On this page本页目录
What is a cohort return rate?什么是群组退货率？
A cohort return rate is the share of eligible orders, units, or merchandise value from one original sales group that later becomes a physical return. The cohort is usually defined by fulfillment or delivery week, and every cohort must be measured at the same age—such as 30 days after delivery—before the rates are compared.
群组退货率是指某个原始销售群组中的合格订单、件数或商品金额，后来形成实体退货的比例。群组通常按履约周或签收周划分；比较前，每个群组必须达到相同年龄，例如签收后 30 天。
This article is for ecommerce operators, finance teams, and analysts who compare return performance across weeks, campaigns, collections, or channels. It covers physical-return attribution and maturity. It does not measure customer repeat purchase or website retention, even though analytics tools also use the word “cohort” for those tasks.
本文面向需要按周、活动、系列或渠道比较退货表现的电商运营、财务和分析团队，重点讨论实体退货的归因与成熟期。尽管分析工具也用“cohort”描述客户复购或网站留存，本文不计算这两类指标。
Define the sales cohort before counting returns统计退货前先定义销售群组
A cohort is a group that shares a defined starting event during a defined period. For return analysis, the most defensible starting event is normally fulfillment or delivery because the return opportunity begins only after the customer receives the item. Order date can still work when delivery timestamps are unavailable, but shipping-time changes may then distort comparisons.
群组是在特定时间段经历同一开始事件的一组记录。进行退货分析时，履约或签收通常是更稳妥的开始事件，因为客户收到商品后才真正拥有退货机会。如果缺少签收时间，也可以使用下单日期，但配送时长变化可能扭曲比较。
Orders fulfilled or delivered in the same week or month. Use it for order-level return incidence.在同一周或同一月履约或签收的订单，适合计算订单级退货发生率。
Units from the same fulfilled-sales period. Use it when basket sizes vary or product analysis is the goal.来自同一履约销售期的商品件数，适合购物篮大小变化或商品分析。
Eligible merchandise value from the same sales period. Use it to evaluate revenue and margin exposure.来自同一销售期的合格商品金额，适合评估收入与利润风险。
Return requested, carrier-scanned, received, inspected, or refunded. Pick one stage and keep it stable.申请退货、承运商扫描、仓库收货、质检或退款；只能选择一个阶段并保持一致。
Google Analytics defines a cohort as users sharing a common characteristic and allows daily, weekly, or monthly granularity. That general structure is useful, but GA4’s “return criterion” means a later user event—not a merchandise return. Do not copy a retention-cohort metric into a physical-return report.
Google Analytics 把 cohort 定义为具有共同特征的一组用户，并支持按日、周或月划分。这种结构可以借鉴，但 GA4 的“return criterion”表示用户后续再次发生事件，并不是商品退回。不要把留存群组指标直接复制到实体退货报告。
Use one denominator for every cohort return rate所有群组退货率必须使用同一种分母
Replace “units” with orders or merchandise value only when both numerator and denominator change together. Shopify’s official sales-report definition likewise calculates returned quantity rate as physically returned units divided by ordered units. The cohort method adds original-sale attribution and a fixed observation age to that unit-level relationship.
只有在分子和分母同时改变时，才能把“件数”替换为订单或商品金额。Shopify 官方销售报告同样把 returned quantity rate 定义为实体退回件数除以订购件数；群组方法则在这一件数关系上增加原始销售归因与固定观察年龄。
| Metric指标 | Numerator分子 | Denominator分母 | Use用途 |
|---|---|---|---|
| Order cohort rate订单群组率 | Orders with ≥1 qualifying return至少有 1 件合格退货的订单 | Eligible fulfilled orders合格履约订单 | Customer/order incidence客户或订单发生率 |
| Unit cohort rate件数群组率 | Qualifying returned units合格退回件数 | Eligible fulfilled units合格履约件数 | Product and category comparison商品与品类比较 |
| Value cohort rate金额群组率 | Returned merchandise value退回商品金额 | Eligible merchandise sales value合格商品销售额 | Revenue exposure收入风险 |
Exclude cancellations that occurred before fulfillment from a physical-return denominator unless your documented business definition intentionally combines them. Exchanges, store credit, partial returns, replacements, and rejected returns also need explicit rules. The general return-rate calculation guide covers those numerator and denominator choices.
