InfiniSynapse
Retention rate formula: the quick answer留存率公式：快速答案
Retention rate is the percentage of an eligible starting population that remains active, subscribed, or otherwise retained at the end of a defined period. For a customer or subscriber base, subtract new additions from the ending count before dividing by the starting count.
留存率是在明确时间段结束时，期初符合条件的人群中仍然活跃、仍在订阅或仍满足留存条件的百分比。对于客户或订阅者总体，应先从期末人数中减去期间新增人数，再除以期初人数。
Here, S is the eligible population at the start, E is the total eligible population at the end, and N is the people added during the period. Example: start with 1,000 customers, end with 940, and add 140. The retained starting customers are 800, so retention is ((940 − 140) ÷ 1,000) × 100 = 80%.
其中，S 是期初符合条件的人数，E 是期末符合条件的总人数，N 是期间新增人数。示例：期初有 1,000 名客户，期末有 940 名，期间新增 140 名。来自期初且仍被保留的客户是 800 名，因此留存率为 ((940 − 140) ÷ 1,000) × 100 = 80%。
Product teams often need a different but related cohort formula: retained members of the original cohort ÷ original cohort size × 100. It answers “what share of users who first qualified together returned at Day 7, Week 4, or another age?” It should not be mixed with the aggregate customer formula unless the population and event definitions are truly equivalent.
产品团队经常需要另一种相关的队列公式：原始队列中仍满足留存条件的人数 ÷ 原始队列人数 × 100。它回答“同一时间首次满足条件的用户，在第 7 天、第 4 周或其他队列年龄仍回访的比例是多少？”除非人群和事件定义完全等价，否则不要把它与总体客户留存公式混用。
Choose the right retention rate formula如何选择正确的留存率公式
The phrase “retention rate formula” covers several business questions. The arithmetic is simple; the measurement contract is not. First decide what is being retained, what qualifies someone to enter the population, what behavior proves retention, and whether you are measuring a changing account base or a fixed acquisition cohort.
“留存率公式”可能对应多种业务问题。算术很简单，测量契约却不简单。首先要确定保留的对象、进入人群的资格、证明留存的行为，以及测量对象是持续变化的客户总体还是固定的获客队列。
Use ((E − N) ÷ S) × 100 when accounts can be added and lost during a period. Define “customer” consistently: paid account, active contract, or purchaser.
当期间内既有新增也有流失时，使用 ((E − N) ÷ S) × 100。必须统一“客户”的定义，例如付费账户、有效合同或购买者。
Use returning cohort members ÷ original cohort members × 100. Define the entry event, return event, cohort date, observation window, and identity key.
使用回访队列人数 ÷ 原始队列人数 × 100。明确进入事件、回访事件、队列日期、观察窗口和用户身份键。
Use employees from the starting workforce who remain at period end ÷ eligible starting employees × 100. Exclude new hires from retained staff.
使用期末仍在职的期初员工人数 ÷ 符合条件的期初员工人数 × 100。新入职员工不能算作期初员工留存。
Do not reuse a headcount formula. Gross revenue retention and net revenue retention use recurring revenue, with different treatment of expansion and contraction.
不要复用人数公式。毛收入留存和净收入留存以经常性收入为分母，并对增购和缩减采用不同处理方式。
If your question is “how many existing customers stayed?”, use customer retention. If it is “what percentage of sign-ups returned in Week 4?”, use cohort retention. If it is “how much starting recurring revenue remained?”, use revenue retention. A single dashboard can show all three, but each metric needs its own name and definition.
如果问题是“现有客户中有多少留下来”，使用客户留存；如果问题是“注册用户中有多少在第 4 周回访”，使用队列留存；如果问题是“期初经常性收入保留了多少”，使用收入留存。一个仪表盘可以同时展示三者，但每个指标都必须拥有独立名称和定义。
Inputs you need before calculating retention计算留存率前需要准备的输入
A trustworthy retention calculation starts with a written metric definition. Do this before opening a spreadsheet or writing SQL. Otherwise teams can produce different answers from the same records while each query looks technically valid.
可信的留存计算始于书面的指标定义，应在打开表格或编写 SQL 之前完成。否则，不同团队可能从同一批记录得出不同答案，而且每条查询在技术上看起来都正确。
- Unit of analysis: person, account, workspace, subscription, device, household, or employee. Choose one stable identity key and document deduplication rules.
