Customer retention rate: quick answer and formula客户留存率：快速答案与公式
Customer retention rate is the percentage of customers present at the start of a period who are still customers at the end. The common formula removes customers acquired during the period from the ending count so growth does not disguise losses from the starting group.
客户留存率是期初客户中，在周期结束时仍然是客户的比例。常用公式从期末客户数中扣除周期内新增客户，避免新增增长掩盖期初群体的流失。
S is eligible customers at the start, E is customers at the end, and N is customers first acquired during the period. The formula is reliable only when all counts use the same customer identity, eligibility definition, timezone, and boundary rules.
S 是期初符合条件的客户，E 是期末客户，N 是周期内首次获得的客户。只有三个计数使用相同的客户身份、资格定义、时区和边界规则时，公式才可靠。
What customer retention rate does—and does not—tell you客户留存率能说明什么，又不能说明什么
Teams search for customer retention rate because acquisition alone cannot show whether customers continue to receive value. A consistent rate helps monitor continuity, compare like-for-like cohorts, and detect where deeper investigation is needed. It is an outcome metric, not a causal explanation.
团队关注客户留存率，是因为获客数据无法单独说明客户是否持续获得价值。口径一致的留存率可用于监测客户关系连续性、比较可比同期群，并发现需要深入调查的位置。它是结果指标，不是因果解释。
Stable customer populations, subscription renewals, recurring relationships, membership continuity, and repeatable cohort comparisons.
稳定客户群、订阅续约、持续性客户关系、会员连续性，以及可重复的同期群比较。
Explaining why customers left, measuring expansion revenue, proving an intervention worked, or judging products with long and irregular repurchase cycles.
解释客户为何离开、衡量扩张收入、证明某项干预有效，或评价复购周期很长且不规律的产品。
For non-contractual commerce, “still a customer” can be ambiguous. A customer who did not buy this month may still be active in a six-month purchase cycle. In that case, define an inactivity threshold, repeat purchase rate, or probabilistic active state before calculating retention.
在非合同型商业中，“仍是客户”可能含义模糊。本月没有购买的客户，在六个月购买周期中仍可能活跃。此时应先定义不活跃阈值、复购率或概率活跃状态，再计算留存。
Prepare the data before calculating customer retention rate计算客户留存率前先准备数据
A trustworthy calculation begins with a short metric contract. Write the rules before viewing the result so a favorable number does not influence the definition.
可信的计算始于一份简短的指标契约。应在看到结果前写下规则，避免一个看似有利的数字反过来影响定义。
- Customer identity: choose person, household, account, workspace, or paying organization, and document how duplicates and merges are handled.客户身份：选择个人、家庭、账户、工作区或付费组织，并记录重复项和合并账户的处理方式。
- Retained state: define the evidence that makes a customer active at period end, such as an active contract, valid subscription, or purchase within an approved interval.留存状态：定义客户在期末仍活跃的证据，例如有效合同、有效订阅，或在规定间隔内发生购买。
- Period: set start and end timestamps, timezone, inclusive or exclusive boundaries, and a cadence matching the natural customer cycle.周期：设定起止时间、时区、边界是否包含，并让统计频率匹配自然客户周期。
- Eligibility: specify trials, free plans, test accounts, employees, fraud, paused accounts, refunds, and reactivations.资格：明确试用、免费计划、测试账户、员工账户、欺诈、暂停账户、退款与重新激活的规则。
- Source fields: preserve customer ID, acquisition timestamp, state-change timestamps, plan or segment, cancellation or expiration, and a snapshot timestamp.来源字段：保留客户 ID、获得时间、状态变化时间、计划或细分、取消或到期时间，以及快照时间。
Snapshot warning: querying only current account status can rewrite history. Use dated snapshots or an append-only event history so late cancellations, backfills, and reactivations can be traced.
