On this page本页目录
Returns analysis example, in one minute: this synthetic 90-day cohort contains 8,500 eligible shipped units and 797 returned units, for a 9.38% unit return rate. Refund value is 10.92% of eligible sales. Return-related operating cost is $17,173, or $21.55 per returned unit. Fit is the leading recorded reason at 38.27%. SKU-A has the highest return rate at 18%, but SKU-E has the largest modeled economic loss at $17,550, so rate alone would produce the wrong first priority.
一分钟看懂这个退货分析示例：该模拟的 90 天批次包含 8,500 件符合口径的已发货商品和 797 件退货，件数退货率为 9.38%。退款金额占符合口径销售额的 10.92%。退货相关运营成本为 17,173 美元，单件退货成本为 21.55 美元。尺码/适配是最多的记录原因，占 38.27%。SKU-A 的退货率最高，为 18%，但 SKU-E 的模拟经济损失最大，为 17,550 美元，因此只按退货率排序会选错第一优先级。
The conclusion is not “fix SKU-E immediately.” The defensible conclusion is “investigate SKU-E first.” Its 60 damage codes and 40 not-as-described codes are observed labels, not proven root causes. Packaging photos, inspection outcomes, customer comments, batch data, and product-page changes are still needed before intervention.
这里的结论不是“立刻修复 SKU-E”，而是“优先调查 SKU-E”。其中 60 条破损代码和 40 条与描述不符代码是已观察标签，不是已经证实的根因。采取干预前仍需核对包装照片、质检结果、客户评论、生产批次和商品页变更。
Case setup: define the cohort before calculating anything案例口径：计算前先定义批次
This returns data analysis example uses eight fictional U.S. ecommerce SKUs. Eligible units were shipped from January 1 through March 31, 2026. Every unit had a 30-day return window, and the analysis was frozen after May 1 so late returns could not unfairly depress the newest cohort. Currency is USD. The denominator is shipped units, not orders, customers, or website sessions.
这个退货数据分析示例包含八个虚构的美国电商 SKU。符合口径的商品在 2026 年 1 月 1 日至 3 月 31 日发货，每件商品都适用 30 天退货窗口；分析在 5 月 1 日之后冻结，避免尚未成熟的新批次人为拉低退货率。币种为美元，分母是已发货件数，而不是订单数、客户数或网站会话数。
Eligible shipped units, returned units, eligible sales, refund value, four operating-cost components, four standardized reason groups, and sellable recovered units.
符合口径的已发货件数、退货件数、销售额、退款金额、四类运营成本、四个标准化原因组及可售回收件数。
Contribution margin, exchanges, customer cohorts, channel, warehouse, supplier batch, carrier, inspection photos, support text, and post-return repurchase.
贡献利润、换货、客户批次、渠道、仓库、供应商批次、承运商、质检照片、客服文本和退货后复购。
That missing-field list limits what can be claimed. For example, the sample can rank recorded loss but cannot prove a margin impact, attribute damage to a carrier, identify a supplier defect, or compare new and repeat customers.
这份缺失字段清单限定了可下结论的范围。例如，该样本可以按记录损失排序，但不能证明贡献利润影响，不能把破损归因给某家承运商，也不能识别供应商缺陷或比较新老客户。
Download the synthetic ecommerce returns analysis sample下载模拟电商退货分析样本
Eight-SKU sample dataset八个 SKU 的样本数据集
The CSV contains eight aggregate rows and 21 documented columns. It is synthetic, free to reuse for training, and contains no customer or personal data.
CSV 包含八条聚合数据和 21 个已说明字段。它是模拟数据，可免费用于训练，不含客户或个人数据。
↓ Download sample CSV下载样本 CSV| Field group字段组 | Included fields所含字段 | Use用途 |
|---|---|---|
| Cohort批次 | cohort_start, cohort_end, return_window_days, currency | Make time, maturity, and money comparable保证时间、成熟度和金额可比 |
| Demand需求 | eligible_units, eligible_sales_usd | Provide denominators提供分母 |
| Returns退货 | returned_units, refund_value_usd | Measure unit and value exposure separately分别衡量件数与金额敞口 |
| Operations运营 | shipping, handling, inspection, value_loss | Model return-related operating cost模拟退货相关运营成本 |
| Diagnosis诊断 | fit, damaged, not_as_described, changed_mind | Locate reason concentrations, not prove cause定位原因集中点，不用于证明根因 |
| Recovery回收 | sellable_recovered_units | Measure inventory outcome衡量库存处置结果 |
Step 1: clean and validate the returns dataset步骤一：清洗并校验退货数据
Before making a chart, reconcile the dataset. NIST's exploratory data analysis guidance starts with quality checks for entry problems, unusual values, and missing data. In a production return file, also confirm that return events join to order items, reason categories map consistently, money signs are normalized, and a refund without a physical return is not silently treated as returned inventory.