除非业务定义明确要把取消与实体退货合并，否则履约前取消不应进入实体退货分母。换货、商店积分、部分退货、补发和拒收也需要明确规则。完整的退货率计算指南说明这些分子与分母选择。
Prepare six fields that preserve original-sale attribution准备六个字段以保留原始销售归因
The analysis needs a stable join between the sales line and the return line. A return-date-only report cannot produce a true cohort return rate because it shows when the reverse event happened, not which sales group created it. Keep raw timestamps in addition to the derived cohort label so definitions can be changed without rebuilding the source export.
分析需要在销售行与退货行之间建立稳定关联。只有退货日期的报告无法计算真正的群组退货率，因为它只说明逆向事件何时发生，却没有说明来自哪个销售群组。除派生的群组标签外，还应保留原始时间戳，以便调整定义时无需重建源数据。
- Stable order and line IDs稳定的订单与订单行 ID
Use line-level keys when an order can contain multiple products or partial returns.当一张订单包含多个商品或部分退货时，使用订单行级键。 - Order, fulfillment, and delivery dates下单、履约与签收日期
Choose one cohort start event and retain the others for transit-time checks.选择一个群组开始事件，并保留其他日期用于配送时长检查。 - Return event and timestamp退货事件与时间
Record request, carrier scan, receipt, inspection, and refund separately when available.如有条件，分别记录申请、承运商扫描、收货、质检与退款。 - Eligible quantity and value合格件数与金额
Keep units and merchandise value so order, unit, and value rates are not mixed.同时保留件数与商品金额，避免混用订单率、件数率和金额率。 - Return-window rule退货窗口规则
Store the allowed window by market, product, or policy version instead of assuming one rule.按市场、商品或政策版本保存允许窗口，不要假设所有订单相同。 - Exclusion flags排除标记
Identify cancellations, tests, fraud, duplicate events, replacements, and rejected returns.标记取消、测试、欺诈、重复事件、补发与拒收。
The CSV provides line-level fields, cohort-age columns, a maturity flag, and clearly labeled synthetic examples.
CSV 提供订单行字段、群组年龄、成熟标记以及明确标注的模拟示例。
Download CSV template下载 CSV 模板 ↓Give every cohort the same return opportunity让每个群组拥有相同的退货机会
A recent cohort often looks healthier because many customers are still inside the delivery, decision, shipping, and warehouse-processing cycle. Choose an age that reflects the policy window plus operational lag. For a 30-day return policy, a 30-day-after-delivery view may still exclude parcels requested on day 30 but received later. A 45- or 60-day maturity point can be more complete, but it delays reporting.
近期群组往往看起来更健康，因为许多客户仍处于配送、决策、寄回和仓库处理周期内。成熟年龄应覆盖政策窗口与运营延迟。对于签收后 30 天可申请退货的政策，观察到第 30 天仍可能漏掉第 30 天申请、之后才到仓的包裹；45 天或 60 天更完整，但报告会更晚。
| View观察方式 | Question answered回答的问题 | Risk风险 |
|---|---|---|
| Age 7第 7 天 | Which cohorts produce very early return signals?哪些群组出现早期退货信号？ | Large share of outcomes still pending大量结果尚未发生 |
| Age 30第 30 天 | How do cohorts compare after one common month?相同一个月后群组如何比较？ | Late shipments and receipts may remain open延迟寄回与收货可能尚未完成 |
| Age 60第 60 天 | What is the more mature physical-return outcome?更成熟的实体退货结果是什么？ | Slower feedback for operational decisions运营反馈更慢 |
| Final closed最终关闭 | What happened after the defined maximum window?定义的最长窗口后最终发生了什么？ | Different policies can create different close dates不同政策可能产生不同关闭日期 |
Do not compare a seven-day-old cohort with a sixty-day-old cohort and call the difference an improvement. Either compare all groups at age 7, age 30, or age 60, or model the incomplete recent cohorts separately and label the result as estimated.
