What advertising analytics means in practice广告分析在实践中意味着什么
Advertising analytics is the disciplined use of delivery, cost, audience, conversion, and business-outcome data to understand paid-media performance and decide what to do next. It goes beyond reading a dashboard: the work reconciles sources, defines metrics, tests explanations, states uncertainty, and separates attributed credit from incremental effect.
广告分析是对投放、成本、受众、转化和商业结果数据进行规范使用，以理解付费媒体表现并决定下一步行动。它不只是阅读仪表板，还要对账数据源、定义指标、检验解释、说明不确定性，并区分归因功劳与增量效果。
A useful analysis begins with a decision such as whether to pause a creative, repair a landing page, change a bid, expand an audience, or reallocate a channel budget. It then identifies the evidence needed for that decision. Platform-reported impressions and conversions are valuable operational signals, but they may use different identity rules, attribution windows, time zones, or modeled estimates than web analytics, CRM, billing, or finance data.
有用的分析从决策开始，例如是否暂停素材、修复落地页、调整出价、扩大受众或重新分配渠道预算，然后确定该决策所需证据。平台报告的展示和转化是重要运营信号，但其身份规则、归因窗口、时区或模型估算可能不同于网站分析、CRM、计费或财务数据。
Scope boundary: use this guide for your own paid-media campaign performance. Competitive ad intelligence—tracking other brands' creatives, placements, or estimated spend—is a different data and research problem. For the broader discipline across paid, owned, product, CRM, and market signals, use the Marketing Data Analysis guide.
范围边界：本指南用于分析自身付费媒体活动表现。追踪其他品牌素材、版位或估算花费的竞争广告情报属于不同的数据与研究问题。如需了解涵盖付费、自有渠道、产品、CRM 和市场信号的更广泛学科，请参阅营销数据分析指南。
When advertising analytics helps—and when it does not广告分析何时有用，以及何时不够
Detect delivery failures, spend pacing, cost inflation, frequency buildup, tracking breaks, and sudden conversion-rate changes.
发现投放失败、花费节奏、成本上涨、频次累积、追踪中断和转化率突变。
Trace a result through the metric tree to distinguish auction cost, click response, landing-page behavior, lead quality, and revenue lag.
沿指标树追踪结果，区分竞价成本、点击响应、落地页行为、线索质量和收入滞后。
Compare scenarios using reconciled cost, capacity, marginal performance, uncertainty, and channel constraints—not one platform's reported return alone.
使用已对账成本、容量、边际表现、不确定性和渠道约束比较情景，而不是只看单个平台报告回报。
Compare descriptive reporting, attribution, experiments, and longer-horizon models so each method answers the question it can support.
比较描述性报告、归因、实验和长期模型，使每种方法只回答其证据能够支持的问题。
Advertising analytics is not enough when the question concerns creative quality before launch, brand meaning without appropriate research, legal compliance, or causal lift without a credible counterfactual. It also cannot rescue missing conversion instrumentation, uncontrolled campaign naming, or a business outcome that occurs outside every available source. In those cases, fix collection, add qualitative research, obtain legal advice, or design an appropriate experiment before asking the dashboard for a precise answer.
当问题涉及上线前素材质量、缺少适当研究时的品牌含义、法律合规，或没有可信反事实时的因果提升，广告分析本身并不够。它也无法挽救缺失的转化埋点、失控的活动命名，或所有可用数据源都没有记录的业务结果。此时应先修复采集、补充定性研究、获得法律意见或设计适当实验，而不是要求仪表板给出精确答案。
Prepare an advertising analytics data contract before analysis分析前先准备广告分析数据合同
Create a small data contract that names the owner, grain, identifiers, time zone, currency, update latency, retention rule, and metric definitions for each source. Decide which system is authoritative for each fact. An ad platform is usually the operational source for delivered spend and impressions; a governed order or billing system may be stronger for recognized revenue and refunds. Preserve raw exports, transformation versions, and reconciliation totals so an analyst can reproduce the result.
