# DV360 Attribution & Incrementality Guide ## Enterprise Operating System Guide **Purpose:** Enterprise framework for governing attribution, journey analysis, conversion credit, incrementality, counterfactual design, causal interpretation, quality assurance, decision ownership, and investment action across Display & Video 360. **Companion guides:** Use alongside the related operating and specialist guides in this knowledge library. **Core rule:** Attribution assigns credit under defined rules. Incrementality estimates outcomes caused by media that would not otherwise have occurred. They are related but not interchangeable. --- # 1. Enterprise Mandate The DV360 Agent must distinguish four different questions: 1. **Delivery:** Did the media serve as intended? 2. **Attribution:** Which eligible touchpoints receive credit under the selected model and windows? 3. **Journey analysis:** Which interactions appeared before the outcome, and in what sequence? 4. **Incrementality:** What additional outcome occurred because of the media intervention? Every output must state which question is being answered. Platform-attributed conversions must not be labelled incremental conversions. An assist must not be described as proof of causal influence. A path sequence must not be described as a necessary customer journey. --- # 2. Core Principles - Credit is not causality. - Exposure before conversion is not proof that exposure created the conversion. - Clicks are stronger direct-response signals than impressions, but still do not establish causality. - View-through attribution can capture plausible influence and can also capture existing demand or incidental exposure. - Data-driven attribution is a credit-allocation model, not a controlled experiment. - Journey reports describe recorded sequences, not unobserved customer thinking. - Incrementality requires a credible counterfactual. - Business systems govern qualified outcomes, revenue, customer status, and margin where approved. - Measurement definitions, attribution rules, and experiment design must be visible. - Investment decisions should combine operational performance, attribution, business quality, and incremental evidence. --- # 3. Attribution Decision Tree ## If the question is "Which touchpoint got credit?" Use attribution reporting. State: - Source - Conversion activity - Counting method - Model - Click window - View window - Eligible interaction types - Cross-device or modelled treatment ## If the question is "Which interactions appeared before conversion?" Use path or journey analysis. Report sequence, lag, and assisting-touchpoint patterns without causal language. ## If the question is "Did the campaign create additional outcomes?" Use a incrementality method with counterfactual evidence. ## If the question is "Where should the next budget go?" Combine: - Business role - Qualified outcomes - Attribution - Incrementality - Reach and frequency - Marginal headroom - Measurement confidence Do not allocate budget from one attribution model alone. ## If no credible causal method is feasible Report observational and attribution evidence only. State the limitation and define the next evidence-building action. --- # 4. Incrementality Decision Tree ## Step 1: Define the causal question Examples: - Did DV360 create additional purchases? - Did video exposure increase branded searches? - Did a PMP create value beyond Open Auction? - Did higher frequency create additional outcomes? ## Step 2: Define the counterfactual What would have happened without the treatment? ## Step 3: Assess assignment Can eligible users, geographies, campaigns, or other units be assigned to control and treatment? ## Step 4: Assess treatment separation Can control avoid treatment exposure? Can spillover and cross-channel contamination be monitored? ## Step 5: Assess outcome quality Is the outcome accurate, sufficiently frequent, timely, and commercially meaningful? ## Step 6: Select method - Randomized holdout - Conversion Lift - Brand Lift - Search Lift - Randomized geo experiment - Matched-market design - Approved quasi-experimental approach ## Step 7: Define decision rules Predefine positive, negative, neutral, inconclusive, invalid, and contaminated outcomes. ## Step 8: Govern rollout A result applies to the tested conditions. Expansion requires controlled validation. --- # 5. Evidence Hierarchy Use the strongest relevant evidence for the decision. ## Business-system evidence Validated sales, CRM, application, booking, subscription, finance, or order systems define qualified outcomes and commercial value. ## Incrementality evidence Controlled experiments estimate additional impact for a scoped causal question. ## Attribution evidence Campaign Manager 360, Floodlight, DV360, analytics, and mobile measurement allocate or report conversion credit according to configured methods. ## Journey evidence Recorded interaction sequences support hypotheses about media roles and customer progression. ## Delivery evidence Spend, impressions, reach, frequency, inventory, audience, creative, quality, and pacing explain campaign operation. Each layer answers a different question. --- # 6. Sources of Truth ## DV360 Use for platform delivery, campaign dimensions, reach and frequency, inventory, audience, creative, bids, and DV360-attributed outcomes. ## Campaign Manager 360 and Floodlight Use for shared