快照提醒：只查询当前账户状态可能重写历史。应使用带日期的快照或只追加的事件历史，确保延迟取消、回填与重新激活可追溯。
How to calculate customer retention rate step by step如何逐步计算客户留存率
- Freeze the metric contract冻结指标契约Record population, period, timezone, active-state rule, exclusions, and treatment of reactivations before running a query.在运行查询前记录统计人群、周期、时区、活跃状态规则、排除项与重新激活处理方法。
- Count the starting population S计算期初群体 SDeduplicate eligible customer identities active at the start boundary. Save the member list, not only the count.对期初边界上符合条件且活跃的客户身份去重。保存成员清单，而不只是总数。
- Count ending customers E and new customers N计算期末客户 E 与新增客户 NApply the same identity and eligibility rules at the end, then identify customers whose first eligible acquisition occurred inside the period.在期末应用相同的身份和资格规则，再识别首次符合条件的获得时间落在周期内的客户。
- Calculate and reconcile计算并对账Compute ((E − N) ÷ S) × 100, then compare E − N with a direct intersection of the starting and ending member lists. They should agree under the simple model.计算（E − N）÷ S × 100，再把 E − N 与期初和期末成员清单的直接交集比较。在简单模型下，两者应一致。
- Segment without changing definitions在不改变定义的前提下细分Repeat by acquisition cohort, plan, region, channel, tenure, or behavior. Keep the original denominator and period rules visible.按获得同期群、计划、地区、渠道、客户年限或行为重复分析，并始终显示原始分母和周期规则。
- Interpret, investigate, and test解读、调查并测试Treat differences as signals. Review qualitative evidence and operational changes, propose alternatives, and validate any intervention with a suitable experiment or comparison.把差异视为信号。检查定性证据和运营变化，提出替代解释，并用合适的实验或比较验证干预措施。
Customer retention rate example with a validation check带验证检查的客户留存率示例
Hypothetical example: a subscription service starts April with 1,000 eligible paying accounts. It ends April with 980 eligible accounts, including 120 first acquired during April. The retained starting accounts are 980 − 120 = 860, so monthly customer retention rate is 860 ÷ 1,000 × 100 = 86%.
假设示例：某订阅服务在 4 月初有 1,000 个符合条件的付费账户。4 月末有 980 个符合条件的账户，其中 120 个是在 4 月首次获得。保留的期初账户为 980 − 120 = 860，因此月度客户留存率为 860 ÷ 1,000 × 100 = 86%。
The team should also intersect the saved April 1 and April 30 account lists. If that direct intersection returns 855 instead of 860, the five-account difference must be resolved before reporting. Possible causes include reactivated accounts misclassified as new, timezone boundaries, merged identities, or delayed status updates.
团队还应直接求 4 月 1 日与 4 月 30 日账户清单的交集。如果交集结果是 855 而不是 860，就必须在报告前解释这 5 个账户的差异。可能原因包括重新激活账户被误判为新增、时区边界、身份合并或状态更新延迟。
Choose aggregate, cohort, or segment retention analysis选择总体、同期群或细分留存分析
| View视图 | Best for适合用途 | Main limitation主要限制 |
|---|---|---|
| Aggregate CRR总体客户留存率 | A stable headline metric and executive monitoring稳定的总体指标与管理监测 | Mix changes can hide weak new cohorts or strong mature cohorts群体构成变化可能掩盖较弱的新同期群或较强的成熟同期群 |
| Acquisition cohort获得同期群 | Comparing customers acquired in the same week, month, or quarter比较同一周、月或季度获得的客户 | Recent cohorts need enough observation time before comparison近期同期群需要足够观察时间后才能比较 |
| Segment retention细分留存 | Finding meaningful differences by plan, channel, region, or behavior发现计划、渠道、地区或行为上的有意义差异 | Small groups are noisy and correlation does not establish cause小群体波动大，相关关系也不能证明因果 |
| Revenue retention收入留存 | Understanding contraction, expansion, and retained recurring revenue理解收缩、扩张与保留的经常性收入 | It is a different metric and can exceed 100% when expansion is included这是不同指标；包含扩张收入时可能超过 100% |
Use the aggregate rate as a monitoring summary, then use cohorts and segments for diagnosis. Keep raw counts beside percentages: 50% of 20 customers is not as stable as 50% of 20,000. When comparing groups, report observation windows and uncertainty instead of ranking every small difference.
可用总体留存率进行汇总监测，再用同期群和细分完成诊断。百分比旁应保留原始数量：20 名客户的 50% 与 20,000 名客户的 50% 稳定性不同。比较群体时，应报告观察窗口与不确定性，而不是给每个微小差异强行排序。
What is a good customer retention rate?什么是好的客户留存率？
There is no universal good customer retention rate. A valid comparison needs the same customer unit, retained-state definition, product category, lifecycle stage, contract or purchase cycle, and observation period. A monthly subscription rate cannot be compared directly with annual contract renewal or a repeat-purchase rate.
