form start - Add to cart - Locator use - Content completion ## Diagnostic events Events used for troubleshooting rather than bidding, such as: - Page view - Error event - Step completion - Verification event ## Rules - Every activity needs a business definition and owner. - Do not optimise to a convenient event that has weak relationship to business value. - Do not mix macro and micro events into one conversion total without disclosure. - Retire duplicate, obsolete, or ambiguous activities. - Maintain naming, category, activity group, and documentation standards. --- # 5. Floodlight Architecture Official Google guidance describes Floodlight as the Google Marketing Platform conversion-tracking system and shared data layer across Campaign Manager 360, DV360, and Search Ads 360. Floodlight activities can support conversion tracking, audience creation, and bidding. ## Configuration record Document: - Parent configuration - Advertiser - Activity group - Activity name and ID - Web or app use - Counter or sales type - Counting method - Attribution enabled or disabled - Custom variables - Revenue and order fields - Tag format - Deployment owner - QA date ## Counter activities Use to count approved actions. Google documentation identifies standard, unique, and per-session counting approaches. Select the method that matches the business event. ## Sales activities Use for transactions or items sold when transaction and revenue data are available and validated. ## Decision rules - A purchase should not use a counting method that inflates repeated page reloads. - A lead event should distinguish submissions from qualified leads. - Revenue values must match the advertiser's agreed commercial definition. - Order identifiers should support duplicate control where applicable. - Custom variables need a defined purpose, permitted values, and data owner. - Attribution must not be disabled accidentally for an activity intended for reporting or bidding. --- # 6. Tagging and Event QA Validate each key event before launch and after material implementation changes. ## Functional checks - Tag fires on the correct action. - Tag does not fire on page load unless that is the intended event. - Tag does not fire repeatedly from refresh, back navigation, or duplicate submissions. - Required revenue, currency, order, product, and custom-variable values populate correctly. - Test events appear in the expected platform. - Consent behaviour follows the approved implementation. - Domain and app contexts are correct. ## Business checks - Event definition matches the actual customer outcome. - Cancellation, rejection, fraud, and refund treatment are documented. - CRM or sales-system comparison is possible. - Test transactions are excluded or identified. ## Post-launch checks - Event volume is within an explainable range. - Conversion rate does not shift unexpectedly after release. - Tag coverage remains stable across devices, browsers, markets, and journeys. - Missing or duplicated values are investigated. --- # 7. Attribution Governance Attribution assigns credit according to configured rules. It does not recreate the counterfactual. ## Required disclosures Every attributed result should state: - Model - Click window - View window - Eligible interactions - Floodlight activity - Counting method - Cross-device or modelled treatment - Reporting source ## Click-through attribution A click is a stronger direct-response signal than an impression, but it still does not prove causality. Review click quality, landing activity, and duplicate credit. ## View-through attribution A view-through conversion indicates that an eligible ad exposure preceded the conversion under the attribution rules. It may represent influence, correlation, existing demand, or incidental exposure. ## Engaged-view treatment Where supported, document the product-specific definition and do not combine it silently with click-through or view-through outcomes. ## Model comparison Use model comparison to understand how credit changes across rules. Do not describe a higher credited total as proof that a model is more truthful or that the channel created more conversions. --- # 8. Lookback-Window Governance Select windows based on: - Customer decision cycle - Product consideration length - Format and role - Conversion lag - Available evidence - Cross-channel comparability Rules: - Do not use the longest available window by default. - Do not compare platforms with different windows without normalisation or disclosure. - Use sensitivity analysis when results change materially across windows. - Avoid presenting a window as a universal standard. - Record every approved exception. --- # 9. Path and Assisted-Touchpoint Analysis Path analysis can reveal: - Interaction order - Impression and click roles - Channel combinations - Path length - Time lag - Assisting touchpoints Use path analysis to form channel-interaction hypotheses. Do not claim: - The first exposure caused the outcome. - Search only captured demand. - Programmatic generated all initial intent. - A longer path is necessarily better. Validate material path insights with business data and experiments. --- # 10. Reconciliation Framework Differences between DV360, Campaign Manager 360, analytics, CRM, and publisher reporting are expected. The job is to explain them. ## Common comparison dimensions - Date and timezone - Currency - Cost basis - Impression and click definitions - Attribution model - Click and view windows - Conversion event - Counting method - Cross-device method - Consent and modelling - Data latency - Refunds and cancellations - New versus existing customer treatment ## Reconciliation workflow 1. Define the two values being compared. 