- 分析单位：个人、账户、工作区、订阅、设备、家庭或员工。选择稳定的身份键，并记录去重规则。
- Eligibility: the event or status that admits someone to the denominator. Free trials, refunded orders, internal users, and test accounts may require explicit treatment.
- 资格条件：决定某个对象是否进入分母的事件或状态。免费试用、退款订单、内部用户和测试账户都需要明确处理。
- Start and end boundaries: use a stated timezone and unambiguous inclusive or exclusive timestamps. Calendar months and rolling 30-day windows are not interchangeable.
- 起止边界：使用明确的时区以及清晰的包含或排除规则。自然月与滚动 30 天窗口不能互换。
- Retention condition: an active subscription, purchase, login, meaningful event, or other observable behavior. “Any event” may overstate durable product value.
- 留存条件：有效订阅、购买、登录、有意义的产品事件或其他可观察行为。“任意事件”可能夸大产品的持续价值。
- Late-arriving and deleted data: decide when a period is considered complete and whether historical results are restated after identity merges, refunds, or backfills.
- 迟到与删除数据：决定一个周期何时视为完整，以及身份合并、退款或数据回填后是否重算历史结果。
Data prerequisite: the aggregate formula can work from three counts, but cohort retention needs row-level identities and dated events or status snapshots. Keep raw identifiers protected and use only the minimum access required for analysis.
数据前提：总体公式只需要三个计数，但队列留存需要行级身份和带日期的事件或状态快照。应保护原始标识符，并仅授予分析所需的最小权限。
How to calculate retention rate step by step如何逐步计算留存率
- Write the question and period写明问题和周期
Example: “What percentage of paying accounts active at 00:00 UTC on July 1 remained active at 23:59:59 UTC on July 31?” A precise question prevents silent shifts between monthly, rolling, and cohort logic.
例如：“7 月 1 日 00:00 UTC 有效的付费账户中，有多少在 7 月 31 日 23:59:59 UTC 仍然有效？”精确问题可以避免自然月、滚动窗口和队列逻辑之间悄然切换。
- Build the starting population建立期初人群
Freeze the list or reproducible query for S. Confirm the identity grain and exclude records that do not meet eligibility. Save the row count and extraction time.
固定 S 的名单或可复现查询。确认身份粒度，排除不符合资格的记录，并保存行数与提取时间。
- Determine the ending state and new additions确定期末状态与新增
Count E using the same eligibility rule at period end. Count N only for entities first added after the start boundary. Reinstated or reactivated accounts need an explicit policy.
使用相同资格规则统计期末 E。N 只统计起始边界之后首次新增的对象。恢复订阅或重新激活的账户必须有明确政策。
- Apply the formula at full precision以完整精度应用公式
Calculate ((E − N) ÷ S) × 100 and round only for display. Preserve numerator, denominator, and unrounded value so reviewers can reproduce the result.
计算 ((E − N) ÷ S) × 100，只在展示时四舍五入。保留分子、分母和未舍入值，便于审核者复现。
- Validate the membership, not only the totals验证成员关系，而不只是总数
Where possible, directly join the starting population to retained end-state members. Investigate duplicates, negative counts, retention above 100%, missing identities, and results that change materially after data refresh.
在可能的情况下，将期初人群与期末留存成员直接关联。调查重复、负数、超过 100% 的留存率、缺失身份以及数据刷新后大幅变化的结果。
Retention rate examples for customers and product users客户与产品用户的留存率计算示例
Example 1 — monthly customer retention. This hypothetical subscription business starts April with 2,400 paying accounts. It ends April with 2,310 paying accounts, including 270 accounts first acquired during April. Retained starting accounts equal 2,310 − 270 = 2,040. Customer retention is 2,040 ÷ 2,400 × 100 = 85%. Under the same population and mutually exclusive outcomes, customer churn is 15%.