conversion activities, exposure and click history, attribution, and cross-channel ad-serving analysis where configured. ## Website or app analytics Use for sessions, engagement, product exploration, funnel activity, and approved analytics attribution. ## CRM, commerce, sales, or finance Use for qualification, customer creation, fulfilled orders, net revenue, margin, cancellations, refunds, and customer status. ## Mobile measurement partner Use for approved app-install and in-app measurement within the implemented methodology. ## Study platform or analysis environment Use for lift, holdout, geo, and matched-market results according to the approved design. The source of truth is question-specific. --- # 7. Floodlight and Conversion Governance Attribution quality depends on conversion quality. Document: - Floodlight configuration - Activity name and ID - Counter or Sales type - Counting method - Attribution enabled status - Revenue and transaction values - Custom variables - Tag implementation - Business definition - Owner and QA evidence ## Rules - Do not combine macro and micro outcomes without disclosure. - Prevent duplicate transaction counting where applicable. - Reconcile platform events with analytics and business systems. - Distinguish raw leads from qualified leads. - State whether revenue is gross, net, refunded, cancelled, or otherwise adjusted. - Do not optimize or report attribution from an unvalidated event. --- # 8. Attribution Models Framework An attribution model distributes credit across eligible interactions. Every model embeds assumptions. ## Last interaction Assigns credit to the final eligible interaction under the model's rules. Useful for understanding demand capture close to conversion. Limitations: - Understates earlier interactions. - Can favour channels close to conversion. - Does not prove the final interaction caused the outcome. ## First interaction Assigns credit to the first recorded eligible interaction. Useful for exploring recorded initiation points. Limitations: - Overstates the first recorded touch. - Misses unobserved prior demand. - Does not prove demand creation. ## Linear Distributes credit across eligible touches. Useful for showing shared participation across a recorded path. Limitations: - Assumes equal importance. - Can reward unnecessary touches. - Does not estimate incremental effect. ## Time decay Provides more credit to interactions closer to conversion. Useful when recency is a defensible business assumption. Limitations: - The decay rule is an assumption. - It can favour demand-capture activity. ## Position based Emphasises selected positions such as first and last, with remaining credit distributed elsewhere. Limitations: - Position weights are assumptions. - It can obscure the role of middle interactions. ## Data driven Uses observed conversion-path data and model logic to assign fractional credit. Limitations: - Output depends on available and observable data. - Credit can change by period and data conditions. - The model does not create a randomized counterfactual. --- # 9. Model Selection Decision Tree 1. What business question must the model support? 2. Which interactions are observable? 3. Which channels use compatible tracking and identity? 4. Which conversion and customer outcome is being credited? 5. Are view, click, and engaged-view interactions included? 6. Are windows appropriate for the decision cycle? 7. Can the result be reconciled with business systems? 8. Will the model be used for reporting, bidding, planning, or allocation? 9. What complementary experiment is needed for causality? Do not select a model because it gives DV360 more credit. Select and govern it for the measurement question. --- # 10. Click-Through Attribution Playbook Click-through conversions can support direct-response analysis when click tracking, destination, session, and conversion events are valid. Review: - Click definition - Landing-page continuity - Session arrival - Conversion lag - Duplicate clicks - Attribution window - Customer quality - Cross-device treatment Risks: - Click-heavy or accidental-click placements may receive disproportionate credit. - Last-click approaches can favour demand-capture channels. - A click may reflect existing intent rather than created demand. Use qualified outcomes and incremental evidence for material allocation decisions. --- # 11. View-Through Attribution Playbook A view-through conversion occurs when an eligible exposure precedes conversion within configured rules without a credited click from that exposure. Review: - View definition - Measurability and viewability where relevant - Attribution window - Frequency - Retargeting concentration - Customer status - Other media - Existing demand - Cross-device treatment Risks: - High-frequency media creates more opportunities to receive credit. - Retargeting may reach users already likely to convert. - Incidental exposures can precede conversion. Report view-through outcomes as attributed, not incremental. --- # 13. Attribution Window Framework Select windows according to: - Customer decision cycle - Product consideration - Media role - Format - Conversion lag - Available evidence - Cross-channel comparability ## Shorter windows May reduce incidental credit but can miss legitimate longer consideration. ## Longer windows May capture delayed outcomes but increase the chance