不要把 7 天群组与 60 天群组直接比较并称作改善。应统一比较所有群组的第 7、30 或 60 天状态；如果单独建模未成熟的近期群组，必须把结果标为估计值。
Calculate cohort return rate in seven reproducible steps用七个可复现步骤计算群组退货率
- Declare the business question明确业务问题
Choose whether the decision concerns orders, units, merchandise value, a category, or a channel.确定决策对象是订单、件数、商品金额、品类还是渠道。 - Choose the cohort start选择群组起点
Use fulfillment or delivery week when possible and document the timezone.尽可能使用履约周或签收周，并记录时区。 - Join returns to original lines把退货关联到原始订单行
Connect each qualifying reverse event to the source order line; do not group only by return date.把每个合格逆向事件连接到原始订单行，不能只按退货日期分组。 - Apply eligibility rules应用合格规则
Remove documented exclusions from both numerator and denominator where appropriate.根据既定规则从分子和分母中移除排除项。 - Calculate cohort age计算群组年龄
Measure days between the chosen start and the observation date or return event.计算所选起点与观察日期或退货事件之间的天数。 - Freeze equal-age snapshots冻结相同年龄快照
Store rates at age 7, 30, 45, or 60 rather than repeatedly overwriting one value.分别保存第 7、30、45 或 60 天的比率，不要反复覆盖一个数字。 - Reconcile and explain对账并解释
Tie totals back to source exports, then inspect category, product, reason, channel, and policy mix.把总数与源导出对账，再检查品类、商品、原因、渠道与政策结构。
Worked example: equal-age comparison changes the conclusion示例：相同年龄比较会改变结论
Assume three synthetic weekly cohorts each contain 1,000 eligible fulfilled units. At the current reporting date, the oldest cohort has accumulated 140 returns, the middle cohort 110, and the newest cohort 55. A naive current-total comparison reports 14.0%, 11.0%, and 5.5%, suggesting rapid improvement.
假设三个模拟周群组都包含 1,000 件合格履约商品。截至当前报告日，最早群组累计 140 件退货，中间群组 110 件，最新群组 55 件。直接比较当前累计值会得到 14.0%、11.0% 和 5.5%，看起来像是快速改善。
| Synthetic cohort模拟群组 | Current age当前年龄 | Returns at current age当前累计退货 | Rate now当前比率 | Returns by age 14第 14 天退货 | Age-14 rate第 14 天比率 |
|---|---|---|---|---|---|
| Week A | 56 days | 140 | 14.0% | 70 | 7.0% |
| Week B | 35 days | 110 | 11.0% | 78 | 7.8% |
| Week C | 14 days | 55 | 5.5% | 55 | 5.5% |
At the common age of 14 days, the rates are 7.0%, 7.8%, and 5.5%. The newest cohort still looks lower, but the apparent gap is 1.5–2.3 percentage points rather than 5.5–8.5 points. Even that result is an early signal, not a final outcome. Wait for the chosen maturity point and check whether product, promotion, or channel mix changed.
在统一的第 14 天观察时，比率分别为 7.0%、7.8% 和 5.5%。最新群组依然较低，但差距变成 1.5–2.3 个百分点，而不是 5.5–8.5 个百分点。即便如此，它也只是早期信号，不是最终结果；仍需等待既定成熟点，并核对商品、促销或渠道结构是否变化。
All store figures in this example are synthetic and are included only to demonstrate the calculation. They are not InfiniSynapse customer results or an ecommerce benchmark.本示例中的商店数据全部为模拟值，仅用于说明计算方法，不代表 InfiniSynapse 客户结果或电商行业基准。
Separate a measured difference from a possible explanation区分已测得差异与可能解释
A cohort table can establish that one sales group produced a different return outcome under the selected rules. It cannot by itself prove why. Treat the rate difference as an observed pattern, then test candidate explanations using fields that existed before the outcome: product, size, color, supplier, acquisition channel, promotion, shipping promise, destination, and policy version.
群组表可以证明某个销售群组在既定规则下产生了不同的退货结果，但不能单独证明原因。应把比率差异视为已观察模式，再利用结果发生前就存在的字段检验可能解释，例如商品、尺码、颜色、供应商、获客渠道、促销、配送承诺、目的地和政策版本。
The age-30 unit return rate increased from 9.2% to 11.1%.第 30 天件数退货率从 9.2% 升至 11.1%。
The newer cohort contains more high-return-size variants or promotion traffic.新群组可能包含更多高退货尺码或促销流量。
Matched product-size-channel tables, return reasons, delivery time, and policy version.同口径的商品—尺码—渠道表、退货原因、配送时长与政策版本。
Act only when the difference is material, repeated, sufficiently mature, and operationally addressable.只有差异足够重要、重复出现、已经成熟且可操作时才采取行动。
Pair cohort rates with returned units, merchandise value, contribution margin, support contacts, and repeat purchase. Use return reason analysis to narrow the investigation. External averages such as the dated U.S. benchmark belong in a separate context layer; the average ecommerce return-rate guide explains why unmatched benchmarks are not targets.