示例 1——月度客户留存。这是一个假设的订阅业务：4 月初有 2,400 个付费账户，4 月末有 2,310 个，其中 270 个是在 4 月首次获得的账户。留存的期初账户为 2,310 − 270 = 2,040，因此客户留存率为 2,040 ÷ 2,400 × 100 = 85%。在相同人群且结果互斥的前提下，客户流失率为 15%。
Example 2 — Day 7 product retention. A hypothetical cohort contains 500 users whose first qualifying activation occurred on August 1. If 185 of those same users perform the defined core action on exactly Day 7, exact-day retention is 185 ÷ 500 × 100 = 37%. Users acquired after August 1 are not added to the denominator or numerator. If the team instead counts activity on or after Day 7, that is rolling retention and must be labeled differently.
示例 2——第 7 天产品留存。一个假设队列包含 500 名在 8 月 1 日首次完成合格激活的用户。如果其中 185 名同一用户在恰好第 7 天完成定义的核心行为，则精确日留存率为 185 ÷ 500 × 100 = 37%。8 月 1 日之后新获得的用户既不进入分母，也不进入分子。如果团队统计第 7 天或之后任意时间的活动，那是滚动留存，必须使用不同标签。
| Check检查项 | Customer example客户示例 | Cohort example队列示例 |
|---|---|---|
| Denominator分母 | 2,400 accounts active at start期初有效的 2,400 个账户 | 500 activated users in original cohort原始队列中的 500 名激活用户 |
| Numerator分子 | 2,310 ending accounts − 270 new = 2,040期末 2,310 − 新增 270 = 2,040 | 185 original users completed return event185 名原始用户完成回访事件 |
| Result结果 | 85% | 37% |
| Main risk主要风险 | Reactivation or account-status ambiguity重新激活或账户状态含糊 | Exact-day versus rolling-window ambiguity精确日与滚动窗口口径含糊 |
Retention rate vs churn and revenue retention留存率与流失率、收入留存的区别
Related metrics can reveal different failure modes. Logo retention treats each account equally. User cohort retention focuses on return behavior. Revenue retention weights the economic value that remains. Churn measures loss rather than survival. Report the metric name, unit, period, and formula together.
相关指标揭示不同的失败模式。客户数量留存将每个账户等权处理，用户队列留存关注回访行为，收入留存按保留下来的经济价值加权，而流失率衡量损失而非存续。报告时应同时给出指标名称、单位、周期与公式。
| Metric指标 | Best for适用问题 | Core calculation核心计算 | Caution注意事项 |
|---|---|---|---|
| Customer retention rate客户留存率 | Existing account survival现有账户存续 | ((E − N) ÷ S) × 100 | Define new, reactivated, and merged accounts定义新增、重新激活与合并账户 |
| Cohort user retention队列用户留存 | Return behavior by cohort age按队列年龄衡量回访 | Returning original users ÷ original cohort回访的原始用户 ÷ 原始队列 | State event and exact/rolling window说明事件与精确/滚动窗口 |
| Churn rate流失率 | Loss from the starting population期初人群的损失 | Lost starting entities ÷ starting entities流失的期初对象 ÷ 期初对象 | Only complements retention under identical rules仅在规则相同时与留存互补 |
| Gross revenue retention毛收入留存 | Recurring revenue protected before expansion不含增购时保住的经常性收入 | Starting revenue less churn and contraction ÷ starting revenue期初收入减流失与缩减 ÷ 期初收入 | Do not include expansion revenue不要包含增购收入 |
| Net revenue retention净收入留存 | Revenue change within existing base现有客户群内部的收入变化 | Starting revenue plus expansion less contraction and churn ÷ starting revenue期初收入加增购减缩减与流失 ÷ 期初收入 | Can exceed 100%; not a customer-count metric可能超过 100%，且不是客户数量指标 |
Calculate retention from event or warehouse data从事件数据或数据仓库计算留存率
A repeatable retention analysis should preserve membership lists, not only a final percentage. In a warehouse, build a starting-population table at the chosen grain, a dated status or events table, and a metric-definition record. Then derive retention with an explicit join. This makes it possible to inspect who was counted, identify identity changes, and reproduce earlier periods.
可重复的留存分析应保留成员名单，而不仅是最终百分比。在数据仓库中，按选定粒度建立期初人群表、带日期的状态或事件表，以及指标定义记录，再通过明确关联得出留存。这使团队能够检查谁被计入、识别身份变化并复现历史周期。
- Create a metric contract创建指标契约
Record entity grain, qualifying entry, retained state or return event, timezone, period boundaries, exclusions, reactivation treatment, and rounding. Version this definition when it changes.