不存在通用的“好”客户留存率。有效比较需要相同的客户单位、留存状态定义、产品类别、生命周期阶段、合同或购买周期，以及观察周期。月度订阅留存率不能直接与年度合同续约率或复购率比较。
Build an internal baseline first: recalculate at least several comparable periods, annotate product, pricing, incident, and policy changes, and compare equivalent cohorts at the same age. If using an external benchmark, record its source, date, population, denominator, and calculation method. If those fields are missing, treat it as orientation rather than a target.
应先建立内部基线：重新计算多个可比周期，标注产品、定价、事故与政策变化，并在相同客户年龄上比较等价同期群。如果使用外部基准，应记录来源、日期、人群、分母和计算方法。若这些字段缺失，就只能把它当作方向参考，而不是目标。
Retention and churn: under identical rules, retained and churned starting customers are mutually exclusive, so retention rate + churn rate = 100%. If churn counts cancellations during a month but retention uses active accounts at month-end, the relationship may not hold. See the separate customer churn analysis guide for root-cause and risk workflows.
留存与流失：在完全相同的规则下，期初客户只能是留存或流失之一，因此留存率 + 流失率 = 100%。如果流失统计月内取消事件，而留存统计月末活跃账户，两者就可能不满足该关系。根因与风险工作流请参阅独立的客户流失分析指南。
Analyze customer retention with prepared, governed data使用已准备且受治理的数据分析客户留存
After the metric contract and reconciliation totals are ready, analysis can move beyond a spreadsheet percentage. Compare acquisition cohorts, plans, regions, and early behaviors; inspect supporting records; and produce reviewable tables, charts, and written findings. Do not upload personal or sensitive data unless your organization authorizes the source, purpose, access, and retention rules.
当指标契约和对账总数准备好后，分析就能超越电子表格中的单一百分比。可以比较获得同期群、计划、地区和早期行为，检查支持记录，并生成可审阅的表格、图表与文字结论。除非组织已经批准来源、目的、访问和保留规则，否则不要上传个人或敏感数据。
Prepare deduplicated customer IDs, acquisition and state dates, cohort or segment fields, period boundaries, exclusions, a data dictionary, and trusted reconciliation totals. Then use the InfiniSynapse AI data analysis web app to examine the prepared sources through a natural-language analytical workflow and review the resulting artifacts. The app is not presented as a dedicated retention calculator, CRM, or campaign executor.
请准备去重后的客户 ID、获得与状态日期、同期群或细分字段、周期边界、排除项、数据字典及可信对账总数。随后可使用 InfiniSynapse AI 数据分析 Web 应用，通过自然语言分析工作流检查已准备的数据源并审阅分析产物。本页不把该应用描述为专用留存计算器、CRM 或活动执行器。
Analyze prepared retention data分析已准备的留存数据For useful grouping variables, the existing customer segmentation guide explains how to define decision-ready groups, while the behavioral segmentation workflow focuses on event-based features. These are analysis aids, not substitutes for a valid retention denominator.
如需构建有用的分组变量，现有客户细分指南说明如何定义可用于决策的群体，行为细分工作流则聚焦事件特征。它们是分析辅助方法，不能替代有效的留存分母。
Common customer retention rate errors and result checks客户留存率常见错误与结果检查
| Problem问题 | Why it misleads为何误导 | Check检查方法 |
|---|---|---|
| Using E ÷ S直接使用 E ÷ S | New acquisition hides losses from the starting group新增客户掩盖期初群体流失 | Subtract N or directly intersect member lists扣除 N，或直接求成员清单交集 |
| Mixed customer units混用客户单位 | Users, accounts, and organizations have different denominators用户、账户与组织的分母不同 | Declare one unit and test duplicate mappings声明一种单位并检查重复映射 |
| Future leakage未来信息泄漏 | Later statuses rewrite what was knowable at the measurement date后续状态改写测量日当时可知的信息 | Use dated snapshots and as-of queries使用带日期快照与截至日期查询 |
| Unequal cohort ages同期群年龄不等 | Older cohorts have had more time to churn较老同期群经历流失的时间更长 | Compare each cohort at the same elapsed age在相同经过时间上比较各同期群 |
| Claiming causality声称因果 | Observed segment differences may reflect selection or confounding观察到的细分差异可能来自选择偏差或混杂因素 | List alternatives and run a suitable test列出替代解释并开展合适测试 |
Before publishing, confirm 0% ≤ customer CRR ≤ 100%, S is nonzero, counts reconcile to saved member lists, every segment rolls up as expected, missing identities are quantified, and rerunning the query from the same snapshot reproduces the result. Add the formula, period, population, exclusions, raw counts, and data freshness to the report.