群组退货率应与退回件数、商品金额、贡献利润、客服咨询和复购一起查看，再使用退货原因分析缩小调查范围。带年份的美国平均值应放在独立背景层；电商平均退货率指南解释了为什么口径不匹配的基准不能作为目标。
Handle policy changes, exchanges, and incomplete joins处理政策变化、换货与关联缺失
- Policy changes: retain policy version and maximum return window on each order so a 30-day cohort is not compared with a 60-day cohort without adjustment.政策变化：在每张订单保存政策版本和最长退货窗口，避免直接比较 30 天与 60 天政策群组。
- Holiday extensions: use a longer maturity point or analyze extended-window orders separately.节日延长：使用更长成熟期，或单独分析延长窗口订单。
- Exchanges: decide whether the original returned unit counts once and whether the replacement begins a new eligibility record.换货：明确原商品是否只计一次退货，以及替换商品是否建立新的合格记录。
- Partial returns: prefer line-level unit logic; order-level rates lose the size of the return.部分退货：优先使用订单行件数逻辑；订单级比率会丢失退货规模。
- Missing order-line joins: report the unmatched-return count and rate. Do not silently discard unmatched records.订单行关联缺失：报告无法匹配的退货数量与比例，不要静默删除。
- Cross-border delays: segment or control for transit time when delivery and return shipping vary materially by market.跨境延迟：如果不同市场的配送与寄回时间差异明显，应分组或控制运输时长。
Run eight checks before publishing a cohort comparison发布群组比较前完成八项检查
| Check检查项 | Pass condition通过条件 |
|---|---|
| Join coverage关联覆盖率 | Matched and unmatched return records are reported报告已匹配与未匹配的退货记录 |
| Unit consistency单位一致性 | Numerator and denominator are both orders, units, or value分子分母同时为订单、件数或金额 |
| Event consistency事件一致性 | The same return stage is used across cohorts所有群组使用同一退货事件阶段 |
| Age consistency年龄一致性 | Every compared cohort uses the same observation age所有比较群组使用相同观察年龄 |
| Eligibility consistency合格规则一致性 | Cancellations, tests, and exclusions follow one versioned rule取消、测试和排除项遵循同一版本规则 |
| Reconciliation对账 | Cohort totals tie to the source sales and returns exports群组总数与销售、退货源导出一致 |
| Mix check结构检查 | Product, channel, market, and promotion mix are shown展示商品、渠道、市场与促销结构 |
| Evidence label证据标签 | Measured, calculated, estimated, and synthetic values are distinguished区分实测、计算、估计与模拟值 |
Avoid seven cohort return-rate mistakes避免七个群组退货率错误
- Grouping returns by return date and calling the result a sales cohort.按退货日期分组，却把结果称作销售群组。
- Comparing cohorts at different ages without an incomplete-data warning.比较不同年龄群组，却不提示数据尚未成熟。
- Using order counts in the numerator and unit counts in the denominator.分子使用订单数，分母却使用件数。
- Mixing return requests, warehouse receipts, and refunds in one trend line.在同一趋势线混合退货申请、仓库收货和退款。
- Overwriting early snapshots instead of preserving age-7, age-30, and final values.覆盖早期快照，而不是保存第 7 天、第 30 天和最终值。
- Treating a cohort difference as proof of a product, campaign, or supplier cause.把群组差异直接当作商品、活动或供应商原因的证明。
- Comparing an internal unit cohort with an external sales-value benchmark.把内部件数群组与外部销售金额基准直接比较。
Build a comparable return cohort before diagnosing causes先建立可比退货群组，再诊断原因
Prepare order-line IDs, fulfillment or delivery dates, return-event timestamps, quantities, values, policy windows, and exclusion flags. Return Compass can help organize file-based loss analysis after the cohort rules are documented; it does not supply missing events or decide the policy definition for you.