为每个数据源建立小型数据合同，记录责任人、粒度、标识符、时区、币种、更新延迟、保留规则和指标定义，并决定每项事实由哪个系统作为权威来源。广告平台通常是已投放花费和展示的运营来源；受治理的订单或计费系统可能更适合确认收入和退款。保留原始导出、转换版本和对账总数，使分析结果可重现。
| Source来源 | Use it for用于 | Required controls必要控制 | Common mismatch常见错配 |
|---|---|---|---|
| Ad platforms广告平台 | Spend, delivery, auction, placement, creative, platform-attributed actions花费、投放、竞价、版位、素材、平台归因行动 | Account and campaign IDs, currency, time zone, attribution setting, export timestamp账户与活动 ID、币种、时区、归因设置、导出时间 | Platform totals use different windows or modeled conversions平台总数使用不同窗口或模型转化 |
| Web or app analytics网站或应用分析 | Sessions, events, landing behavior, source and campaign dimensions会话、事件、落地行为、来源与活动维度 | UTM policy, click-ID handling, consent state, event versions, cross-domain rulesUTM 规范、点击 ID 处理、同意状态、事件版本、跨域规则 | Identity loss, self-referrals, overwritten campaign parameters身份丢失、自引荐、活动参数被覆盖 |
| CRM or lead systemCRM 或线索系统 | Qualified leads, stages, owners, opportunities, delayed outcomes合格线索、阶段、负责人、商机、延迟结果 | Stable lead and campaign keys, stage history, duplicate and reopen rules稳定线索与活动键、阶段历史、重复与重开规则 | Latest state replaces history; offline source is missing最新状态覆盖历史；线下来源缺失 |
| Orders, billing, or finance订单、计费或财务 | Recognized revenue, margin, refunds, taxes, cancellations, customer status确认收入、利润、退款、税费、取消、客户状态 | Order and customer keys, accounting date, currency conversion, refund policy订单与客户键、会计日期、汇率转换、退款规则 | Gross sales are compared with net revenue or mixed periods总销售额与净收入或不同周期被混比 |
Before joining data, write the grain explicitly. A campaign-day table cannot safely join directly to an order-line table without aggregation or a documented bridge. Define how one-to-many relationships, late conversions, duplicate click IDs, missing campaign IDs, refunds, and channel reassignment behave. Aggregate when user-level identity is unnecessary, restrict access to raw identifiers, and retain only what the stated purpose requires.
连接数据前必须明确写出粒度。活动-日期表不能在未聚合或缺少已记录桥接规则时直接连接订单行表。应定义一对多关系、延迟转化、重复点击 ID、缺失活动 ID、退款和渠道重分配的处理方式。无需用户级身份时应聚合数据，限制原始标识符访问，并只保留既定目的所需信息。
Choose advertising analytics metrics from the decision backward从决策反推广告分析指标
Do not ask one metric to represent the whole system. Build a metric tree from delivery to response to commercial outcome, and pair each ratio with its numerator, denominator, sample size, and time window. Segment new versus returning customers when acquisition is the decision. Use contribution margin rather than revenue when product costs differ materially. For delayed purchases, use mature cohorts or clearly label incomplete revenue.