绘图前先完成数据对账。NIST 的探索性数据分析指南建议先检查录入问题、异常值和缺失数据。真实退货文件还要确认退货事件能关联到订单行、原因分类映射一致、金额正负号统一，并避免把“仅退款未实物退回”悄悄计入退货库存。
- Check keys and duplicates.检查主键与重复。 Aggregate only after each order-item or return-event key is unique at the intended grain.只有订单行或退货事件主键在目标粒度上唯一时才可汇总。
- Reconcile reason totals.对账原因合计。 For every sample row, fit + damaged + not_as_described + changed_mind equals returned_units.每一行的四类原因合计必须等于 returned_units。
- Test valid ranges.检查有效范围。 Returned and recovered units cannot be negative; recovered units cannot exceed returned units.退货与回收件数不能为负，可售回收件数不能超过退货件数。
- Freeze a mature cohort.冻结成熟批次。 Wait until the 30-day window closes for the last shipment before comparing rates.最后一批发货的 30 天窗口关闭后，再进行退货率比较。
- Record exclusions.记录排除项。 Document test orders, cancellations before shipment, warranty replacements, chargebacks, and unlinked refunds rather than guessing where they belong.记录测试订单、发货前取消、保修换新、拒付和无法关联的退款，不要猜测其归属。
Shopify's current reporting documentation also distinguishes physical returned quantity from broader sales reversals, which may include refunds, cancellations, edits, and other adjustments. Google Analytics similarly records refund events only when they are implemented and sent. Platform exports are inputs; your metric contract decides what belongs in the numerator.
Shopify 当前报告文档也把实体退货件数与更广义的销售冲销区分开，后者可能包含退款、取消、编辑和其他调整。Google Analytics 也只有在正确实施并发送 refund 事件后才会记录退款。平台导出只是输入，真正决定分子内容的是你的指标口径。
Step 2: calculate five baseline returns metrics步骤二：计算五项退货基线指标
1. Unit return rate1. 件数退货率
Use this to describe frequency at the unit grain. If your operation cares about orders, calculate a separate order return rate with returned orders over eligible orders; do not mix the two.
该指标描述件数粒度上的发生频率。如果运营更关注订单，则应另算“退货订单数 ÷ 符合口径订单数”，不要混用。
2. Refund-value rate2. 退款金额率
The value rate exceeds the unit rate, suggesting returned units are slightly more expensive than the average eligible unit. That is an observed mix effect, not proof that returns caused a margin decline.
金额率高于件数率，说明退回商品的平均售价略高于符合口径商品的平均售价。这是已观察到的结构效应，并不能证明退货导致贡献利润下降。
3–5. Operating cost, recovery, and reason mix3–5. 运营成本、回收与原因结构
This cost definition excludes refund principal because refund value is reported separately. It also excludes customer-support labor, payment fees, lost contribution margin, acquisition cost, warehouse occupancy, exchange economics, and future customer value because those fields are absent.
这里的成本口径不含退款本金，因为退款金额已经单独报告；也不含客服人力、支付费用、损失的贡献利润、获客成本、仓储占用、换货经济性和未来客户价值，因为样本没有这些字段。
Step 3: rank SKUs by rate, volume, and economic exposure步骤三：按退货率、数量和经济敞口排序 SKU
A product return analysis example becomes useful when it avoids a single leaderboard. Return rate identifies concentration; returned units identify workload; refund value plus operating cost identifies modeled cash and operational exposure. The sample produces three different leaders.
商品退货分析只有避免单一排行榜才真正有用。退货率定位集中度，退货件数反映工作量，退款金额加运营成本反映模拟的现金与运营敞口。这个样本在三种排序下出现了不同的第一名。
| SKU | Return rate退货率 | Returned units退货件数 | Refund + operating cost退款 + 运营成本 | Interpretation解读 |
|---|---|---|---|---|
| SKU-A | 18.00% | 180 | $14,400 | Highest rate and workload; fit codes dominate退货率与工作量最高；尺码原因占主导 |
| SKU-E | 15.00% | 135 | $17,550 | Highest modeled loss; damage and description need validation模拟损失最高；需验证破损与描述问题 |
| SKU-C | 15.00% | 120 | $8,400 | Fit concentration with moderate value exposure尺码问题集中，金额敞口中等 |
| SKU-G | 14.00% | 98 | $10,388 | Fit-led and higher-value than SKU-C尺码主导，单件价值高于 SKU-C |
| Other four | 4.00–7.00% | 264 combined | $12,767 | Keep monitored; do not ignore aggregate exposure继续监测，不忽略合计敞口 |
Observed SKU-E has the largest refund-plus-operating-cost total. Inferred It may be the best first investigation because a validated fix could address more modeled loss. To verify Add gross margin, resale proceeds, exchanges, batch, warehouse, carrier, and inspection condition before setting a financial priority.