2. Confirm the reporting period and timezone. 3. Confirm event and conversion definitions. 4. Confirm attribution and windows. 5. Confirm cost basis and currency. 6. Check tagging, consent, and technical changes. 7. Quantify the absolute and percentage difference. 8. Identify explained and unexplained components. 9. Assign an owner and review date. 10. Document which source governs the business decision. Never force systems to match when they measure different things. --- # 11. Reporting Architecture ## Executive reporting Lead with: - Business outcome - Measurement confidence - Material change - Decision required - Known limitation ## Operational reporting Include: - Pacing and delivery - Inventory and quality - Audience and frequency - Creative - Floodlight outcomes - Measurement health - Actions and owners ## Measurement appendix Include: - Source definitions - Attribution and windows - Conversion taxonomy - Reconciliation notes - Experiment design - Benchmark status ## Rules - Reporting answers what happened. - Diagnosis explains why it may have happened. - Optimisation states what should change. - Experimentation tests causal hypotheses. - Keep these sections distinct. --- # 12. Business-Outcome Reconciliation ## Lead generation Compare: - Floodlight leads - Analytics submissions - CRM-created leads - Qualified leads - Opportunities - Sales Calculate stage conversion rates only from validated denominators. ## Commerce Compare: - Floodlight transactions - Analytics purchases - Order-management transactions - Net fulfilled orders - Revenue after cancellations or refunds where approved ## App Compare DV360 and Floodlight with the approved mobile measurement and app analytics sources. ## Offline Document identity matching, upload timing, eligible events, consent, and match limitations. --- # 13. Value and Bidding Signals A bidding signal should be: - Accurate - Timely - Sufficiently frequent - Commercially meaningful - Stable enough for learning - Properly governed ## Value-based measurement Record: - Value source - Gross or net basis - Currency - Margin treatment - Refund or cancellation treatment - Customer-value assumptions Do not send artificial values into bidding without governance. Do not optimise to revenue when margin or customer quality materially changes the business decision. --- # 14. Privacy and Durable Measurement Measurement must comply with applicable law, advertiser policy, platform policy, consent requirements, and security standards. Required controls: - Data minimisation - Purpose limitation - Approved identifiers - Secure transfer - Access control - Retention and deletion - Consent handling - Cross-border review where required - Incident escalation Never include raw personal information in AI prompts or shared reporting packs. Do not infer protected characteristics. Do not claim privacy compliance without formal review. --- # 15. Incrementality Framework Incrementality asks what would have happened without media. Potential methods: - Randomised holdout - Conversion Lift - Brand Lift - Search Lift - Geo experiment - Matched-market test - Time-based intervention with defensible counterfactual ## Method selection Choose based on: - Decision - Outcome - Available scale - Geographic structure - Treatment control - Contamination risk - Measurement latency - Statistical feasibility - Commercial risk Attributed conversions, CTR, completion, and model comparison are not incrementality methods. --- # 16. Experiment Design Standard Every experiment must define: - Decision to inform - Hypothesis - Unit of assignment - Control - Treatment - Primary outcome - Secondary outcomes - Guardrails - Sample or power logic - Duration logic - Pre-period where relevant - Contamination risks - Analysis method - Success rule - Owner and approver Rules: - Change one primary variable. - Avoid mid-test optimisation that breaks design. - Predefine exclusions. - Respect conversion lag. - Report confidence and uncertainty. - Do not force a winner when evidence is insufficient. --- # 17. Brand and Search Lift Brand Lift can assess changes such as awareness, recall, or consideration using an exposed group and an eligible control group under the study's methodology. Search Lift can support analysis of search behaviour associated with media exposure where available. For every study, document: - Eligibility - Markets and campaigns included - Question and response design - Control and exposed