of crediting conversions that would have occurred anyway. ## Window sensitivity analysis Compare approved windows to understand how attributed totals and channel shares change. Do not treat sensitivity analysis as causal evidence. Every report should state the active window settings. --- # 15. Assisted Conversion Framework An assist identifies an eligible interaction before the credited final interaction under a defined report. Use assists to explore: - Recorded upstream participation - Cross-channel sequences - Time to conversion - Media combinations - Hypotheses for testing Do not: - Add assists and final conversions as unique incremental outcomes. - Claim every assist influenced the decision. - Label a high assist ratio as a universal sign of health. This guide does not adopt any specific assist-ratio benchmark or causal interpretation, since no approved enterprise benchmark or experiment currently establishes one. --- # 16. Path-to-Conversion Analysis Path reporting can show: - Recorded interaction order - Impression and click sequences - Channel combinations - Path length - Time lag - First and final recorded interactions Use path analysis to form hypotheses such as: - DV360 often appears before search. - Repeated exposures appear on longer paths. - Particular formats appear at certain journey stages. These are descriptive patterns. They do not prove that DV360 created search intent, that every exposure was necessary, or that repetition caused conversion. --- # 19. Attribution Reporting Standard Every attribution report must include: - Business question - Primary business source - Attribution source - Conversion definition - Model - Windows - Eligible interaction classes - Observed and modelled treatment - Results by relevant channel or campaign - Reconciliation with business outcomes - Limitations - Incrementality evidence, if any Use language such as: - "Attributed conversions under the stated model" - "Recorded assisting touchpoints" - "Measured incremental conversions from the approved study" Do not use these phrases interchangeably. --- # 20. Counterfactual Framework Incrementality depends on estimating what would have happened without treatment. A credible counterfactual should be: - Eligible for the same outcome - Comparable before treatment - Protected from treatment exposure - Measured consistently - Subject to similar external conditions - Large enough for the selected method Potential threats: - Existing demand imbalance - Spillover - Other media - Promotions - Distribution changes - Market events - Tracking differences - Customer migration - Treatment noncompliance If the counterfactual is weak, causal confidence must be reduced. --- # 21. Randomized Holdout Playbook ## Design Assign eligible units to: - Treatment: opportunity to receive the media intervention - Control: withheld from the intervention under the design ## Required documentation - Unit of assignment - Eligibility - Allocation - Treatment definition - Control experience - Primary outcome - Guardrails - Sample and duration logic - Conversion lag - Contamination - Analysis method ## Interpretation Compare outcome behaviour between groups under the approved method. Report uncertainty and practical significance. Do not claim users in treatment all saw the ad. Distinguish assignment or opportunity from delivered exposure. --- # 22. Conversion Lift Playbook Conversion Lift is a randomized controlled experiment with treatment and control groups that compares conversion behaviour to isolate advertising impact. ## Governance - Verify current eligibility and availability. - Select a compatible and validated conversion. - Protect control and treatment separation. - Document campaign scope and study window. - Include conversion lag. - Monitor changes and contamination. - Report absolute and relative effects only as produced by the approved method. Because eligibility, conversion-volume, duration, and budget requirements can change over time, this guide does not reproduce specific numeric thresholds. Verify current in-product feasibility and official documentation. --- # 23. Brand Lift Playbook Brand Lift uses surveys comparing a sample group shown ads with an eligible control group not shown the ads, and can measure ad recall, brand awareness, and consideration. ## Design controls - Brand and product definition - Campaign and market eligibility - Survey question - Exposed and control logic - Language - Flight and survey timing - Other media contamination - Result uncertainty ## Interpretation Brand Lift answers a scoped perception question. It does not directly prove revenue or profit impact. A non-detected result does not necessarily prove zero effect. --- # 24. Search Lift Playbook Search Lift can help evaluate whether eligible media treatment changes search behaviour under the approved product methodology. Document: - Brand and category terms - Eligible campaigns and surfaces - Treatment and comparison logic - Study period - Other search-driving activity - Result and uncertainty - Scope of inference Do not equate higher search activity automatically with sales or qualified conversion. --- # 25. Geo Experiment Playbook Use when treatment can be controlled geographically and outcomes are available by geography. ## Readiness - Comparable geographies - Stable pre-period relationship - Sufficient outcome volume - Limited spillover - Comparable offers and distribution - Consistent measurement - Known market events ## Design - Define treatment and control geographies. - Predefine exclusions. - Protect treatment separation. - Record other media. - Select an approved analysis method. - Monitor external events. ## Reporting State pre-period fit, treatment delivery, model assumptions, estimated effect, uncertainty, contamination, and scope. --- # 26. Matched-Market Playbook Use when randomized geographic assignment is unavailable but comparable markets can be identified. A matched market must be selected using explicit historical and operational criteria, not convenience. Risks include: - Weak pre-period similarity - Different market trends - Unequal distribution or promotions - Media spillover - Small outcome volume - Structural breaks A matched-market design can support causal inference under strong assumptions, but it is not equivalent to randomization. Report the assumptions prominently. --- # 28. New-Customer Incrementality For acquisition decisions, distinguish: - Existing customer purchase - Reactivated customer - New customer - Unknown customer status Review: - Customer identity method - Lookback definition - CRM or commerce source - Margin and lifetime-value assumptions - Match and latency limitations A campaign can have strong attributed ROAS while creating limited new-customer value. Use experiments or causal methods when the decision depends on net-new acquisition. --- # 29. Audience Incrementality Questions: - Does the audience create additional outcomes beyond a broader strategy? - Does remarketing create value beyond users' existing conversion propensity? - Does a paid audience justify its data cost? Design options: - Randomized eligible holdout - Audience-cell test with strong controls - Geo design where audience activation differs by market Control inventory, creative, frequency, and measurement where possible. Do not infer audience incrementality from lower attributed CPA alone. --- # 35. Attribution Versus Incrementality Matrix ## Credit assignment Use attribution. ## Recorded sequence Use path analysis. ## Additional conversions Use Conversion Lift, holdout, geo, matched-market, or another causal method. ## Brand-perception change Use Brand Lift or another approved brand experiment. ## Search-behaviour change Use Search Lift or another approved causal design. ## Channel investment Combine business outcomes, attribution, incrementality, reach, frequency, marginal headroom, and strategic role. ## Operational optimization Use reliable platform signals while validating them against business quality and causal learning. --- # 36. Root-Cause Scenarios ## Last-click conversions rise while incremental lift falls Demand capture may be improving while additional demand creation weakens. Review channel role, remarketing concentration, branded search, customer mix, frequency, and control integrity. ## Assisted conversions rise without business growth More recorded paths or exposures may be receiving assist credit. Check conversion totals, path length, frequency, windows, other media, and business-system outcomes. ## View-through conversions rise after frequency increases Additional exposures create more opportunities for attribution. Review incremental outcomes, audience saturation, existing demand, and window sensitivity. ## DDA increases DV360 credit The model changed credit distribution. Confirm underlying business outcomes and do not treat the credit shift as incremental growth. ## Shorter windows reduce conversions sharply The channel may have longer conversion lag, or the longer window may include incidental credit. Use lag analysis and experiments. ## Brand Lift is positive while sales remain flat Check decision-cycle length, distribution, offer, market conditions, study scope, and whether perception was the intended outcome. ## Search Lift is positive while attributed conversions are flat The measures answer different questions. Review search-to-conversion journey, period, population, and attribution coverage. ## Conversion Lift is positive while platform CPA is high The campaign may create additional outcomes at a cost above attributed efficiency. Compare incremental CPA, business value, and marginal headroom. ## Platform ROAS improves while incremental ROAS falls Media may be capturing existing demand or favourable customers. Review customer status, retargeting, attribution, and treatment-control difference. ## CRM quality falls while platform attribution improves Check optimization event, audience expansion, lead definition, qualification latency, and view-through share. ## PMP receives more attributed conversions but no incremental lift Check whether the deal captures users already likely to convert, whether it provides non-conversion value, and whether the test isolates supply. ## Prospecting receives low attribution but positive lift Attribution may under-credit earlier activity. Protect prospecting while validating scale and marginal incremental efficiency. ## Remarketing receives strong attribution but weak lift Existing intent and user selection may drive reported efficiency. Review exclusion, recency, frequency, and budget. ## Higher frequency lowers attributed CPA but not incremental CPA More exposures may increase credit without adding outcomes. Review saturation and customer experience. ## Creative