# DV360 Approved APAC Benchmarks Guide ## Project Knowledge for Claude and ChatGPT **Version:** 1.0 **Purpose:** Govern how DV360 benchmarks are sourced, approved, segmented, applied, monitored, refreshed, and retired across APAC markets. **Companion guides:** Use alongside the related operating and specialist guides in this knowledge library. **Default status:** No numerical benchmark is approved unless a complete registry record carries an approved status, owner, approver, approval date, and valid review date. --- # 1. Purpose This guide prevents the DV360 Agent from inventing, misapplying, or over-promising benchmarks. A benchmark is a structured reference point used to support planning, diagnosis, target-setting, forecasting, or performance interpretation. It is not a guarantee, universal truth, contractual commitment, or substitute for campaign-specific evidence. DV360 performance varies materially by: - Market - Currency - Objective - Format - Inventory environment - Buying route - Publisher and supply path - Audience - Creative - Device - Seasonality - Brand category - Offer - Destination - Measurement setup - Attribution window - Conversion definition - Data quality The DV360 Agent must use benchmarks only when the comparison is sufficiently relevant and the benchmark has an approved source, owner, methodology, scope, and freshness status. --- # 2. Core Benchmark Rules 1. Never invent a number, range, threshold, percentile, or market multiplier. 2. Never convert a planning assumption into an approved benchmark without review. 3. Never describe a benchmark as a guarantee. 4. Never compare unlike objectives, formats, markets, currencies, attribution settings, or conversion definitions without disclosure. 5. Prefer campaign-specific baselines and matched historical evidence over broad industry averages when the internal data is reliable and comparable. 6. Use benchmarks as diagnostic starting points, not automatic optimisation triggers. 7. Always disclose source, scope, effective date, and confidence. 8. Separate media-cost benchmarks from total-cost benchmarks. 9. Separate platform outcomes from business-qualified outcomes. 10. Separate attributed outcomes from incremental outcomes. 11. Require accountable human approval before publishing or operationalising a benchmark. 12. Retire benchmarks that are stale, unsupported, or no longer comparable. --- # 3. Benchmark Types ## Planning benchmark Used to estimate expected delivery, cost, reach, response, or outcome before launch. Planning benchmarks should be expressed as: - An approved range - A scenario assumption - A confidence level - A set of conditions They should not be presented as performance promises. ## Diagnostic benchmark Used to identify where performance may require investigation. Example uses: - Viewability below a validated comparable range - Completion materially different from similar creative and inventory - CPM movement inconsistent with comparable market history A diagnostic benchmark triggers questions, not an automatic action. ## Target benchmark An operational target approved for a specific campaign, insertion order, line item, deal, or test. Targets may differ from external or historical benchmarks because the advertiser has different economics, inventory, creative, risk, or ambition. ## Guardrail benchmark A minimum or maximum condition used to protect quality, cost, frequency, delivery, suitability, or business value. Guardrails require an owner, rationale, monitoring method, and exception process. ## Historical baseline A previous comparable campaign or period used as the reference. Historical results are useful only after adjusting for material changes in market, budget, objective, audience, supply, creative, offer, destination, and measurement. ## Peer-group benchmark A reference built from a defined group of comparable campaigns. The peer group must be documented. "APAC average" is not sufficiently specific. ## External industry benchmark A reference published by a platform, verification provider, research organisation, publisher, trade body, or another documented source. External benchmarks require methodology review and should be treated cautiously when the source does not disclose sample, market, format, measurement, or period. ## Experimental baseline A control or pre-period used in an approved experiment. This is not automatically transferable to unrelated campaigns. --- # 4. Source Hierarchy Use the highest-quality comparable source available. ## Tier 1: Advertiser business outcomes Examples: - Revenue - Margin - Qualified leads - Approved applications - New customers - Offline sales Strength: - Most relevant to commercial value Limitations: - May be delayed - Can contain operational or CRM bias - Often requires reconciliation ## Tier 2: Validated campaign and ad-server history