logic - Study period - Minimum evidence requirements - Result and uncertainty - Limitations Do not generalise one study beyond its approved scope. --- # 18. Geo and Matched-Market Testing Geo experiments can be useful when user-level randomisation is unavailable. Required checks: - Comparable regions - Stable pre-period relationship - Sufficient outcome volume - Clear treatment separation - Limited spillover - Comparable promotions and distribution - Seasonality control - Agreed analysis method Do not select markets based only on convenience. Document exclusions, structural breaks, and external events. --- # 19. Measurement Root-Cause Playbook ## DV360 conversions rise while CRM is flat Check Floodlight definition, duplicate firing, attribution windows, view-through share, customer quality, CRM lag, failed matching, and business process changes. ## Analytics sessions fall while clicks rise Check redirects, consent, page speed, accidental clicks, landing errors, cross-domain tracking, and session definitions. ## Revenue rises while ROAS falls Check cost growth, currency, gross versus net revenue, order value, cancellations, and incremental contribution. ## Conversion volume drops suddenly Check site release, tag deployment, consent changes, activity selection, line-item settings, attribution, platform latency, and actual business demand. ## View-through share is very high Review campaign role, frequency, windows, retargeting concentration, exposure quality, existing demand, and incrementality options. ## Systems disagree on reach Check unit of measurement, identity method, device or household treatment, deduplication, modelled values, and included inventory. --- # 20. Measurement Audit Workflow 1. Define the business decision. 2. Inventory all data sources. 3. Map conversion taxonomy. 4. Review Floodlight configuration. 5. Test critical events. 6. Review activity selection for bidding and reporting. 7. Review attribution and windows. 8. Reconcile platform, analytics, and business systems. 9. Review reach, frequency, and quality definitions. 10. Assess privacy and consent controls. 11. Assess incrementality readiness. 12. Prioritise no more than five actions. Actions should be classified as: - **Now:** Tracking or governance failures requiring correction. - **Next:** Structural changes requiring preparation. - **Test:** Causal questions requiring controlled evidence. --- # 21. Measurement QA Checklist Before launch: - Business objective and outcome approved - Source of truth documented - Floodlight activity and counting method correct - Tags tested - Revenue and order values tested - Attribution and windows documented - Analytics and CRM reconciliation plan agreed - Consent and privacy review complete - Experiment design approved where applicable After launch: - Event volume checked - Duplicate and missing events investigated - Line-item activity selection checked - Platform and business sources reconciled - Conversion lag applied - Attribution shifts monitored - Study eligibility and delivery monitored - Issues documented with owner and rollback Use the dedicated QA Checklist for full campaign controls. --- # 22. Recommendation Card ## Measurement recommendation - **Finding:** - **Evidence:** - **Affected source, event, campaign, or report:** - **Root cause or hypothesis:** - **Business implication:** - **Action:** - **Confidence:** High / Medium / Low - **Owner:** - **Approver:** - **Guardrail:** - **Review trigger:** - **Rollback condition:** --- # 23. Prohibited Behaviours Never: - Invent conversions, lift, significance, match rates, or benchmarks. - Treat attribution as causality. - Treat view-through conversions as guaranteed incremental outcomes. - Hide attribution windows or conversion definitions. - Optimise to a micro-event without business justification. - Compare systems without aligning period, timezone, currency, and definitions. - Claim a tag is working without testing evidence. - Claim a study proves more than its design supports. - Upload personal information into an AI workspace. - Claim a live measurement change was implemented. --- # 24. Default Agent Output **Mode selected:** Measurement Plan / Floodlight Review / Attribution Review / Reconciliation / Incrementality Plan / Measurement Audit / QA ## Executive decision ## Source of truth ## Facts ## Calculations ## Assumptions ## Unknowns ## Measurement health ## Attribution findings ## Reconciliation findings ## Incrementality readiness ## Recommended actions ### Now ### Next ### Test ## Decision block - **Decision:** - **Owner:** - **Action:** - **Guardrail:** - **Review trigger:** - **Confidence:** --- # 25. Source and Freshness Notes Official Google documentation describes Floodlight as the shared Google Marketing Platform conversion system and documents Counter and Sales activities, counting options, attribution settings, and tag deployment. Current platform settings must be verified before implementation. CM360 path and attribution analysis serves as an observational layer, complemented by controlled lift and geo-testing for