已观察 SKU-E 的退款加运营成本总额最高。推断 它可能是最值得优先调查的对象，因为经过验证的修复可能覆盖更多模拟损失。待验证 在确定财务优先级前，还需补充毛利、转售收入、换货、批次、仓库、承运商与质检状态。
Step 4: treat return reasons as clues, not verdicts步骤四：把退货原因当线索，而不是定论
| Recorded reason记录原因 | Units件数 | Share占比 | Next evidence下一步证据 |
|---|---|---|---|
| Fit尺码/适配 | 305 | 38.27% | Variant-level size, chart version, measurements, review and support text变体尺码、尺码表版本、测量、评论与客服文本 |
| Not as described与描述不符 | 175 | 21.96% | PDP content version, imagery, package contents, customer comments商品页内容版本、图片、包装清单和客户评论 |
| Damaged破损 | 171 | 21.46% | Inspection photos, batch, warehouse, carrier, lane and packaging质检照片、批次、仓库、承运商、线路与包装 |
| Changed mind改变主意 | 146 | 18.32% | Policy exposure, customer cohort, acquisition source and order pattern政策曝光、客户批次、获客来源和订单模式 |
The aggregate result says fit is the largest coded group. The SKU result is more actionable: 120 of SKU-A's 180 returns and 55 of SKU-G's 98 returns are fit-coded, while SKU-E is led by 60 damage codes. A single sitewide “improve descriptions” project would therefore be less precise than separate SKU hypotheses.
总体结果显示尺码/适配是最大的原因组，但 SKU 结果更可执行：SKU-A 的 180 件退货中有 120 件标记为尺码，SKU-G 的 98 件中有 55 件标记为尺码，而 SKU-E 以 60 件破损代码为主。因此，相比分配一个全站“优化描述”项目，更准确的做法是为不同 SKU 建立独立假设。
For deeper taxonomy design, use the dedicated return reason analysis guide. For the fuller SKU workflow, see product return analysis.
Step 5: connect return cost to recovery outcome步骤五：把退货成本与回收结果连接起来
Refund value is not the same as return operating cost. In the sample, $49,505 was refunded and another $17,173 was modeled across return shipping, handling, inspection, and inventory value loss. Those figures should remain separate so finance can reconcile money returned to customers, operations can own process cost, and inventory can own disposition.
退款金额并不等于退货运营成本。样本中退款为 49,505 美元，另有 17,173 美元模拟用于退货运费、处理、质检与库存价值损失。两类金额应分开记录，便于财务对账客户退款、运营负责流程成本、库存团队负责处置。
Only 495 of 797 returned units were recovered as sellable inventory. The 62.11% recovery rate is a result, not a diagnosis. To explain the remaining 302 units, add condition grade, disposition route, days to inspection, refurbishment cost, resale proceeds, donation, liquidation, recycling, and disposal. That extension belongs in a full cost of returns model.
797 件退货中只有 495 件恢复为可售库存。62.11% 的可售回收率是结果，不是诊断。要解释其余 302 件，需要补充状态等级、处置路径、质检时长、翻新成本、转售收入、捐赠、清算、回收和报废字段。更完整的扩展可参考退货成本模型。
The worked returns analysis report: findings, limits, and decisions退货分析报告示例：发现、限制与决策
| Status状态 | Statement陈述 | Owner / next check负责人 / 下一步 |
|---|---|---|
| Observed | 797 of 8,500 eligible units were returned; unit return rate is 9.38%.8,500 件符合口径商品中退货 797 件，件数退货率为 9.38%。 | Analytics: reconcile to source exports数据团队：与源导出对账 |
| Observed | SKU-E has $17,550 in refund value plus modeled operating cost, the largest in the sample.SKU-E 的退款金额加模拟运营成本为 17,550 美元，是样本最高值。 | Finance: add margin and resale proceeds财务：补充毛利和转售收入 |
| Inferred | SKU-E may deserve the first investigation even though SKU-A has a higher return rate.虽然 SKU-A 的退货率更高，SKU-E 可能更值得优先调查。 | Operations: confirm materiality and capacity运营：确认重要性与处理能力 |
| To verify | SKU-E damage may originate in packaging, handling, a batch, a warehouse, a carrier, or customer classification.SKU-E 的破损可能来自包装、搬运、批次、仓库、承运商或客户分类。 | Quality: inspect photos and segment evidence质量团队：检查照片与分层证据 |
| Action | Run a focused evidence review before changing product, packaging, policy, or carrier.在修改商品、包装、政策或承运商前，先进行聚焦的证据复核。 | Named cross-functional owner指定跨职能负责人 |
This hierarchy prevents a common reporting failure: presenting a plausible story as a fact. It also makes the page easier for AI answer engines to quote accurately because the numerator, denominator, time window, currency, status, and limitation appear next to the claim.