记录对象粒度、进入资格、留存状态或回访事件、时区、周期边界、排除项、重新激活处理和舍入规则。定义变化时应保留版本。
- Snapshot or reconstruct eligibility快照或重建资格
Use valid-from and valid-to timestamps or reliable snapshots. A current-status column alone cannot reconstruct who was eligible months ago.
使用有效起止时间戳或可靠快照。仅有当前状态字段无法重建数月前哪些对象符合资格。
- Join on the stable identity按稳定身份键关联
Resolve anonymous-to-known user merges, account ownership changes, and duplicate devices consistently. Run a one-to-many join check before aggregating.
一致处理匿名用户合并到已知用户、账户归属变化和重复设备。聚合前应检查一对多关联。
- Aggregate by comparable segments按可比细分聚合
Break out plan, acquisition channel, activation path, country, device, or customer age only when sample sizes remain useful. Compare cohorts at the same age.
可按套餐、获客渠道、激活路径、国家、设备或客户年龄拆分，但必须保证样本量仍有意义，并在相同队列年龄上比较。
- Save evidence for review保存审核证据
Store the query, parameters, source freshness, counts before and after joins, exclusions, and result. A chart without this trail is hard to defend when numbers change.
保存查询、参数、数据新鲜度、关联前后计数、排除项与结果。没有这些记录的图表在数字变化时很难解释。
For product retention, the return condition must be observable. Google Analytics describes a cohort as users sharing a characteristic and its cohort exploration requires both an inclusion criterion and a return criterion. That distinction is useful beyond GA4: it prevents teams from calling any later record “retained” without defining what meaningful return means.
对于产品留存，回访条件必须可观察。Google Analytics 将队列定义为共享某种特征的用户群，其队列探索同时要求进入条件与回访条件。该区分不只适用于 GA4，它能避免团队在没有定义“有意义回访”的情况下，把任何后续记录都称为留存。
Analyze retention across your own data sources跨自有数据源分析留存
Prepare a read-only connection or export containing a stable user or account ID, qualification date, status or event timestamps, and the metric definition you want to test. InfiniSynapse is an AI-powered data analysis tool that can connect to supported databases and warehouses and analyze multi-source data with natural-language questions. It is not presented here as a dedicated retention calculator: you remain responsible for validating the population, query logic, and result.
请准备只读连接或数据导出，其中包含稳定的用户或账户 ID、资格日期、状态或事件时间戳，以及待验证的指标定义。InfiniSynapse 是 AI 辅助数据分析工具，可连接其支持的数据库与数据仓库，并通过自然语言问题分析多源数据。本页不将其描述为专用留存率计算器；你仍需验证人群、查询逻辑和结果。
Open InfiniSynapse Online Data Analyst打开 InfiniSynapse 在线数据分析工具How to interpret retention rate without a false benchmark如何解释留存率并避免错误基准
There is no universal “good retention rate.” A grocery app, annual enterprise contract, social product, marketplace seller, and employee population have different natural cadences and eligibility rules. Public benchmarks frequently mix products, stages, countries, pricing models, and retention definitions. Treat them as context, not a target.
不存在适用于所有场景的“良好留存率”。杂货应用、年度企业合同、社交产品、平台卖家与员工人群具有不同的自然使用频率和资格规则。公开基准经常混合产品类型、发展阶段、国家、定价模式和留存定义，只能作为背景，不能直接作为目标。
A stronger interpretation compares like with like: the same metric contract over time, cohorts at the same age, and segments with adequate counts. Show the numerator and denominator beside the percentage. A move from 40% to 44% means something different for 50 users than for 50,000, and a recent cohort may be incomplete.
更可靠的解释应进行同类比较：在时间上保持同一指标契约，在相同队列年龄比较，并确保细分样本量充足。百分比旁应展示分子与分母。50 名用户从 40% 提升到 44%，与 50,000 名用户的相同变化含义不同；最近的队列还可能尚未成熟。
Track a stable baseline, annotate product or policy changes, and compare mature cohorts. Avoid declaring success from one noisy period.