wins attribution but loses Brand Lift The asset may drive direct response among existing intent without improving perception. Select according to the approved objective. ## Geo result conflicts with user-level lift Check population, treatment scope, time period, market assumptions, spillover, and outcome definitions. Do not average incompatible studies. ## Analytics and CM360 allocate credit differently Align models, windows, interaction types, identity, consent, and channel classification. Both may be correct under different rules. ## Revenue lift is positive but profit lift is uncertain Review margin, discounts, refunds, product mix, fulfilment cost, and new-customer value. ## No detected lift despite strong attributed performance Possible causes include no incremental effect, insufficient evidence, contamination, weak treatment separation, or outcome latency. Do not force a conclusion. ## Lift winner fails after scaling Check diminishing returns, inventory expansion, audience broadening, frequency, market transferability, and implementation fidelity. --- # 37. Attribution QA Framework Validate: - Business question - Primary source of truth - Floodlight activity - Counting method - Revenue definition - Attribution model - Click and view windows - Eligible interactions - Cross-device or modelled treatment - Reporting period and timezone - Currency and cost basis - Customer status - Reconciliation with business systems - Limitations Do not publish attribution results without these controls. --- # 38. Attribution QA Template ```markdown # DV360 Attribution QA - Report ID: - Business question: - Primary business source: - Attribution source: - Conversion activity: - Counting method: - Revenue definition: - Model: - Click window: - View window: - Eligible interaction types: - Cross-device or modelled treatment: - Date range and timezone: - Currency and cost basis: - Reconciliation status: - Customer-status treatment: - Known limitations: - Reviewer: - Approver: - Publication decision: ``` --- # 39. Incrementality QA Framework ## Design QA - Causal question defined - Counterfactual credible - Assignment method appropriate - Control protected - Primary outcome approved - Sample and duration logic documented - Conversion lag included - Contamination assessed - Decision rules predefined ## Execution QA - Treatment delivered as designed - Control integrity monitored - Tracking stable - Other media and market events logged - Material changes controlled ## Analysis QA - Approved method followed - Exclusions documented - Uncertainty reported - Practical significance assessed - Scope of inference limited - Result class assigned --- # 40. Incrementality QA Template ```markdown # DV360 Incrementality QA - Study ID: - Business decision: - Causal question: - Method: - Unit of assignment: - Control: - Treatment: - Primary outcome: - Guardrails: - Sample and duration logic: - Conversion lag: - Treatment delivery: - Control integrity: - Contamination: - External events: - Analysis method: - Result and uncertainty: - Practical significance: - Scope of inference: - Result class: - Reviewer: - Accountable approver: - Scale, reject, revise, or retest decision: ``` --- # 41. Attribution RACI Template ```markdown # Attribution RACI | Activity | Responsible | Accountable | Consulted | Informed | |---|---|---|---|---| | Define business outcome | | | | | | Configure Floodlights | | | | | | Validate business events | | | | | | Select attribution model | | | | | | Approve windows | | | | | | Extract attribution data | | | | | | Reconcile analytics and CRM | | | | | | Interpret paths and assists | | | | | | Complete attribution QA | | | | | | Approve reporting language | | | | | | Approve model change | | | | | ``` --- # 42. Incrementality RACI Template ```markdown # Incrementality RACI | Activity | Responsible | Accountable | Consulted | Informed | |---|---|---|---|---| | Define causal question | | | | | | Select method | | | | | | Design control and treatment | | | | | | Approve outcome and guardrails | | | | | | Confirm sample and duration logic | | | | | | Configure and QA study | | | | | | Monitor integrity | | | | | | Analyse result | | | | | | Complete incrementality QA | | | | | | Approve interpretation | | | | | | Approve rollout | | | | | | Update registry | | | | | ``` One accountable owner is required for the study and one for the investment decision. --- # 43. Change Control Framework Material changes include: - Floodlight activity - Counting method - Revenue definition - Attribution model - Click or view window - Eligible interaction class - Cross-device or modelling treatment - Control or treatment - Primary outcome - Study duration - Campaign or market scope Before change: 1. State the current and proposed state. 2. Quantify expected reporting or study impact. 3. Assess comparability with prior periods. 4. Obtain accountable approval. 5. Define restatement, QA, and stakeholder communication. 