Examples: - Comparable DV360 and Campaign Manager 360 campaigns - Floodlight outcomes - Reach and frequency - Media quality Strength: - Operationally relevant Limitations: - Attribution and setup differences - Historical strategy may not match the new plan ## Tier 3: Approved cross-client or peer-group dataset Strength: - Larger sample - Useful for planning ranges Limitations: - Category, objective, and confidentiality differences - Aggregation can hide material variation ## Tier 4: Platform, verification, publisher, or research source Strength: - External methodology or broad scale may be available Limitations: - Commercial interest - Different inventory or measurement population - Limited transparency ## Tier 5: Expert assumption Use only when no reliable benchmark exists. Requirements: - Label as an assumption - State rationale - State uncertainty - Obtain approval - Replace with observed evidence after launch --- # 5. Approval Standard A benchmark is approved only when the registry contains: - Benchmark ID - Metric name - Metric definition - Formula - Objective - Format - Environment - Market - Currency - Buying route - Audience scope - Creative scope - Measurement source - Attribution model and windows - Conversion definition where relevant - Cost basis - Sample description - Reference period - Approved value or range - Confidence - Limitations - Owner - Approver - Approval date - Review date - Status If any critical field is missing, the Agent must treat the benchmark as provisional or unusable. --- # 6. Benchmark Confidence ## High confidence Use when: - The source is directly comparable - Definitions match - Sample is adequate for the stated use - Data is recent enough for the market - Measurement is validated - Material confounders are controlled or documented ## Medium confidence Use when: - The source is useful but some scope, sample, timing, or measurement differences remain - The benchmark can guide planning or diagnosis but should not set a hard commitment ## Low confidence Use when: - The source is directional - Methodology is incomplete - Market or format comparability is weak - The sample is sparse - The data is old or volatile - The benchmark is based mainly on expert judgement The Agent must state what evidence would increase confidence. --- # 7. APAC Market Governance APAC is not one benchmark market. A benchmark must identify the specific market or approved market group. Grouping markets requires a documented rationale based on comparable media economics, inventory, audience, language, device use, regulation, measurement, and business context. ## Market requirements - Use local currency for market planning where practical. - If converted to a common currency, record the foreign-exchange source and effective date. - Do not compare local-currency cost metrics without conversion. - Do not assume one market's CPM, CTR, completion, CPA, or reach applies to another. - Do not create market multipliers without an approved method. - Consider local publisher and deal availability. - Consider language and localisation. - Consider market-specific privacy and targeting restrictions. - Consider seasonality and major commercial periods. ## Regional roll-up A regional value must disclose: - Markets included - Weighting method - Currency treatment - Missing markets - Whether the result is spend-weighted, impression-weighted, conversion-weighted, or unweighted Do not average market percentages or unit costs without an appropriate weighting method. --- # 8. Objective Segmentation Benchmarks must reflect the campaign objective. ## Awareness and reach Relevant metric families may include: - CPM - Reach - Frequency - Cost per reached user - Viewability - Completion - Brand lift - Search lift Do not judge awareness solely by clicks or attributed conversions. ## Consideration Relevant metric families may include: - Qualified sessions - Cost per qualified session - Engagement quality - Journey starts - Product views - Consideration lift CTR and CPC may be diagnostics but do not prove consideration quality. ## Conversion or acquisition Relevant metric families may include: - Qualified conversion rate - CPA - Cost per qualified lead - Revenue - ROAS - Margin - New-customer acquisition Platform conversions require reconciliation with the advertiser's business system. ## Retargeting and retention Separate these from prospecting because audience intent, attribution, frequency, and incremental value differ. --- # 9. Format Segmentation Never combine formats into one benchmark unless the approved methodology explicitly supports the roll-up. ## Display Possible metrics: - CPM - CTR - CPC - Measurability - Viewability - Qualified session rate - Conversion rate - CPA Segment by static, HTML5, responsive, rich media, environment, device, and buying route where differences are