跟踪稳定基线，标注产品或政策变化，并比较成熟队列。不要根据单个噪声周期宣布成功。
An overall rate can hide high retention in one channel and severe loss in another. Segment only on pre-existing or clearly timed attributes.
总体比例可能掩盖某渠道的高留存和另一渠道的严重流失。仅使用预先存在或时间关系明确的属性分群。
Customer-count retention does not show contract value. Review revenue retention, margin, support cost, and customer concentration where relevant.
客户数量留存无法反映合同价值。必要时同时查看收入留存、利润率、服务成本和客户集中度。
A retention change identifies where to investigate. It does not prove that a release, campaign, or feature caused the change.
留存变化可以指出调查方向，但不能证明某次发布、活动或功能导致了变化。
Common retention calculation mistakes and limitations常见留存率计算错误与局限
- Keeping new customers in the numerator: acquisition can make the ending total look healthy while the original base is shrinking. Subtract N or directly intersect starting and ending members.
- 把新增客户保留在分子中：获客可能让期末总数看起来健康，却掩盖原始客户群缩小。应减去 N，或直接取期初与期末成员交集。
- Changing the denominator: excluding dormant, trial, refunded, or low-value users after seeing the result creates a biased metric. Establish eligibility first.
- 事后改变分母：看到结果后再排除休眠、试用、退款或低价值用户会产生偏差。必须预先确定资格。
- Mixing exact and rolling retention: “active on Day 7” and “active on or after Day 7” answer different questions. Rolling retention cannot be placed on the same curve without a clear label.
- 混用精确留存与滚动留存：“第 7 天活跃”和“第 7 天或之后活跃”回答不同问题。滚动留存不能在未清楚标注的情况下放到同一曲线。
- Comparing cohorts at different ages: a cohort that has existed for two weeks cannot have a complete Month 3 rate. Use only matured cells or mark incomplete data.
- 比较不同年龄的队列：仅存在两周的队列不可能拥有完整的第 3 月留存率。只使用已成熟单元格，或明确标记不完整数据。
- Assuming retention plus churn always equals 100%: the identity holds only with matching population, unit, period, and mutually exclusive states. Reactivation, pauses, expansion, and revenue weighting complicate it.
- 假设留存率与流失率总和始终为 100%：只有在人群、单位、周期相同且状态互斥时才成立。重新激活、暂停、增购和收入加权都会使关系复杂化。
- Ignoring survivorship and observation limits: retention describes observed behavior within available data. Missing events, blocked tracking, identity resets, seasonality, and a short horizon can distort it.
- 忽略幸存者偏差与观察限制：留存只描述可用数据中的已观察行为。事件缺失、追踪受阻、身份重置、季节性和过短观察期都可能造成失真。
Retention is descriptive unless paired with a credible causal design. Use experiments or carefully designed quasi-experiments when you need to determine whether a product change caused retention to improve. A before-and-after chart alone is vulnerable to channel mix, seasonality, pricing changes, and external events.
如果没有可信的因果设计，留存率只是描述性指标。当你需要判断产品改动是否导致留存改善时，应使用实验或经过严谨设计的准实验。单纯的改动前后图表容易受到渠道结构、季节性、定价变化和外部事件影响。
Retention analysis validation checklist留存分析结果验证清单
Before publishing a dashboard or decision memo, run these checks. A retention percentage should be traceable to its population and raw evidence, not merely plausible.
在发布仪表盘或决策备忘录之前，执行以下检查。留存百分比必须能够追溯到人群与原始证据，而不只是看起来合理。
- Metric name identifies customer, user cohort, employee, or revenue retention.
- 指标名称明确是客户、用户队列、员工还是收入留存。
- Start and end timestamps, timezone, and exact or rolling window are visible.
- 起止时间戳、时区以及精确或滚动窗口清晰可见。
- Entry event, return event or retained status, exclusions, and reactivation policy are documented.
- 进入事件、回访事件或留存状态、排除项与重新激活政策均有记录。
- Numerator, denominator, unrounded result, and display rounding are retained.
- 保留分子、分母、未舍入结果与展示舍入规则。
- New additions cannot be misclassified as retained starting members.