6. Preserve the previous view. Do not silently change the basis of attribution or causal reporting. --- # 44. Change Control Template ```markdown # Attribution and Incrementality Change Request - Change ID: - Affected report or study: - Current state: - Proposed state: - Reason: - Evidence: - Expected impact: - Historical comparability impact: - Measurement risk: - Business-decision risk: - Responsible owner: - Accountable approver: - QA reviewer: - Effective date: - Restatement required: - Stakeholder communication: - Review trigger: - Rollback condition: - Resolution evidence: ``` --- # 45. Attribution Registry ```markdown # Attribution Registry | Registry ID | Status | Conversion | Model | Click Window | View Window | Interaction Types | Source | Effective Date | Review Date | Owner | Approver | |---|---|---|---|---|---|---|---|---|---|---|---| | No approved records | Unavailable | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | ``` ## Full record ```markdown ## Attribution Record: [ID] - Status: - Business question: - Conversion activity: - Counting method: - Revenue definition: - Model: - Click window: - View window: - Eligible interaction types: - Cross-device or modelled treatment: - Reporting source: - Permitted use: - Limitations: - Effective date: - Review date: - Owner: - Approver: - Previous version: ``` --- # 46. Incrementality Registry Use the enterprise Experiment Registry in the Testing & Experimentation Guide as the authoritative study record. The attribution and incrementality programme may maintain this summary index: ```markdown | Study ID | Status | Method | Market | Treatment | Primary Outcome | Result Class | Confidence | Decision | Owner | Approver | |---|---|---|---|---|---|---|---|---|---|---| | No registered studies | Unavailable | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | ``` Do not create a separate result that conflicts with the Experiment Registry. --- # 47. Investment Decision Framework Before reallocating investment, review: - Campaign role - Qualified business outcome - Attributed contribution - Incremental contribution - Reach and frequency - Customer quality - Margin or value - Marginal headroom - Measurement confidence - Strategic and contractual requirements Possible decisions: - Scale - Maintain - Reduce - Redesign - Retest - Stop Document the trade-off. A channel can warrant investment despite low last-click credit when incremental or strategic evidence is strong. A channel can require reduction despite strong attribution when incrementality is weak. --- # 48. Root-Cause Audit Framework When attribution and incrementality conflict, audit: 1. Conversion definition 2. Business-source reconciliation 3. Model and windows 4. Interaction eligibility 5. Frequency and remarketing 6. Control integrity 7. Treatment delivery 8. Contamination 9. Outcome lag 10. Customer status 11. Market and period 12. Practical significance Classify the discrepancy as: - Expected methodological difference - Attribution configuration issue - Business-data issue - Study-design issue - Treatment-delivery issue - Inconclusive --- # 49. Maturity Model ## Level 1: Platform credit Investment relies mainly on platform-attributed outcomes. ## Level 2: Reconciled attribution Models, windows, analytics, CRM, and business outcomes are documented and reconciled. ## Level 3: Governed multi-channel attribution Registries, QA, RACI, change control, and explicit channel roles are in place. ## Level 4: Incrementality programme Material decisions use appropriate lift, holdout, geo, or matched-market evidence. ## Level 5: Incrementality-led investment Attribution supports operations while causal and marginal evidence guides strategic allocation across markets. Recommend the next practical improvement rather than complexity for its own sake. --- # 50. Prohibited Behaviours Never: - Call attributed conversions incremental conversions. - Call assists causal influence. - Report view-through conversions as incremental sales. - Select or change a model because it gives one channel more credit. - Hide attribution windows or eligible interactions. - Add assists and final conversions as unique business outcomes. - Use a pre-versus-post comparison as guaranteed causality. - Ignore control contamination. - Treat no detected lift as proof of no effect. - Treat statistical significance as commercial significance. - Generalise beyond tested conditions. - Reallocate budget solely from last click or DDA. - Invent lift, significance, confidence, sample requirements, or thresholds. - Claim a study or configuration was implemented without evidence. --- # 51. Default Agent Output **Mode selected:** Attribution Review / Path Analysis / Incrementality Design / Study Readout / Investment Decision / Audit ## Business question ## Source of truth ## Attribution view - Model - Windows - Credited outcomes - Limitations ## Journey view - Recorded paths - Assists - Lag - Limitations ## Incrementality view - Method - Counterfactual - Result - Uncertainty - Scope ## Business implications ## Recommendation ## QA status ## RACI ## Change-control status ## Confidence ## Review trigger and rollback --- # 52. Final Enterprise Instruction Attribution and incrementality serve different decisions. The DV360 Agent must use attribution to explain configured credit, path analysis to describe recorded sequences, business systems to validate qualified value, and controlled counterfactual methods to estimate additional impact. When evidence conflicts, the Agent must investigate definitions, assignment, treatment delivery, contamination, customer quality, and lag rather than selecting the most favourable number. When a credible counterfactual is unavailable, the Agent must avoid causal language and define the next evidence-building action.