material. ## Video Possible metrics: - CPM - CPV under a stated view definition - View rate - Quartile rates - Completion rate - Cost per completed view - Viewability - Reach and frequency - Brand or business outcomes Do not compare unlike view definitions, durations, skippability, or environments. ## Connected television Possible metrics: - CPM - Reach - Frequency - Completion - Incremental reach - Brand lift - Search lift - Qualified business outcome State whether reach is user, device, household, deduplicated, or modelled. ## Audio Possible metrics: - CPM - Reach - Frequency - Completion where supported - Companion response - Brand lift - Search lift - Qualified outcome Do not compare podcast, streaming audio, digital radio, and YouTube audio without documenting inventory and measurement differences. ## Native Possible metrics: - CPM - CTR - CPC - Qualified session rate - Conversion rate - CPA - Viewability where supported State rendering and measurement limitations. --- # 10. Buying-Route Segmentation Separate when material: - Open Auction - Private Auction - Preferred Deal - Programmatic Guaranteed - Curated package - YouTube auction - YouTube or publisher reservation A private or guaranteed route may have higher cost because it provides priority, certainty, context, audience, or unique access. It should not be judged against Open Auction cost alone. A benchmark must document whether media is bought through comparable publishers, placements, formats, and devices. --- # 11. Cost Basis Every cost benchmark must identify the numerator. Possible bases: - Media cost - Billable media cost - Total media cost - Total cost including data - Total cost including verification - Total advertiser cost ## Rules - Do not compare media-cost CPM with total-cost CPM. - Do not calculate CPA from one cost basis and compare it with a benchmark using another. - Record currency and tax treatment. - Record whether agency or service fees are included. - Record whether data, platform, verification, serving, creative, and production costs are included. - Never invent missing fee values. --- # 12. Metric Definition Registry Definitions must come from the reporting system used for the analysis. ## Impressions Record the platform's exact impression definition and whether filtered, billable, measured, or monitored populations differ. ## Click-through rate Typical formula: `Clicks / Impressions` Confirm which click and impression fields are used. ## CPM Typical formula: `Cost / Impressions × 1,000` State the cost basis. ## CPC Typical formula: `Cost / Clicks` State the click definition and cost basis. ## Measurability Use the reporting platform's defined measurable and eligible populations. Do not confuse measurability with viewability. ## Viewability Use the reported viewable-impression rate and state provider, format, eligible population, and methodology. Official DV360 guidance states that predicted viewability targeting uses Active View and that stricter thresholds can reduce eligible inventory. Predicted viewability is a targeting input, not the same as observed viewability. ## Video completion rate Confirm the numerator and denominator used by the selected report. Do not assume every system uses the same population. ## View rate and CPV State the view definition because definitions can differ by video format. ## Conversion rate State whether the denominator is impressions, clicks, sessions, or another event. ## CPA Typical formula: `Approved cost basis / Approved conversion count` State conversion source and attribution. ## ROAS Typical formula: `Approved revenue / Approved cost basis` State whether revenue is gross, net, cancelled, refunded, tax-inclusive, or otherwise adjusted. ## Reach and frequency State whether reporting is based on users, devices, households, accounts, deduplicated populations, or modelled estimates. --- # 13. Measurement Alignment Before using an outcome benchmark, align: - Floodlight activity - Counter or sales activity - Counting method - Attribution model - Click window - View window - Conversion lag - Revenue value - Analytics event - CRM qualification - New versus existing customer status - Online versus offline outcome Two campaigns can report different conversion results because definitions and attribution differ. A benchmark is unusable when those differences are not understood. --- # 14. Benchmark Creation Workflow ## Step 1: Define the decision Examples: - Planning budget - Setting a target - Diagnosing performance - Reviewing a publisher deal - Evaluating market maturity ## Step 2: Define the comparison cell Specify market, objective, format, environment, audience role, buying route, device, creative, and measurement. ## Step 3: Collect candidate sources Capture internal history, advertiser data, peer-group data, and qualified external references. ## Step 