5 Whys Root Cause Analysis: A Practical, Evidence-Led Guide5 Whys 根因分析：以证据为导向的五问法实操指南

What is 5 Whys root cause analysis?什么是 5 Whys 根因分析？

The method is useful because it turns a vague reaction—“fix the broken part” or “retrain the operator”—into a causal conversation. Its output is not automatically proof. Each answer is a hypothesis that should be tied to a record, measurement, observation, test, or qualified account. The final step is not naming a root cause; it is implementing a proportionate control and checking whether the expected outcome changes.

这种方法的价值在于，它能把“修好损坏部件”或“重新培训操作员”等模糊反应，转化为因果讨论。但分析结果不会自动成为证明。每个答案都是假设，应关联记录、测量、观察、试验或合格人员的陈述。最后一步不是给根因命名，而是实施相称的控制，并检查预期结果是否真的改变。

When to use 5 Whys—and when another method fits better何时使用五问法，何时改用其他方法

Use the InfiniSynapse data analysis blog to review related anomaly-detection and investigation topics. Use 5 Whys as one instrument in an investigation, not as a substitute for domain expertise or mandated procedures.

可通过 InfiniSynapse 数据分析博客查看相关异常检测与调查主题。应把五问法作为调查工具之一，而不是替代领域专业知识或强制程序。

Prepare a 5 Why analysis before asking the first question第一次追问前如何准备 5 Why 分析

Start with a neutral problem statement: what happened, where and when it happened, the measured scope, and the expected condition. “Orders released after the cutoff missed same-day dispatch on 18 of 240 cases during the example week” is testable. “The team ignored the procedure” embeds an unverified cause and steers the discussion toward blame.

先写中性问题陈述：发生了什么、何时何地发生、测量范围多大、预期状态是什么。“示例周内，截止时间后释放的 240 个订单中有 18 个未能当天发运”可以检验；“团队无视程序”则提前嵌入了未经验证的原因，并把讨论导向归责。

• Bring people close to the work: include process owners, frontline knowledge, data owners, and an independent facilitator when power dynamics may suppress evidence.让熟悉现场的人参与：包括流程负责人、一线人员、数据负责人；权力关系可能压制证据时，加入独立主持人。
• Freeze a timeline: preserve timestamps, logs, versions, alarms, photos, samples, tickets, interviews, and relevant baseline periods before records change.固定时间线：在记录变化前保存时间戳、日志、版本、告警、照片、样本、工单、访谈和相关基线区间。
• Define evidence quality: label direct observations, system records, measurements, recollections, and inference separately. Note missing or conflicting evidence.定义证据质量：分别标注直接观察、系统记录、测量、回忆和推断，并记录缺失或冲突证据。
• Set the decision boundary: state whether the session can recommend containment, process change, engineering control, policy review, or only further investigation.设定决策边界：说明会议能否建议遏制措施、流程变更、工程控制、政策复核，或只能建议继续调查。

How to run a 5 Whys session without steering the answer如何主持五问分析而不把答案引向预设结论

A productive session is usually short enough to maintain focus but structured enough to resist groupthink. Name a facilitator who is responsible for the method, not for defending a department. Put the problem statement, scope, timeline, and evidence register where everyone can see them. Invite the people who performed the work, people who designed or supervise the process, and someone who can retrieve the relevant records. Clarify at the start that the purpose is to understand how the system produced the outcome—not to negotiate blame or approve a favorite fix.

高质量分析会议既要足够紧凑以保持专注，也要有足够结构来抵抗群体思维。指定一名负责方法、而非替某个部门辩护的主持人；把问题陈述、范围、时间线和证据清单放在所有参与者都能看到的位置。邀请实际执行工作的人、设计或监督流程的人，以及能够调取相关记录的人。会议开始时说明目标是理解系统如何产生该结果，而不是协商责任归属或批准某个偏好的修复方案。

Time-box the first pass, then assign evidence collection rather than filling gaps by consensus. If participants disagree, write competing chains and define the observation that would distinguish them. Close with named owners and due dates for missing evidence, hypothesis tests, containment, corrective-action design, and effectiveness review. The meeting record should show who attended, which version of each source was used, what assumptions remain, and who has authority to approve closure.

为首轮分析设定时限；出现证据缺口时，应分配取证任务，而不是用共识把空白填满。参与者有分歧时，记录相互竞争的因果链，并定义能够区分它们的观察。结束前为缺失证据、假设检验、遏制、纠正措施设计和有效性复核指定负责人及截止日期。会议记录还应列明参与者、各项来源的版本、尚存假设，以及谁有权批准结案。