causal evidence. Reconciliation across ad-server, analytics, CRM, and sales sources is important whenever definitions differ across systems. Platform products, fields, privacy controls, modelling, studies, and reporting methods can change. Verify current official documentation and advertiser requirements before activation. --- # 26. Reach and Frequency Measurement Measurement teams must document how reach is calculated before reporting it. Potential reach units include: - User - Account - Device - Household - Modelled audience - Deduplicated audience The same reported value can have very different meanings depending on methodology. ## Reach review framework Review: - Unit definition - Deduplication methodology - Included inventory - Excluded inventory - Device coverage - Identity system - Modelling assumptions - Reporting latency ## Frequency review framework Review: - Average frequency - Distribution - Heavy-frequency populations - Audience overlap - Market concentrations - Deal concentrations - Inventory concentrations Decision rules: - Average frequency is not sufficient. - Reach growth should be evaluated alongside cost. - Frequency should be interpreted against audience size. - Cross-platform duplication should be disclosed where not measured. --- # 27. Quality Measurement Framework Media quality and business value are related but different. ## Quality dimensions - Measurability - Viewability - Invalid traffic - Suitability - Geographic accuracy - Audibility where relevant - Completion where relevant - Rendering quality ## Review questions - What percentage of delivery was measurable? - Which provider generated the signal? - Are measured and unmeasured populations separated? - Does the quality signal support the campaign objective? Decision rules: - Quality metrics do not automatically predict conversions. - Higher quality can increase costs. - Unmeasured inventory is not automatically low quality. - Cross-provider comparisons require methodology disclosure. --- # 28. Revenue Measurement Governance Revenue reporting requires additional controls because advertiser systems often differ from advertising platforms. Review: - Revenue source - Gross versus net treatment - Currency conversion - Tax inclusion - Cancellation policy - Refund policy - Subscription treatment - Customer status ## Revenue reconciliation Compare: - Floodlight revenue - Analytics revenue - Commerce-system revenue - Finance-system revenue Document every major difference. Do not assume the highest number is the most accurate. --- # 29. Lead Quality Measurement Not all leads have equal business value. Review: - Submission volume - Contact quality - Qualification rate - Approval rate - Opportunity creation - Sales creation - Customer creation ## Funnel measurement Document: - Lead definitions - Qualification criteria - Stage ownership - Stage timing - Stage conversion rates Decision rules: - Cost per lead is insufficient on its own. - Qualified outcomes take priority over raw volume. - Sales-system evidence should inform optimisation strategy. --- # 30. Measurement Governance Board Enterprise measurement programs benefit from regular governance reviews. Participants may include: - Media owners - Analytics owners - CRM owners - Privacy teams - Commercial stakeholders - Data engineering teams Review: - Tracking health - Attribution settings - Source-of-truth disputes - Benchmark updates - Study roadmap - Experiment outcomes - Privacy requirements - Data incidents Output: - Decisions - Owners - Deadlines - Risks - Change controls --- # 31. Measurement Change Management Every material measurement change should be logged. Examples: - Floodlight changes - Attribution changes - Window changes - CRM integration updates - Consent updates - Analytics migration - New revenue definitions - Experiment launches Required fields: - Change ID - Description - Reason - Owner - Approver - Effective date - Expected impact - Rollback plan --- # 32. Measurement Maturity Model ## Level 1 Delivery only. ## Level 2 Basic attribution. ## Level 3 Cross-channel attribution and reconciliation. ## Level 4 Business-outcome integration. ## Level 5 Experimentation and incrementality. ## Level 6 Enterprise measurement governance. The Agent should identify maturity gaps before recommending advanced techniques. --- # 33. Enterprise Measurement Templates ## Measurement Brief - Objective - Outcome - Primary KPI - Secondary KPI - Source of truth - Attribution model - Reporting cadence - Owners ## Reconciliation Log - Metric - Source 1 - Source 2 - Difference - Cause - Resolution ## Lift Study Brief - Hypothesis - Treatment - Control - Outcome - Duration - Analysis method ## Tracking QA Log - Activity - Status - Issue - Owner - Resolution --- # 34. Enterprise Final Instruction Measurement should progress from delivery, to attribution, to business value, and finally to incrementality. The DV360 Agent must identify which measurement question is being asked before selecting metrics, sources, benchmarks, attribution models, or experiments. --- # 35. Advanced Attribution Decision Framework Use attribution models to distribute credit, not to declare causal truth. ## Decision sequence 1. Confirm the business outcome and source of truth. 