4: Standardise definitions Align currency, cost basis, formulas, attribution, conversion, and reporting fields. ## Step 5: Remove unsuitable records Exclude records with broken tracking, incompatible objectives, unsupported definitions, or material anomalies that cannot be explained. Document exclusions. ## Step 6: Assess sample quality Review campaign count, spend, impressions, outcome count, market coverage, and concentration where available. Do not invent minimum sample thresholds. ## Step 7: Select statistical representation Possible approved approaches include: - Median - Interquartile range - Weighted average - Percentile band - Scenario range The chosen method must match the metric and decision. ## Step 8: Document limitations State seasonality, sparse cells, market concentration, advertiser concentration, and measurement gaps. ## Step 9: Approve Assign owner, approver, approval date, review date, status, and permitted use. --- # 15. Statistical Decision Rules - Prefer ranges over a single point when natural variation is material. - Use medians when extreme values would distort an average. - Use weighted values only when the weighting method is appropriate and disclosed. - Do not average ratios blindly. - Avoid combining campaigns with incompatible denominators. - Do not treat a small sample as precise. - Do not remove inconvenient results without a documented quality reason. - Do not use the same campaign in both a benchmark and a supposedly independent comparison without disclosure. - Separate confidence in the data from confidence in the recommendation. - Never claim statistical significance without the required method and evidence. --- # 16. Sparse-Data Framework When data is limited: 1. Use the nearest defensible comparison. 2. State the differences. 3. Lower confidence. 4. Use a wider scenario range if approved. 5. Treat the benchmark as provisional. 6. Replace it with observed campaign evidence after launch. Possible hierarchy: - Same advertiser, market, objective, and format - Same category, market, objective, and format - Same market, objective, and format - Comparable market group, objective, and format - External directional source - Expert assumption Do not hide sparse data behind false precision. --- # 17. Freshness and Review Freshness should reflect market volatility, platform change, inventory change, privacy change, inflation, seasonality, and measurement change. Every approved record needs: - Source period - Approval date - Next review date - Status ## Status values - Draft - Provisional - Approved - Approved with limitations - Under review - Retired - Rejected A benchmark should move to Under review when: - Metric definitions change - Platform reporting changes - Inventory or format changes materially - A market experiences structural cost change - Attribution or tracking changes - New evidence conflicts with the current range - The review date passes --- # 18. Benchmark Use in Planning For planning, the Agent must provide: - Low, central, and high scenarios where approved - Assumptions - Cost basis - Objective and format - Market and currency - Expected range, not guaranteed outcome - Confidence - Risks - Measurement plan ## Forecast formula examples Use only after inputs are approved. `Estimated impressions = Budget / CPM × 1,000` `Estimated clicks = Estimated impressions × CTR` `Estimated conversions = Approved denominator × Conversion rate` `Estimated revenue = Estimated conversions × Approved value per conversion` Label all outputs as calculations based on assumptions, not observed results. --- # 19. Benchmark Use in Optimisation A benchmark variance should trigger diagnosis. ## If CPM is above benchmark Investigate: - Market and seasonality - Audience restriction - Inventory quality - Deal or floor - Device and format - Viewability targeting - Competition ## If CTR is below benchmark Investigate creative, audience, placement, device, and objective. Do not optimise to CTR if clicks are not the business objective. ## If viewability is below benchmark Investigate measurability, publisher, app, exchange, position, creative, and control settings. Do not assume the benchmark applies to every environment. ## If completion is below benchmark Investigate duration, skippability, autoplay, device, inventory, creative, and view definition. ## If CPA is above benchmark Decompose: `CPA = Cost / Conversions` Then diagnose cost, eligible reach, click or session quality, conversion rate, conversion definition, attribution, and business qualification. Do not pause solely because one metric misses a benchmark. --- # 20. Benchmark Use in Reporting Every benchmark comparison should state: - Actual result - Benchmark or range - Variance - Scope match - Confidence - Explanation - Recommended action, if any Avoid statements such as "underperformed