不要让一个指标代表整个系统。应从投放、响应到商业结果构建指标树，并为每个比率保留分子、分母、样本量和时间窗口。决策涉及获客时，应区分新客与老客；产品成本差异显著时，应使用贡献利润而非收入；购买存在延迟时，应使用成熟队列或明确标记未完整收入。
| Question问题 | Useful measures可用指标 | Calculation计算 | Interpretation warning解释警告 |
|---|---|---|---|
| Did delivery occur efficiently?投放是否有效执行？ | Reach, frequency, CPM, viewability, impression share覆盖、频次、CPM、可见率、展示份额 | CPM = spend ÷ impressions × 1,000CPM = 花费 ÷ 展示 × 1,000 | Cheap impressions may be low quality or outside the eligible audience便宜展示可能质量低或不在合格受众内 |
| Did people respond?用户是否响应？ | CTR, CPC, landing-page engagement, qualified visit rateCTR、CPC、落地页互动、合格访问率 | CTR = clicks ÷ impressions; CPC = spend ÷ clicksCTR = 点击 ÷ 展示；CPC = 花费 ÷ 点击 | A click can reflect curiosity, accidental action, or poor expectation setting点击可能来自好奇、误触或预期设置不当 |
| Did the desired action occur?期望行动是否发生？ | CVR, CPA, qualified lead rate, cost per qualified leadCVR、CPA、合格线索率、每个合格线索成本 | CVR = conversions ÷ eligible visits; CPA = spend ÷ conversionsCVR = 转化 ÷ 合格访问；CPA = 花费 ÷ 转化 | Conversion definitions, duplication, and windows must match转化定义、去重和窗口必须一致 |
| Did the economics work?经济性是否成立？ | New-customer CAC, ROAS, contribution return, payback, LTV:CAC新客 CAC、ROAS、贡献回报、回收期、LTV:CAC | ROAS = attributed revenue ÷ ad spendROAS = 归因收入 ÷ 广告花费 | Attributed revenue is not automatically incremental revenue or profit归因收入不自动等于增量收入或利润 |
Treat a dashboard as a monitoring interface, not a verdict. Show actual versus target, comparison period, freshness, attribution setting, exclusions, and reconciliation status. A statistically noisy small campaign should not receive the same confidence as a stable mature cohort. When multiple metrics move, trace the identity spend → impressions → clicks → eligible visits → conversions → qualified outcomes → revenue or margin before choosing an explanation.
应把仪表板视为监控界面，而不是最终裁决。展示实际值与目标、对比周期、新鲜度、归因设置、排除项和对账状态。统计噪声较大的小活动不应获得与稳定成熟队列相同的置信度。多个指标同时变化时，应沿花费 → 展示 → 点击 → 合格访问 → 转化 → 合格结果 → 收入或利润追踪，再选择解释。
A repeatable paid media analytics workflow可重复的付费媒体分析工作流
- Define the decision and evidence threshold定义决策与证据门槛State the owner, deadline, eligible campaigns, outcome, guardrails, and what result would trigger pause, repair, test, scale, or no change. Avoid the vague request “tell me what happened.”写明负责人、截止时间、合格活动、结果、保护指标，以及何种结果会触发暂停、修复、测试、扩量或不变。避免“告诉我发生了什么”这种模糊要求。
- Freeze scope and comparison rules冻结范围与比较规则Choose event time or processing time, time zone, currency, conversion definition, maturity window, customer type, and comparison period. Record campaign hierarchy and any renamed or merged entities.选择事件时间或处理时间、时区、币种、转化定义、成熟窗口、客户类型和对比周期，并记录活动层级以及重命名或合并实体。
- Extract and preserve source evidence提取并保留来源证据Save versioned platform, analytics, CRM, and finance exports with timestamps and filters. Keep raw files immutable, restrict sensitive access, and note missing dates, API sampling, modeled fields, or backfills.保存带时间戳和筛选条件的版本化平台、分析、CRM 与财务导出。原始文件保持不可变，限制敏感访问，并记录缺失日期、API 抽样、模型字段或回填。
- Reconcile before joining先对账再连接Compare spend, impressions, clicks, and source counts by account and day. Explain variance using currency, time zone, taxes, invalid traffic adjustments, attribution windows, privacy thresholds, and extraction latency.按账户和日期比较花费、展示、点击和来源计数，并用币种、时区、税费、无效流量调整、归因窗口、隐私阈值和提取延迟解释差异。
- Build the metric tree at a safe grain在安全粒度构建指标树Aggregate each source to the approved campaign-day or campaign-cohort grain, apply the campaign-key map, calculate denominators once, and retain unmatched and duplicate counts rather than silently dropping them.把各来源聚合到批准的活动-日期或活动-队列粒度，应用活动键映射，统一计算分母，并保留未匹配与重复计数，而不是静默删除。