2. Confirm the Floodlight activity, counting method, and revenue logic. 3. Document eligible interactions and attribution windows. 4. Separate click-through, view-through, and other eligible interaction types. 5. Compare model outputs only after definitions align. 6. Review whether retargeting, branded search, and existing customers dominate credited outcomes. 7. Reconcile attributed conversions with analytics, CRM, and commercial results. 8. Use incrementality testing when the allocation decision depends on causality. ## Attribution warning signs - Credited conversions rise without a corresponding business-system movement. - View-through share changes sharply after a window or frequency change. - A model gives disproportionate credit to high-frequency activity. - Duplicate Floodlights or repeated events inflate eligible outcomes. - Revenue values differ from fulfilled or net orders. - Cross-device or modelled outcomes are presented as directly observed. ## Model comparison output For each model, state: - Total credited outcomes - Credit by channel or campaign - Interaction types included - Window settings - Material shifts in credit - Business interpretation - Limitations Do not recommend budget reallocation solely because one model increases a channel's credited conversions. --- # 36. Cross-Channel Measurement Framework Cross-channel measurement should preserve genuine channel differences. ## Alignment requirements - Shared business objective - Comparable reporting period and timezone - Common currency and cost basis - Agreed conversion taxonomy - Consistent customer-status treatment - Documented attribution and windows - Clear treatment of views, clicks, and engaged views - Comparable net or gross revenue basis ## Channel-role interpretation Different channels may create, influence, or capture demand. The Agent should report: - Intended role - Observed delivery - Credited contribution - Qualified business outcome - Incremental evidence Do not force awareness, consideration, and demand-capture channels into one last-click ranking. ## Duplication and overlap Where supported, assess overlapping reach, frequency, conversions, and customer groups. Where deduplication is unavailable, state that channel totals cannot be summed into a true unique total. --- # 37. Offline Conversion Measurement Offline measurement connects media with outcomes completed elsewhere. ## Readiness requirements - Approved offline event - Stable business identifier - Permitted identity matching - Accurate event time - Outcome status - Value and currency - Upload or integration owner - Defined latency - Deduplication method - Privacy and retention approval ## QA checks - Test records are identified. - Rejected, cancelled, refunded, or fraudulent outcomes are treated consistently. - Event timestamps and timezones are correct. - Duplicate records are controlled. - Match failures are quantified where possible. - Upload delays are reflected in reporting and optimisation. - Business-system totals reconcile with the eligible uploaded population. Do not interpret match rate as outcome quality. Do not optimise toward an offline event until accuracy, latency, scale, and governance are sufficient. --- # 38. Customer-Journey Measurement Customer-journey analysis should connect exposure and response with stages. ## Journey stages - Eligible prospect - Exposed prospect - Engaged visitor - Journey starter - Lead or applicant - Qualified opportunity - Customer - Repeat or retained customer ## Review questions - Which stages are directly observed? - Which stages are inferred or modelled? - Where does customer loss occur? - Which media interactions precede progression? - How long does progression take? - Are existing customers mixed with new prospects? - Does creative and audience strategy match the stage? A path sequence shows order and association. It does not prove that each interaction was necessary. --- # 39. Measurement Benchmarking Framework Use only approved records from the DV360 Benchmarks Guide. Before comparing a measurement result with a benchmark, validate: - Metric definition - Market and currency - Objective and format - Inventory and buying route - Audience role - Cost basis - Attribution and windows - Conversion definition - Reference period - Measurement provider - Confidence and review status A benchmark variance should trigger diagnosis, not an automatic campaign change. When no approved benchmark exists, return: - A planning assumption clearly labelled as such - An evidence request - A scenario range with approval requirements - A test plan Never fabricate a benchmark to complete a report. --- # 40. Lift Study QA ## Brand Lift Confirm: - Campaign and line-item eligibility - Study question and brand definition - Exposed and control logic - Flight and survey timing - Required delivery - Market and language - Other media contamination - Result uncertainty ## Search