这种层级可以避免常见报告错误：把看起来合理的故事当成事实。同时，由于分子、分母、时间窗口、币种、状态和限制都紧邻结论，也更利于 AI 答案引擎准确引用。
A 30-day action plan built from the example基于示例形成 30 天行动计划
- Days 1–5: reconcile the evidence.第 1–5 天：完成证据对账。 Confirm SKU-E return events, refunds, cost fields, reason mapping, physical-return status, and recovery disposition.核对 SKU-E 的退货事件、退款、成本字段、原因映射、实体退回状态和回收处置。
- Days 6–10: segment the suspected issue.第 6–10 天：分层调查疑似问题。 Break damage and description codes down by variant, batch, warehouse, carrier, destination, product-page version, and inspection condition.按变体、批次、仓库、承运商、目的地、商品页版本和质检状态拆分破损及描述原因。
- Days 11–15: choose one reversible intervention.第 11–15 天：选择一项可逆干预。 Examples include a packaging test, content clarification, measurement update, or inspection rule—only after the relevant evidence supports it.例如包装测试、内容澄清、测量更新或质检规则，但只有相关证据支持时才执行。
- Days 16–30: define the comparison.第 16–30 天：定义对照。 Pre-register the eligible cohort, success metric, guardrail, return-window delay, owner, and stop condition. Do not call the test early while returns are still maturing.预先登记符合口径的批次、成功指标、护栏指标、退货窗口延迟、负责人和停止条件。退货尚未成熟时不要提前宣布结果。
For the implementation side, connect this example to how to reduce ecommerce returns. For the rate formula and cohort design, use how to calculate return rate.
How to reproduce this returns analysis in Excel, SQL, or BI如何在 Excel、SQL 或 BI 中复制该退货分析
The logic is tool-independent. In Excel, import the CSV, add calculated columns or measures, and build pivots by SKU and reason. In SQL, aggregate a reconciled order-item table after joining return events and disposition. In a BI tool, keep the semantic definitions visible so “returned quantity,” “reversed quantity,” and “refund” cannot be swapped without review.
这套逻辑与工具无关。在 Excel 中导入 CSV，添加计算列或度量值，并按 SKU 与原因建立透视表；在 SQL 中关联退货事件和处置记录后，对已对账的订单行表进行聚合；在 BI 工具中保持语义定义可见，避免未经审核就把“实体退货件数”“冲销件数”和“退款”互相替换。
Scope and maturity → data-quality status → five baselines → top SKU/reason/cost findings → limitations → three actions with owners.
口径与成熟度 → 数据质量状态 → 五项基线 → 关键 SKU/原因/成本发现 → 限制 → 三项带负责人的行动。
Refresh weekly for operations, but compare only cohorts whose return windows have closed. Reconcile financial totals monthly.
运营层面每周刷新，但只比较退货窗口已关闭的批次；财务合计每月对账。
If you need a reusable blank structure rather than a completed example, download the returns report template. If you need the broader measurement system, start with ecommerce returns metrics or the main ecommerce returns analytics hub.
如果需要空白的可复用结构，而不是已完成的示例，请使用退货报告模板。如果需要更完整的衡量体系，请从电商退货指标或主电商退货分析中心页开始。
Use Return Compass to organize the same investigation用逆向罗盘组织同样的调查
Prepare sanitized order-item, return-event, refund, reason, cost, and disposition files. State one business question and attach the metric contract, exclusions, and return-window rule. Confirm the tool's current file, login, pricing, retention, and security requirements before use.