- Diagnose changes and challenge the story诊断变化并质疑叙事Start with the largest absolute contribution, then segment one meaningful dimension at a time. Check tracking, mix, auction cost, creative, audience, landing experience, offer, sales follow-up, seasonality, and conversion lag before treating correlation as cause.先找绝对贡献最大的变化，再一次细分一个有意义维度。依次检查追踪、组合、竞价成本、素材、受众、落地体验、优惠、销售跟进、季节性和转化滞后，避免把相关性当因果。
- Choose an action and verification plan选择行动与验证计划Record the evidence, uncertainty, decision, owner, change size, guardrails, next review date, and how impact will be tested. Use attribution for operational credit; use experiments or another credible causal design when incrementality is the decision.记录证据、不确定性、决策、负责人、变更幅度、保护指标、下次复核日期和影响检验方法。归因用于运营功劳分配；当决策关注增量时，应使用实验或其他可信因果设计。
Advertising attribution is not the same as incrementality广告归因不等于增量效果
Attribution assigns credit for observed conversions to touchpoints under a rule or model. It is useful for reporting and many within-channel operating decisions. Incrementality asks what outcomes the advertising caused compared with what would have happened without it. These are related but different questions. A retargeting campaign can show strong attributed ROAS because it reaches people already close to buying, yet generate a smaller incremental effect.
归因按照规则或模型，把已观察转化的功劳分配给触点，适用于报告和许多渠道内运营决策。增量则询问：与没有广告时相比，广告实际造成了多少结果。两者相关但不是同一个问题。再营销活动可能因为触达本就接近购买的人而显示很高归因 ROAS，但增量效果较小。
| Method方法 | Best-supported question最适合回答的问题 | Strength优势 | Limit限制 |
|---|---|---|---|
| Descriptive reporting描述性报告 | What was delivered and observed?实际投放和观察到什么？ | Fast, granular, operational快速、细粒度、适合运营 | Does not assign cause不能确定因果 |
| Attribution归因 | How should observed conversion credit be allocated?如何分配已观察转化功劳？ | Supports journey and tactic reporting支持旅程与策略报告 | Depends on identity, coverage, windows, and model assumptions依赖身份、覆盖、窗口和模型假设 |
| Randomized lift test随机提升实验 | What did this campaign cause for the tested population?该活动对实验人群造成了什么？ | Credible counterfactual when implemented well实施良好时具有可信反事实 | Needs power, compliance, stable treatment, and careful generalization需要统计功效、合规、稳定处理和谨慎外推 |
| Marketing mix modeling营销组合建模 | How did channels relate to outcomes over longer periods?较长周期内渠道与结果如何关联？ | Can include offline media and aggregate data可纳入线下媒体与聚合数据 | Requires sufficient variation, controls, diagnostics, and calibration需要足够变化、控制、诊断与校准 |
Use multiple methods as a measurement system rather than forcing agreement at every campaign row. Attribution can guide frequent tactical decisions; experiments can calibrate causal expectations; longer-horizon modeling can inform portfolio allocation when its assumptions and uncertainty are appropriate. Document which estimate is being used, its scope, and why it fits the decision.
应把多种方法作为衡量系统，而不是强迫每个活动行完全一致。归因可指导频繁战术决策；实验可校准因果预期；当假设和不确定性适当时，长期模型可支持组合分配。必须记录所用估计、适用范围以及它为何适合该决策。
Hypothetical example: diagnosing a cross-channel ROAS decline假设示例：诊断跨渠道 ROAS 下降
This is a hypothetical example, not customer data or a benchmark. A team observes that attributed ROAS fell from 4.0 to 3.1 between two four-week periods. Spend increased from $100,000 to $120,000, while attributed revenue fell from $400,000 to $372,000. The headline alone does not identify a cause.