准备订单行 ID、履约或签收日期、退货事件时间、件数、金额、政策窗口和排除标记。在记录群组规则后，逆向罗盘可辅助整理基于文件的损失分析；它不会补造缺失事件，也不会替你决定政策口径。
Open Return Compass打开逆向罗盘 →Cohort Return Rate FAQ群组退货率常见问题
It is the share of eligible orders, units, or merchandise value from one original sales cohort that later becomes a qualifying physical return, measured at a declared cohort age.它是某个原始销售群组中的合格订单、件数或商品金额，后来形成合格实体退货的比例，并在声明的群组年龄进行测量。
For a cohort return rate, assign the return to the original order, fulfillment, or delivery cohort. Keep return date as the event timestamp, but do not use it as the sales-cohort identity.计算群组退货率时，应把退货归回原始订单、履约或签收群组；退货日期保留为事件时间，但不能替代销售群组身份。
Use a documented age that covers the policy window and expected operational lag. A 30-day policy may need a 45- or 60-day physical-receipt snapshot, depending on shipping and warehouse timing.应选择覆盖政策窗口与预期运营延迟的固定年龄。30 天政策可能需要在第 45 或 60 天观察实体收货，具体取决于运输与仓库时效。
No. Cohort return rate tracks merchandise returns tied to original sales. Customer retention rate tracks whether users or customers return to perform a later activity such as another purchase.不等于。群组退货率追踪归属于原始销售的商品退回；客户留存率追踪用户或客户是否再次完成购买等后续行为。
They can be modeled, but the result must be labeled estimated, the method and uncertainty must be documented, and the estimate should not replace equal-age measured comparisons.可以建模估计，但必须标明为估计值、记录方法与不确定性，而且不能替代相同年龄的实测比较。
Sources, evidence labels, and limitations来源、证据标签与限制
- Google Analytics Help: Cohort exploration — official source for the general cohort definition, inclusion and return criteria, daily/weekly/monthly granularity, and standard/rolling/cumulative calculations; it is not a merchandise-return definition.Google Analytics 帮助：群组探索——一般群组定义、纳入与返回条件、日/周/月粒度以及标准/滚动/累计计算的官方来源；它并不是商品退货定义。
- Shopify Help Center: Sales reports — official source for returned quantity, returned quantity rate, sales-reversal timing, and the distinction between physical returns and broader reversals.Shopify 帮助中心：销售报告——实体退回件数、退回件数率、销售冲回时间，以及实体退货与广义冲回区别的官方来源。
- National Retail Federation and Happy Returns: 2025 Retail Returns Landscape — primary market-context source illustrating why external annual estimates require explicit year, sample, channel, and methodology before comparison.美国零售联合会与 Happy Returns：2025 零售退货报告——用于说明外部年度估计在比较前必须保留年份、样本、渠道与方法的一手市场背景来源。
Evidence statement: The worked cohort data is synthetic. External documentation is used for general cohort mechanics and physical-return metric definitions. No customer result, universal maturity window, causal claim, or guaranteed improvement is asserted. Source review was frozen on September 15, 2026, and a named reviewer is required before publication.证据声明：计算示例为模拟数据；外部文档仅用于说明一般群组机制和实体退货指标定义。本文不声称客户结果、通用成熟窗口、因果结论或保证改善。来源核验冻结于 2026 年 9 月 15 日，发布前仍需具名审核人。
Compare cohorts at the same age, then investigate the difference先按相同年龄比较群组，再调查差异
A defensible cohort return rate connects each qualifying physical return to its original sales group, keeps numerator and denominator at the same unit, and freezes every comparison at the same cohort age. Preserve early and final snapshots, disclose unmatched records, and separate observed differences from possible causes. When cohorts are not equally mature, report the limitation or use an explicitly labeled estimate instead of presenting recent sales as automatically healthier.
可靠的群组退货率会把每笔合格实体退货连接到原始销售群组，让分子和分母保持相同单位，并在相同群组年龄冻结比较。应保存早期与最终快照、披露无法关联的记录，并区分已观察差异与可能原因。若群组成熟度不一致，应报告限制，或使用明确标注的估计值，而不是把近期销售描述成天然更健康。