准备脱敏后的订单行、退货事件、退款、原因、成本和处置文件。提出一个业务问题，并附指标口径、排除项和退货窗口规则。使用前请核验工具当前的文件、登录、收费、留存与安全要求。
Move from a blended rate to a reviewable hypothesis从总体退货率走向可复核假设
Return Compass can help organize a file-based diagnosis when the required fields are present. Human reviewers remain responsible for product, policy, refund, fraud, supplier, carrier, inventory, and financial decisions. This page does not claim native integrations, automatic execution, guaranteed savings, or guaranteed return-rate reduction.
字段齐备时，逆向罗盘可辅助组织基于文件的诊断。商品、政策、退款、欺诈、供应商、承运商、库存和财务决策仍由人工负责。本页不声称原生集成、自动执行、保证节省或保证降低退货率。
Open Return Compass打开逆向罗盘Method sources, observation date, and limitations方法来源、观察日期与限制
Method references were checked on September 15, 2026. The numerical example is first-party synthetic content: it was constructed for reproducibility and is not presented as market research, a benchmark, a customer case study, or an InfiniSynapse product result.
方法资料核对于 2026 年 9 月 15 日。数字示例属于第一方模拟内容，为可复现性而构建，不作为市场研究、行业基准、客户案例或 InfiniSynapse 产品效果。
- Shopify Help — Sales reportsShopify 帮助——销售报告 — current distinctions among returned quantity, reversed quantity, reasons, and sales reversals.——当前对实体退货件数、冲销件数、原因与销售冲销的区分。
- Google Analytics Help — Ecommerce events and refundsGoogle Analytics 帮助——电商事件与退款 — implementation context for full and partial refund events.——全额及部分退款事件的实施背景。
- NIST/SEMATECH e-Handbook of Statistical MethodsNIST/SEMATECH 统计方法电子手册 — data-quality checks for entry problems, unusual values, and missingness.——录入问题、异常值与缺失情况的数据质量检查。
- NIST SP 800-122NIST SP 800-122 — protecting personally identifiable information when preparing real files.——准备真实文件时保护个人身份信息。
Commercial disclosure: InfiniSynapse publishes this educational page and provides Return Compass, so this is not an independent product review. The download is synthetic and no outcome claim is based on customer data.
商业披露：本教育页面由 InfiniSynapse 发布，逆向罗盘也由 InfiniSynapse 提供，因此本页不是独立产品评测。下载文件为模拟数据，没有任何效果声明基于客户数据。
Frequently asked questions常见问题
Define the cohort and denominator, check data quality, calculate unit and value rates, rank SKUs by rate and economic exposure, break down recorded reasons, measure cost and recovery, and separate observations from hypotheses and actions.
应定义批次和分母、检查数据质量、计算件数率和金额率、按退货率与经济敞口排序 SKU、拆分记录原因、衡量成本与回收，并区分观察、假设与行动。
Divide returned units by eligible shipped units. The sample calculation is 797 ÷ 8,500 × 100 = 9.38%.
用退货件数除以符合口径的已发货件数：797 ÷ 8,500 × 100 = 9.38%。
Rate shows concentration, but priority also depends on volume, refund value, handling cost, margin, recovery, and confidence in the suspected cause. SKU-E has a lower rate than SKU-A but a larger modeled loss.
退货率反映集中度，但优先级还取决于数量、退款金额、处理成本、毛利、回收情况和疑似原因的可信度。SKU-E 的退货率低于 SKU-A，但模拟损失更高。
No. A customer-selected reason is a useful observed label. Root cause needs supporting evidence such as measurements, inspections, support text, product-content versions, supplier batches, warehouse, or carrier data.
不等于。客户选择的原因是有用的已观察标签；根因还需要测量、质检、客服文本、商品内容版本、供应商批次、仓库或承运商数据支持。
Yes. It is a synthetic educational dataset. Preserve the definitions, and replace rows only after removing personal data and reconciling your return, refund, cost, and disposition records.
可以。这是模拟教育数据集。请保留字段定义；只有在移除个人数据，并完成退货、退款、成本和处置记录对账后，才用真实行替换样本。
Turn this example into one auditable investigation把这个示例转化为一项可审计的调查
Start with one mature cohort and one question. Keep the numerator, denominator, currency, window, exclusions, missing fields, evidence status, and owner beside every important conclusion. A useful returns analysis report does not merely show what came back—it makes the next decision smaller, testable, and reviewable.
从一个成熟批次和一个问题开始。把分子、分母、币种、窗口、排除项、缺失字段、证据状态和负责人放在每个重要结论旁边。真正有用的退货分析报告不只是展示“退回了什么”，而是让下一项决策更小、更可测试、更可复核。