benchmark" without explaining whether the benchmark is genuinely comparable. Use plain language: - "The result is outside the approved planning range, but the comparison is low confidence because the inventory and attribution differ." - "The result is within the approved range; this does not by itself prove strong business performance." --- # 21. Quality Benchmarks Quality benchmarks may cover: - Measurability - Viewability - Invalid traffic - Brand suitability - Geographic accuracy - Completion - Audibility - Attention where approved ## Rules - State provider and methodology. - Do not compare providers as if definitions are identical. - Separate measured population from total delivery. - Do not use a quality benchmark as proof of business impact. - A stricter quality target can reduce scale and increase cost. - Mandatory advertiser controls override average industry performance. - Do not copy a verification vendor's benchmark into the registry without approval. --- # 22. Deal and Supply Benchmarks Deal benchmarks should compare: - Like publisher or inventory - Like format and device - Like market and dates - Like audience and creative - Like measurement Possible metrics: - Effective CPM - Delivery rate - Win behaviour where applicable - Reach contribution - Viewability - Invalid traffic - Suitable delivery - Qualified outcome - Open Auction premium A PMP or guaranteed deal can justify a higher cost through unique access, priority, certainty, context, audience, quality, or operational value. The benchmark must reflect the reason for the deal. --- # 23. Audience Benchmarks Audience benchmarks require: - Source and owner - First-party, platform, publisher, marketplace, contextual, or broad classification - Prospecting, retargeting, retention, or suppression role - Data cost - Scale - Freshness - Consent and availability - Qualified outcome Do not combine retargeting and prospecting. Do not assume a narrow segment should have a lower CPA. Do not use CTR as proof of audience accuracy. --- # 24. Creative Benchmarks Creative benchmarks must control or disclose: - Concept - Execution - Format - Size, ratio, or duration - Audience - Inventory - Device - Frequency - Market and language - Destination A creative benchmark should support production decisions, not penalise new concepts before fair delivery. Possible metrics: - Reach and frequency - Viewability - View or completion definition - Qualified session rate - Conversion rate - Brand lift - Qualified outcome --- # 25. Benchmark Exceptions A campaign may deliberately operate outside a benchmark when: - Inventory is scarce or premium - A guaranteed commitment exists - A new market or format is being tested - The audience is strategically narrow - Quality controls are stricter - The advertiser prioritises incremental reach - The campaign has a different commercial value - Measurement is experimental Every exception must document: - Reason - Owner - Approver - Expected trade-off - Guardrail - Review trigger - Exit condition --- # 26. Approved Benchmark Registry This is the operational registry. A numerical value is approved only when a completed record below has `Status: Approved` or `Status: Approved with limitations`, plus a named owner and approver. ## 26.1 Registry Index Use one row per approved or provisional benchmark. | Benchmark ID | Status | Market | Objective | Format | Metric | Approved Range | Currency | Cost Basis | Confidence | Source Period | 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 | N/A | N/A | Rules: - The index is a summary, not the complete evidence record. - Every index row must link to a full Benchmark Record. - Do not place a value in the index until the matching record is complete. - Use `Unavailable` when no approved value exists. - Use separate rows when market, objective, format, environment, cost basis, attribution, or conversion definitions differ materially. - Do not place confidential advertiser names or results in a shared registry unless explicitly approved. ## 26.2 Benchmark ID Convention Suggested structure: `DV360-[MARKET]-[OBJECTIVE]-[FORMAT]-[METRIC]-[VERSION]` Example structure only: `DV360-SG-AWARENESS-DISPLAY-CPM-V1` This example is an identifier format, not an approved benchmark. ## 26.3 Full Benchmark Record Template Copy and complete one record for every proposed benchmark. ```markdown ## Benchmark Record: [BENCHMARK ID] ### Status and Governance - Status: Draft / Provisional / Approved / Approved with limitations / Under review / Retired / Rejected - Permitted use: Planning / Diagnosis / Target setting / Reporting / Guardrail / Test design - Owner: - Approver: - Approval date: - Effective date: - Next review date: - Retirement date, if applicable: - Confidentiality classification: ### Scope - Market: - Approved market group, if applicable: - Advertiser category: - Objective: - Funnel role: - Format: - Environment: - Device: - Buying route: - Inventory scope: - Publisher or deal scope: - Audience role: - Creative scope: - Flight or seasonal scope: ### Metric Definition - Metric name: - Business meaning: - Reporting platform: - Source field or fields: - Numerator: - Denominator: - Formula: - Unit: - Currency: - Foreign-exchange source and date, if converted: - Cost basis: - Tax treatment: - Included fees: - Excluded fees: ### Measurement and Attribution - Measurement source: - Floodlight activity or business event: - Counter or sales activity: - Counting method: - Attribution model: - Click window: - View window: - Conversion lag treatment: - Post-click and post-view treatment: - Deduplication method: - Revenue definition: - New versus existing customer treatment: - Online versus offline treatment: ### Evidence Base - Source name: - Source type: Advertiser / Campaign history / Peer group / External / Expert assumption - Source location: - Reference period: - Number of campaigns: - Number of advertisers, if approved for disclosure: - Spend represented: - Impressions represented: - Outcome count represented: - Markets represented: - Concentration risks: - Exclusion rules: - Data-quality checks: - Known anomalies: ### Statistical Method - Method: Median / Weighted average / Interquartile range / Percentile band / Scenario range / Other - Weighting method: - Outlier treatment: - Missing-data treatment: - Minimum evidence rule: - Low value: - Central value: - High value: - Confidence: High / Medium / Low - Confidence rationale: ### Interpretation - What this benchmark can answer: - What this benchmark cannot answer: - Appropriate comparisons: - Inappropriate comparisons: - Expected trade-offs: - Known limitations: - Conditions requiring exception approval: ### Change Control - Previous benchmark ID or version: - Change reason: - Evidence added or removed: - Impacted plans, reports, or agents: - Change owner: - Change approver: ``` ## 26.4 Approval Gate A benchmark can move to `Approved` only when all critical gates pass: - The metric definition and formula are explicit. - Market, objective, format, environment, audience role, and buying route are defined. - Currency and cost basis are aligned. - Source and reference period are documented. - Measurement and attribution settings are comparable. - The sample is adequate for the stated use or limitations are explicit. - Statistical treatment is documented. - Owners, approvers, approval date, and review date are recorded. - Confidentiality and permitted use are clear. - No unsupported or invented value has been introduced. If any critical gate fails, keep the record as `Draft`, `Provisional`, or `Unavailable`. --- # 27. Format-Specific Registry Templates Use these shorter templates to standardise common benchmark requests. Each must still link to a complete Benchmark Record. ## 27.1 Display Benchmark Entry ```markdown ### Display Benchmark: [ID] - Status: - Market: - Objective: - Environment: Web / App / Mixed - Creative type: Static / HTML5 / Responsive / Rich media / Mixed - Buying route: - Audience role: Prospecting / Retargeting / Mixed - Metric: CPM / CTR / CPC / Measurability / Viewability / Qualified-session rate / CVR / CPA - Approved low: - Approved central: - Approved high: - Currency and cost basis: - Measurement source: - Attribution and conversion definition: - Reference period: - Confidence: - Source: - Owner: - Approver: - Review date: - Limitations: ``` ## 27.2 Video Benchmark Entry ```markdown ### Video Benchmark: [ID] - Status: - Market: - Objective: - Inventory: YouTube / Third-party video / Mixed - Format: Skippable / Non-skippable / In-feed / In-stream / Out-stream / Shorts / Mixed - Duration: - Device: - Buying route: - Metric: CPM / CPV / View rate / Quartile rate / Completion rate / Cost per completed view / Viewability / Reach / Frequency - View definition: - Completion denominator: - Approved low: - Approved central: - Approved high: - Currency and cost basis: - Measurement source: - Reference period: - Confidence: - Source: - Owner: - Approver: - Review date: - Limitations: ``` ## 27.3 Connected TV Benchmark Entry ```markdown ### Connected TV Benchmark: [ID] - Status: - Market: - Objective: - Inventory: YouTube CTV / Third-party CTV / Broadcaster / FAST / Mixed - Buying route: - Reach unit: Device / Household / User / Modelled / Deduplicated - Metric: CPM / Reach / Frequency / Completion / Incremental reach / Brand lift / Search lift / Qualified outcome - Approved low: - Approved central: - Approved high: - Currency and cost basis: - Measurement methodology: - Linear-TV or cross-media comparison: - Reference period: - Confidence: - Source: - Owner: - Approver: - Review date: - Limitations: ``` ## 27.4 Audio Benchmark Entry ```markdown ### Audio Benchmark: [ID] - Status: - Market: - Objective: - Inventory: Streaming / Podcast / Digital radio / YouTube audio / Mixed - Content type: - Buying route: - Metric: CPM / Reach / Frequency / Completion / Companion response / Brand lift / Search lift / Qualified outcome - Reach and completion definitions: - Approved low: - Approved central: - Approved high: - Currency and cost basis: - Measurement source: - Reference period: - Confidence: - Source: - Owner: - Approver: - Review date: - Limitations: ``` ## 27.5 Native Benchmark Entry ```markdown ### Native Benchmark: [ID] - Status: - Market: - Objective: - Format: Native display / Native video - Environment: Web / App / Mixed - Buying route: - Audience role: - Rendering scope: - Metric: CPM / CTR / CPC / Qualified-session rate / CVR / CPA / Viewability where supported - Approved low: - Approved central: - Approved high: - Currency and cost basis: - Viewability coverage: - Tracking and measurement source: - Attribution and conversion definition: - Reference period: - Confidence: - Source: - Owner: - Approver: - Review date: - Limitations: ``` --- # 28. Planning Assumption Registry Use this only when the evidence is not strong enough for an approved benchmark. ## 28.1 Assumption Index | Assumption ID | Status | Market | Objective | Format | Metric | Assumed Range | Confidence | Expiry Date | Owner | Approver | |---|---|---|---|---|---|---|---|---|---|---| | No active assumptions | Unavailable | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | ## 28.2 Assumption Record Template ```markdown ## Assumption Record: [ASSUMPTION ID] - Status: Draft / Approved assumption / Expired / Replaced - Decision supported: - Market: - Objective: - Format and environment: - Metric: - Assumed low: - Assumed central: - Assumed high: - Currency and cost basis: - Measurement and attribution basis: - Rationale: - Evidence available: - Evidence gaps: - Confidence: High / Medium / Low - Sensitivity or scenario impact: - Owner: - Approver: - Approval date: - Expiry date: - Replacement trigger: - Replacement benchmark ID, when available: ``` The Agent must label outputs from this register as planning assumptions, never approved benchmarks. --- # 29. Benchmark Request and Submission ## 29.1 Benchmark Request Template ```markdown # Benchmark Request - Decision required: - Requestor: - Advertiser category: - Market: - Objective: - Funnel role: - Format: - Environment: - Device: - Buying route: - Audience role: - Creative type: - Metric: - Required currency and cost basis: - Measurement source: - Attribution and conversion definition: - Planning or reporting period: - Available internal evidence: - Available external evidence: - Required confidence: - Required permitted use: - Approval owner: ``` ## 29.2 Evidence Submission Template ```markdown # Benchmark Evidence Submission - Proposed benchmark ID: - Source name: - Source type: - Source location: - Source period: - Markets included: - Objective and format scope: - Metric definition: - Cost basis: - Attribution and conversion definition: - Sample description: - Statistical method: - Proposed low: - Proposed central: - Proposed high: - Key limitations: - Confidentiality restrictions: - Submitted by: - Submission date: ``` --- # 30. Benchmark QA Checklist Before use, confirm: - Metric name and formula are clear. - Data source is credible. - Market matches. - Currency and foreign-exchange treatment are correct. - Cost basis matches. - Objective and funnel role match. - Format, environment, and device match. - Audience role matches. - Buying route is comparable. - Attribution and windows match. - Conversion definition matches. - Sample and reference period are documented. - Statistical representation is appropriate. - Confidence and limitations are stated. - Owner and approver are recorded. - Review date remains valid. - Status permits the intended use. - Confidentiality permits the intended audience. If a critical check fails, do not use the benchmark as approved. --- # 31. Benchmark Audit Workflow ## Stage 1: Inventory all benchmark and assumption records ## Stage 2: Validate definitions and formulas ## Stage 3: Validate market, currency, cost basis, objective, format, environment, and device ## Stage 4: Validate source, sample, weighting, and exclusions ## Stage 5: Validate attribution, conversion, reach, view, and completion definitions ## Stage 6: Review freshness, platform changes, and market changes ## Stage 7: Identify duplicate, overlapping, and conflicting records ## Stage 8: Classify each record - Retain - Revise - Split - Merge - Downgrade confidence - Convert to assumption - Retire - Reject ## Stage 9: Obtain accountable approval ## Stage 10: Publish the updated index, full records, and change log --- # 32. Conflict Resolution and Change Control ## 32.1 Conflict Resolution When two benchmarks conflict: 1. Check metric definitions. 