发布前应确认客户留存率位于 0% 至 100% 之间、S 不为零、计数可与保存的成员清单对账、各细分可按预期汇总、缺失身份已量化，并且从同一快照重新运行查询能够复现结果。报告中应包含公式、周期、人群、排除项、原始数量与数据新鲜度。
Improve retention without mistaking correlation for proof改进留存，同时避免把相关性误当证明
Start with the largest credible gap, not the loudest anecdote. Pair quantitative patterns with interviews, support records, product incidents, and journey evidence. Translate each finding into a falsifiable hypothesis: a defined population, proposed mechanism, intervention, primary metric, guardrail, and decision date.
应从最可信且影响最大的差距开始，而不是从声音最大的个案开始。把定量模式与访谈、客服记录、产品事故和旅程证据结合。将每项发现转化为可证伪假设：明确人群、作用机制、干预、主要指标、护栏指标和决策日期。
Possible actions include clarifying onboarding, resolving a reliability failure, correcting poor-fit acquisition, improving renewal communication, or changing service recovery. The correct action depends on evidence. Compare treated and suitable control groups where feasible, use the same retention definition, watch for adverse effects, and allow enough time for the natural customer cycle.
可能的行动包括澄清入门流程、解决可靠性故障、纠正不匹配的获客来源、改善续约沟通或调整服务恢复方式。正确行动取决于证据。在可行时比较干预组与适当对照组，保持同一留存定义，观察不利影响，并为自然客户周期留出足够时间。
Customer retention rate FAQ客户留存率常见问题
How do you calculate customer retention rate?
如何计算客户留存率？
Subtract customers acquired during the period from customers at the end, divide by customers at the start, and multiply by 100. Apply identical customer, eligibility, and time-boundary rules to all three counts.
从期末客户数中减去周期内新增客户数，再除以期初客户数并乘以 100。三个计数必须使用完全相同的客户、资格与时间边界规则。
What is a good customer retention rate?
什么是好的客户留存率？
There is no universal good rate. Compare the same customer definition, product, lifecycle stage, and period with your own historical baseline or a genuinely comparable external cohort.
不存在通用的好留存率。应在相同客户定义、产品、生命周期阶段和周期下，与自身历史基线或真正可比的外部同期群比较。
Is customer retention rate the opposite of churn rate?
客户留存率是流失率的反面吗？
They can sum to 100% only when they use the same starting population, period, eligibility rules, and mutually exclusive retained or churned states. Different definitions break that relationship.
只有在使用相同期初人群、周期、资格规则，并且留存与流失状态互斥时，两者之和才会是 100%。定义不同就会破坏这种关系。
Can customer retention rate exceed 100%?
客户留存率可以超过 100% 吗？
Logo or customer retention rate should not exceed 100% because one starting customer can be retained at most once. A result above 100% usually indicates mixed definitions or confusion with revenue retention.
客户数量留存率不应超过 100%，因为一名期初客户最多只能被留存一次。超过 100% 通常表示定义混用，或把客户留存与收入留存混淆。
How often should retention rate be calculated?
应多久计算一次留存率？
Use a cadence that matches the natural purchase or contract cycle, then keep the window fixed for trend comparisons. Fast-use products may need weekly or monthly cohorts; annual contracts need longer observation.
统计频率应匹配自然购买或合同周期，并在趋势比较中保持窗口固定。高频产品可使用周度或月度同期群，年度合同则需要更长观察期。
Official and first-party sources官方与第一方来源
These sources support the formula and retention-analysis concepts in this guide. They do not endorse InfiniSynapse.
以下来源支持本指南使用的公式与留存分析概念，但不代表它们认可 InfiniSynapse。