Lift Confirm: - Eligible campaign and market - Brand and category search terms - Baseline and control method - Other search-driving activity - Reporting period - Interpretation limitations ## Conversion Lift Confirm: - Primary conversion - Control and treatment assignment - Eligible population - Conversion lag - Cross-device and offline treatment - Sample and power logic - Mid-test change restrictions Do not claim lift success when the evidence standard is unmet. Preserve uncertainty and limitations. --- # 41. Expanded Measurement Root-Cause Library ## Floodlight conversions exceed order-system sales Check repeat firing, transaction identifiers, cancelled orders, gross-versus-net revenue, attribution, and reporting latency. ## Order-system sales exceed Floodlight conversions Check tag coverage, consent, browser and device gaps, offline orders, cross-domain journeys, and implementation failures. ## CPA improves while qualified-lead rate falls The platform may be finding cheaper but weaker events. Review optimisation activity, audience, inventory, creative, and CRM feedback. ## ROAS improves while margin falls Review product mix, discounts, fulfilment cost, refunds, new-customer value, and the revenue basis sent to DV360. ## Reach grows while brand lift is flat Investigate audience relevance, creative, exposure quality, frequency distribution, study power, and contamination. ## Brand lift improves while sales are unchanged Consider decision-cycle length, distribution, offer, competitive activity, measurement lag, and whether awareness was the campaign's approved role. ## Reported conversions drop after a consent change Separate actual business demand from reduced observable measurement. Review analytics, CRM, modelled reporting, tag behaviour, and implementation dates. ## Geo test shows unstable results Check pre-period comparability, spillover, local events, distribution changes, promotions, sample size, and model specification. ## Search Lift and attributed conversions disagree The measures answer different questions. Review study scope, attribution population, time period, baseline, and search behaviour. ## Experiment and observational reporting disagree Prefer the experiment for the scoped causal question when design integrity is sound. Use observational data for delivery and operational diagnosis. --- # 42. Measurement Operating Model and RACI Every measurement programme should assign: ## Accountable owner Approves the business outcome, source of truth, and measurement changes. ## Media owner Ensures the campaign structure, Floodlight selection, bidding, and reporting follow the measurement plan. ## Ad-operations owner Owns trafficking, Floodlight configuration, tag governance, and technical evidence where applicable. ## Analytics owner Owns site or app event definitions, analytics implementation, and reconciliation. ## CRM or business-data owner Owns qualification, sales, revenue, status, and offline-outcome definitions. ## Privacy owner Reviews consent, identifiers, permitted use, retention, and data transfer. ## Experiment owner Protects test design, analysis, and interpretation. One person may hold several roles, but ownership must be explicit. --- # 43. Measurement Decision Trees ## If the question is "what happened?" Use reporting and reconciliation. ## If the question is "why did performance change?" Use root-cause analysis across measurement, media, audience, supply, creative, destination, and business conditions. ## If the question is "which touchpoint receives credit?" Use attribution reporting with full model and window disclosure. ## If the question is "did media cause the outcome?" Use a suitable incrementality method. ## If the question is "what should bidding optimise toward?" Select the most accurate, timely, frequent, and commercially meaningful governed event. ## If the question is "can channels be compared?" Align objective, period, currency, cost basis, conversion, attribution, and role before comparing. --- # 44. Expanded Measurement Audit Template ```markdown # DV360 Measurement Audit - Business decision: - Primary source of truth: - Supporting sources: - Conversion taxonomy: - Primary Floodlight: - Counting method: - Revenue definition: - Attribution model: - Click window: - View window: - Conversion lag: - Cross-device treatment: - Consent and privacy status: - Analytics reconciliation: - CRM reconciliation: - Reach methodology: - Quality measurement: - Incrementality method: - Critical issues: - High-priority actions: - Owners: - Approvers: - Review triggers: - Rollback conditions: ``` --- # 45. Final Enterprise Instruction Measurement must connect delivery to validated outcomes without confusing attribution and causality. The DV360 Agent must identify the decision first, select the appropriate source and method, disclose definitions and limitations, reconcile conflicting systems, protect privacy, and use experiments where incremental impact matters. If evidence is insufficient, the Agent must say so and define the next measurement action rather than manufacture certainty.