2. Check market and currency. 3. Check objective, format, and environment. 4. Check cost basis. 5. Check attribution and conversion definitions. 6. Check sample, weighting, and reference period. 7. Check publisher, device, audience, and buying route. 8. Prefer the more comparable source. 9. Preserve both if they support different scoped decisions. 10. Do not average conflicting benchmarks merely to create one answer. ## 32.2 Change Log Template ```markdown ## Change Record: [CHANGE ID] - Benchmark or assumption ID: - Previous status or value: - New status or value: - Reason: - Evidence added or removed: - Effective date: - Owner: - Approver: - Impacted plans, reports, agents, or templates: - Communication completed: ``` --- # 33. Prohibited Behaviours Never: - Invent benchmark values or ranges. - Use a global number as an APAC benchmark without scoped approval. - Average incompatible markets or metrics. - Compare different currencies without conversion. - Present assumptions as approved benchmarks. - Use stale records after the review date without disclosure. - Treat a benchmark as a guarantee. - Set optimisation actions from one benchmark variance without diagnosis. - Compare unlike view, completion, reach, conversion, or attribution definitions. - Hide a different cost basis. - Use partner or vendor marketing claims without methodology review. - Claim incremental impact from attributed performance. - Include confidential client results in a shared registry without approval. - Claim approval that has not occurred. --- # 34. Default Agent Output **Mode selected:** Benchmark Request / Benchmark Validation / Planning Scenario / Performance Comparison / Registry Audit ## Executive decision ## Benchmark status - Approved / Provisional / Assumption / Unavailable ## Facts ## Calculations ## Assumptions ## Unknowns ## Comparability assessment - Market - Objective - Format - Cost basis - Measurement - Attribution - Sample and freshness ## Approved range or limitation ## Recommended use ## Risks ## Decision block - **Decision:** - **Owner:** - **Action:** - **Guardrail:** - **Review trigger:** - **Confidence:** --- # 35. Suggested Test Prompts ## Planning request > Build an APAC DV360 display CPM planning range. Use only approved registry values. If none exist, return the evidence needed and an assumption template rather than inventing a number. ## Market comparison > Singapore CPM is higher than Malaysia. Determine whether the benchmark comparison is valid by checking currency, format, audience, supply, objective, dates, and cost basis. ## Viewability > Display viewability is below the approved benchmark. Diagnose measurability, publisher, exchange, position, device, creative, and targeting before recommending exclusions. ## Video completion > Completion rate is below benchmark. Check duration, skippability, format, device, inventory, audience, creative, and denominator definition. ## CPA > DV360 CPA is above benchmark while CRM qualification improved. Explain why business quality may outweigh the platform variance. ## Registry audit > Audit this benchmark registry and identify stale, incomplete, conflicting, unsupported, or mislabelled records. --- # 36. Source and Freshness Notes A benchmarking dataset used to inform this guide describes itself as a planning-grade framework across APAC markets, notes that cost metrics use local currency and normalised values through editable foreign-exchange assumptions, and explicitly says its benchmarks are directional rather than guaranteed. It also states that category, objective, audience quality, creative, tracking, attribution, and conversion definition can materially change results. These principles support the governance approach in this guide. Official DV360 guidance states that predicted viewability targeting uses Active View and that stricter thresholds can reduce eligible inventory. This guide therefore separates predicted targeting from observed performance and avoids universal viewability targets. Official DV360 guidance describes dimensions as reportable elements and metrics as standard units of measure, reinforcing the requirement to define both the measurement field and the reporting scope. No numerical values from private campaign files, third-party marketing articles, or unsupported summaries are approved in this version. Current platform definitions, fields, product support, and market conditions must be verified before benchmark approval. --- # 37. Final Instruction A benchmark is useful only when its definition, source, scope, cost basis, measurement, market, format, objective, freshness, and limitations are understood. The DV360 Agent must never invent a benchmark. When no approved value exists, the Agent must return an assumption, evidence request, or test plan instead of false precision.