- 新增对象不会被误分类为留存的期初成员。
- Duplicate identities, join multiplication, missing events, and late-arriving data have been checked.
- 已检查重复身份、关联膨胀、事件缺失和迟到数据。
- Cohorts are compared at equal age and incomplete periods are labeled.
- 队列在相同年龄比较，不完整周期有明确标记。
- A reviewer can reproduce the result from the saved query or source extract.
- 审核者能够从保存的查询或源数据导出复现结果。
As a next step, connect this metric to its diagnostic neighbors: activation, feature adoption, engagement frequency, funnel drop-off, and cohort composition. Retention tells you that a population stayed or returned; those related measures help explain where the experience may be changing.
下一步可以把该指标与诊断性邻近指标连接起来：激活、功能采用、互动频率、漏斗流失与队列构成。留存告诉你某个人群是否留下或回访，而这些相关指标有助于解释体验可能在哪些环节发生变化。
Retention rate formula FAQ留存率公式常见问题
What is the retention rate formula?
留存率公式是什么？
For a customer or subscriber base, retention rate equals customers at the end of the period minus new customers acquired during the period, divided by customers at the start of the period, multiplied by 100. For a fixed user cohort, divide cohort members who meet the return condition by the original cohort size and multiply by 100.
对于客户或订阅者总体，留存率等于期末客户数减去期间新增客户数，再除以期初客户数并乘以 100。对于固定用户队列，则用满足回访条件的队列成员数除以原始队列人数并乘以 100。
Why do you subtract new customers from the retention rate formula?
为什么留存率公式要减去新增客户？
New customers were not part of the starting population, so they cannot have been retained from that population. Subtracting them prevents acquisition from masking losses among existing customers.
新增客户不属于期初人群，因此不可能是从该人群中被保留下来的客户。减去新增客户可以避免获客掩盖现有客户的流失。
Is retention rate always 100% minus churn rate?
留存率是否总等于 100% 减去流失率？
Only when retention and churn use the same starting population, period, unit, eligibility rules, and mutually exclusive outcomes. Revenue churn, logo churn, rolling retention, and reactivation can break the simple complement relationship.
只有留存与流失使用相同的期初人群、周期、单位、资格规则且结果互斥时才成立。收入流失、客户数量流失、滚动留存和重新激活都可能打破简单互补关系。
How do you calculate Day 7 user retention?
如何计算第 7 天用户留存？
Create a cohort of users who first qualified on the same date, count how many meet the defined return event on Day 7, divide by the original eligible cohort size, and multiply by 100. State whether Day 7 means an exact day, a window, or on-or-after Day 7.
建立同一天首次满足条件的用户队列，统计其中有多少人在第 7 天满足定义的回访事件，再除以原始合格队列人数并乘以 100。同时说明第 7 天是精确日、一个窗口，还是第 7 天及之后。
What is a good retention rate?
什么样的留存率算好？
There is no universal good retention rate. Compare like-for-like cohorts, periods, products, lifecycle stages, and definitions; then use your own historical baseline and business economics to judge improvement.
不存在通用的良好留存率。应比较口径一致的队列、周期、产品、生命周期阶段和定义，再结合自身历史基线与业务经济性判断改善。
Sources and further reading权威来源与延伸阅读
The standard customer-retention definition and E, N, S notation are explained in Salesforce’s first-party guide to calculating customer retention rate. For product cohorts, see the official Google Analytics cohort exploration documentation, which distinguishes cohort inclusion and return criteria. Google also documents how its Retention overview report presents returning users and cohort retention.
标准客户留存定义以及 E、N、S 记号可参考 Salesforce 的客户留存率计算第一方指南。产品队列可参考官方 Google Analytics 队列探索文档，其中区分了队列进入条件与回访条件；Google 还在 留存概览报告文档中说明了回访用户和队列留存的展示方式。
For the broader product-analysis workflow, read the InfiniSynapse guide to data science for product managers, including funnel and cohort-retention distinctions. Product and analytics teams should adapt every formula to their documented business rules and privacy obligations.
如需了解更广泛的产品分析工作流，可阅读 InfiniSynapse 的产品经理数据科学指南，其中解释了漏斗与队列留存的区别。产品与分析团队应根据书面业务规则和隐私义务调整每个公式。