这是一个假设示例，不是客户数据或行业基准。某团队发现，两个四周周期之间的归因 ROAS 从 4.0 降至 3.1。花费从 100,000 美元增至 120,000 美元，而归因收入从 400,000 美元降至 372,000 美元。标题指标本身不能识别原因。
- Reconcile: platform spend matches finance within the approved tolerance after currency and tax treatment. One social export changed time zone mid-period and is corrected.对账：处理币种与税费后，平台花费与财务数据在批准容差内一致；某社交平台导出在周期中途更改时区，现已修正。
- Decompose: CPM rose 12%, CTR was nearly flat, landing-page CVR fell 18%, and average order value was stable. Most absolute revenue loss came from two mobile landing pages, not every channel.分解：CPM 上升 12%，CTR 基本持平，落地页 CVR 下降 18%，平均订单价值稳定。绝大多数收入损失来自两个移动端落地页，而非所有渠道。
- Challenge: the pages received a release during the period. Device mix also shifted, and conversions have a seven-day lag. The team checks page errors, mature cohorts, offer availability, and source-specific identity loss before naming the release as the cause.质疑：这些页面在周期内发布了新版本，同时设备组合变化，转化还有七天延迟。团队先检查页面错误、成熟队列、优惠可用性和来源特定身份丢失，再判断发布是否为原因。
- Act and verify: repair the affected page, hold unaffected campaigns stable, restore spend gradually under CPA and margin guardrails, and run a controlled page test where feasible. Do not claim that the observed ROAS difference is incremental lift.行动与验证：修复受影响页面，保持未受影响活动稳定，在 CPA 与利润保护指标下逐步恢复花费，并在可行时运行受控页面实验。不要把观察到的 ROAS 差异称为增量提升。
The example shows why campaign performance analysis should preserve the metric tree and operational timeline. A blanket budget cut would treat every channel as equally responsible, hide the landing-page failure, and reduce learning. The defensible output is a bounded diagnosis, a reversible action, and a verification plan—not a confident story assembled from one ratio.
该示例说明，活动表现分析为何必须保留指标树和运营时间线。全面削减预算会把所有渠道视为同等责任，掩盖落地页故障并减少学习。可辩护的输出应是有边界的诊断、可逆行动和验证计划，而不是从单一比率拼出的自信故事。
Common advertising analytics errors, limits, and privacy risks常见广告分析错误、限制与隐私风险
Different currencies, windows, time zones, conversion definitions, and refunds can create a false performance gap.
不同币种、窗口、时区、转化定义和退款处理会制造虚假表现差异。
Joining campaign-day spend to many order lines can multiply cost or revenue. Aggregate first and test row counts.
把活动-日期花费连接到多个订单行会放大成本或收入。应先聚合并检查行数。
CTR or cheap CPA can improve while lead quality, margin, or retention deteriorates. Keep outcome and guardrail metrics visible.
CTR 或低 CPA 改善时，线索质量、利润或留存可能恶化。必须保持结果与保护指标可见。
A model's assigned credit reflects observed data and assumptions. It does not construct the counterfactual required for incrementality.
模型分配的功劳反映观察数据与假设，并不构建增量所需反事实。
Recent cohorts have less time to convert, while retargeting audiences may already have higher purchase intent.
近期队列转化时间更短，再营销受众可能本就具有更高购买意愿。
User-level stitching increases privacy, security, and governance risk. Prefer aggregate analysis when it answers the decision.
用户级拼接增加隐私、安全与治理风险。聚合分析能够回答决策时，应优先使用聚合数据。
Advertising measurement can involve cookies, device identifiers, click IDs, audience data, and data sharing across services. Requirements depend on jurisdiction, role, technology, and purpose. Use data minimization, access control, retention limits, consent or other lawful mechanisms where required, and documented vendor responsibilities. The UK ICO guidance on cookies and similar technologies is an official reference for one regulatory context, not universal legal advice.
广告衡量可能涉及 Cookie、设备标识符、点击 ID、受众数据和跨服务数据共享。要求取决于司法辖区、角色、技术与目的。应采用数据最小化、访问控制、保留期限，并在需要时使用同意或其他合法机制，同时记录供应商责任。英国 ICO 关于 Cookie 与类似技术的指南是特定监管语境下的官方参考，不构成普遍法律建议。
Validate the analysis and establish an operating cadence验证分析并建立运营节奏
Before a result informs spend, another reviewer should be able to reproduce the totals, inspect exclusions, trace a sample record through transformations, and explain the method in plain language. Keep a decision log containing the question, versioned inputs, metric definitions, findings, alternatives considered, uncertainty, action, owner, and review date.
在结果用于花费决策前，另一位复核者应能重现总数、检查排除项、沿转换追踪样本记录，并用普通语言解释方法。决策日志应包含问题、版本化输入、指标定义、发现、备选解释、不确定性、行动、负责人和复核日期。
- Daily or intraday: delivery, pacing, disapprovals, tracking failures, extreme cost changes, and freshness alerts.每日或日内：投放、节奏、拒登、追踪失败、极端成本变化和新鲜度告警。
- Weekly: mature performance by campaign and audience, metric-tree decomposition, creative or landing-page signals, and actions under pre-agreed guardrails.每周：按活动与受众查看成熟表现、分解指标树、检查素材或落地页信号，并在预先约定保护指标下行动。
- Monthly: cross-channel reconciliation, new-customer economics, cohort lag, margin, attribution sensitivity, and data-quality exceptions.每月：跨渠道对账、新客经济性、队列滞后、利润、归因敏感性和数据质量例外。
- Quarterly or campaign-cycle: incrementality experiments, measurement-plan review, taxonomy cleanup, vendor changes, privacy controls, and portfolio allocation.每季度或活动周期：增量实验、衡量计划复核、分类清理、供应商变化、隐私控制和组合分配。
Stop conditions matter. Pause automated or manual recommendations when source freshness fails, unmatched keys exceed tolerance, a conversion definition changes, a cohort is immature, or sample sizes cannot support the requested segment. “No decision yet” is often more responsible than a precise but unstable ranking.
停止条件很重要。当数据新鲜度失败、未匹配键超过容差、转化定义变化、队列尚未成熟或样本量无法支持所请求细分时，应暂停自动或人工建议。“暂不决策”通常比精确但不稳定的排名更负责任。
Use InfiniSynapse after advertising data is prepared广告数据准备完成后使用 InfiniSynapse
An AI analysis layer can reduce the friction of asking follow-up questions across prepared paid-media, analytics, CRM, and revenue sources. It does not replace reliable collection, lawful data use, a campaign-key map, agreed metrics, or causal method selection. The analyst remains responsible for checking source coverage, transformations, alternative explanations, uncertainty, and whether the conclusion supports the proposed action.
AI 分析层可以降低跨已准备付费媒体、分析、CRM 和收入数据源追问的摩擦，但不能替代可靠采集、合法数据使用、活动键映射、已约定指标或因果方法选择。分析人员仍需检查数据源覆盖、转换、备选解释、不确定性，以及结论是否足以支持拟议行动。
Before opening the app, prepare governed exports or connected sources, a campaign and account key map, documented metric definitions, the decision and time window, attribution assumptions, source reconciliation totals, and known limitations. Then use InfiniSynapse for natural-language, multi-source analysis while a responsible reviewer validates the evidence and conclusions.
打开应用前，请准备受治理的导出或已连接数据源、活动与账户键映射、记录完善的指标定义、决策与时间窗口、归因假设、来源对账总数和已知限制。随后可使用 InfiniSynapse 进行自然语言多源分析，同时由负责人员验证证据与结论。
Open the InfiniSynapse analysis app打开 InfiniSynapse 分析应用InfiniSynapse is not described here as the native collection system for every advertising platform, a media-buying automation service, or proof of causal lift. Its evidenced role is a professional AI data analyst for natural-language and multi-source analysis. If the task is comparing broader analysis software categories and implementation tradeoffs, use the Data Analytics Software guide.
本页不会把 InfiniSynapse 描述为每个广告平台的原生采集系统、媒体购买自动化服务或因果提升证明。其已有定位是支持自然语言与多源分析的专业 AI 数据分析工具。如果任务是比较更广泛的分析软件类别与实施权衡，请参阅数据分析软件指南。
Advertising analytics frequently asked questions广告分析常见问题
Advertising analytics is the disciplined use of delivery, cost, audience, conversion, and business-outcome data to describe paid-media performance, diagnose changes, evaluate evidence, and choose the next campaign or budget action.
广告分析是对投放、成本、受众、转化和商业结果数据进行规范使用，以描述付费媒体表现、诊断变化、评估证据，并选择下一项活动或预算行动。
Choose metrics from the decision backward. Delivery decisions may require reach, frequency, CPM, and viewability; response decisions may use CTR, CPC, CVR, and CPA; commercial decisions need reconciled revenue, contribution margin, CAC, payback, and carefully interpreted ROAS.
应从决策反推指标。投放决策可能需要覆盖、频次、CPM 和可见率；响应决策可使用 CTR、CPC、CVR 和 CPA；商业决策需要已对账收入、贡献利润、CAC、回收期，以及经过谨慎解释的 ROAS。
Define the decision and eligible campaign scope, reconcile platform delivery and cost, join conversion and business outcomes at a governed grain, calculate a metric tree, segment one dimension at a time, test alternative explanations, and record the next action with a verification plan.
先定义决策和合格活动范围，对账平台投放与成本，在受治理粒度连接转化和商业结果，计算指标树，每次细分一个维度，检验备选解释，并记录带验证计划的下一步行动。
No. ROAS divides attributed revenue by advertising spend under a reporting rule. Incremental return estimates outcomes caused by the advertising compared with a credible counterfactual, usually through an experiment or another causal design.
不同。ROAS 按报告规则用归因收入除以广告花费；增量回报则通过实验或其他因果设计，与可信反事实比较，估计广告实际造成的结果。
They can differ because of time zones, currencies, conversion definitions, lookback windows, view-through rules, identity loss, modeled conversions, deduplication, refunds, late revenue, and different campaign identifiers. Reconcile definitions before judging performance.
差异可能来自时区、币种、转化定义、回溯窗口、浏览归因规则、身份丢失、模型转化、去重、退款、延迟收入和不同活动标识符。判断表现前必须先对账定义。
With governed exports or connected sources, documented campaign keys and metrics, a clear decision, and reconciliation totals, InfiniSynapse can support natural-language, multi-source analysis. Reviewers still need to validate transformations, assumptions, evidence, and limitations.
在具备受治理导出或已连接来源、记录完善的活动键与指标、明确决策和对账总数后，InfiniSynapse 可支持自然语言多源分析；复核者仍需验证转换、假设、证据和限制。
Official advertising analytics sources and verification references广告分析官方来源与验证参考
- Google Analytics official guide to advertising and attribution reportsGoogle Analytics 广告与归因报告官方指南 — supports the advertising-reporting and attribution context.支持广告报告与归因语境。
- Google Analytics official traffic-source dimension documentationGoogle Analytics 流量来源维度官方文档 — documents campaign dimensions and how advertising traffic-source data can be derived.记录活动维度以及广告流量来源数据的形成方式。
- Google Analytics official campaign data import guidanceGoogle Analytics 活动数据导入官方指南 — supports combining non-Google campaign data with key events and revenue.支持将非 Google 活动数据与关键事件和收入结合。
- IAB UK Measuring Digital Advertising toolkitIAB UK 数字广告衡量工具包 — provides an industry measurement framework and cautions about method limits.提供行业衡量框架及方法限制提示。
- UK ICO guidance on cookies and similar technologies英国 ICO 关于 Cookie 与类似技术的指南 — an official privacy reference relevant to some advertising measurement implementations.与部分广告衡量实施相关的官方隐私参考。