How to conduct a 5 Whys root cause analysis如何执行 5 Whys 根因分析

1. State the observed problem陈述已观察到的问题Use measurable facts and a comparison condition. Separate immediate containment from the analysis so urgent risk is controlled without declaring a cause.使用可测量事实和对照状态。把即时遏制与根因分析分开，在不提前宣布原因的情况下控制紧急风险。
2. Ask why the problem occurred追问问题为何发生Write the answer as a causal statement, not a label. “The queue exceeded available processing capacity for 42 minutes” is more useful than “capacity issue.”把答案写成因果陈述而非标签。“队列在 42 分钟内超过可用处理能力”比“容量问题”更有用。
3. Attach evidence to the answer为答案关联证据Record the source, time range, owner, confidence, and contradiction. If the answer has no supporting evidence, label it as a hypothesis and plan a test.记录来源、时间范围、负责人、置信程度和矛盾点。答案没有支持证据时，应标为假设并计划检验。
4. Ask why that condition existed继续追问该条件为何存在Use the preceding answer as the subject of the next question. Avoid switching topics or jumping directly to a favored organizational explanation.以上一个答案为下一问的主语，避免换题或直接跳到团队偏好的组织层解释。
5. Branch when the evidence branches证据分叉时保留分支A trigger, enabling condition, prevention-control failure, and detection gap can all matter. Give each material path its own chain rather than compressing them into one cause.触发因素、促成条件、预防控制失效和检测缺口都可能重要。为每条重要路径建立独立链条，不要压缩为单一原因。
6. Test the stopping condition检验停止条件Stop only when the candidate explains the mechanism, is supported by evidence, is actionable at an appropriate system level, and predicts what should change if controlled.只有当候选原因能解释机制、有证据支持、可在适当系统层采取行动，并能预测受控后的变化时才停止。
7. Design and verify corrective action设计并验证纠正措施Assign an owner, due date, baseline, target, verification window, side-effect checks, and reopening rule. Completion is not effectiveness; measure recurrence and control performance.指定负责人、截止日期、基线、目标、验证窗口、副作用检查和重新开启规则。完成不等于有效，应测量复发和控制表现。

5 Whys root cause analysis example: delayed data pipeline5 Whys 根因分析示例：数据管道延迟

Hypothetical example: a scheduled reporting pipeline completed 47 minutes late on three weekdays, causing dashboards to miss their agreed refresh time. The numbers below are illustrative and do not describe an InfiniSynapse customer or product incident.

假设示例：某定时报表数据管道在三个工作日延迟 47 分钟完成，导致仪表板错过约定刷新时间。以下数字仅用于说明，不代表 InfiniSynapse 客户或产品事件。

The team should not stop at “the backfill used capacity” and simply reschedule it. A stronger action package could add a workload class and concurrency limit, require resource-impact review for future batch releases, and alert on queue wait before the dashboard deadline. Verification compares queue wait and on-time refresh before and after the change over a defined window. If delay persists without saturation, the chain is incomplete and must reopen.

团队不应停在“回填占用了容量”，然后只调整任务时间。更稳健的措施组合可以增加工作负载分类与并发限制，要求后续批处理发布评估资源影响，并在仪表板截止时间前对队列等待发出告警。验证时应在明确窗口内比较变更前后的队列等待和准时刷新率。如果没有饱和仍然延迟，说明因果链不完整，必须重新开启调查。

Make Five Whys evidence-led, not opinion-led让五问法由证据驱动，而不是由意见驱动

A useful worksheet has more than “Why?” and “Answer.” Add evidence, source owner, confidence, alternatives, disconfirming observations, action, and verification fields. This makes disagreements visible and prevents the loudest participant from turning intuition into fact. When two explanations fit the same observation, preserve both until a discriminating test separates them.

实用工作表不应只有“为什么”和“答案”，还应加入证据、来源负责人、置信程度、替代解释、反证条件、措施和验证字段。这样能把分歧显性化，避免声音最大的人把直觉变成事实。当两个解释都符合观察时，应保留两者，直到区分性检验把它们分开。

Use one worksheet row for each causal link. Record the problem or preceding condition, the next why question, the proposed answer, and the mechanism connecting them. Beside that claim, identify the evidence item by a stable name or ID, its source, collection time, relevant time window, owner, and access location. Add fields for evidence quality, competing explanations, the next test, test owner, due date, and status. This structure keeps the reasoning auditable even when logs, interviews, measurements, and documents come from different systems.

工作表应为每条因果联系单独设置一行。记录问题或上一层条件、下一次为什么追问、建议答案，以及连接两者的机制。在该主张旁，用稳定名称或 ID 标识证据项，并写明来源、收集时间、相关时间窗口、负责人和访问位置；再增加证据质量、竞争解释、下一项检验、检验负责人、截止日期和状态字段。即使日志、访谈、测量与文档来自不同系统，这种结构也能让推理过程保持可审计。

Avoid reducing evidence quality to a single confidence score. Note whether the item is contemporaneous or recalled, complete or sampled, independently generated or repeated from the same account, and whether its timestamps and definitions match the problem scope. A high-confidence system record can still be irrelevant; several consistent interviews can still share the same missing information. When evidence conflicts, retain the conflict and plan a resolving test instead of averaging opinions. Version the worksheet after material changes so reviewers can see when a hypothesis was introduced, challenged, supported, or retired.

不要把证据质量压缩成单一置信分数。应注明证据是事件发生时同步形成还是事后回忆、完整记录还是抽样、独立生成还是重复同一来源，并检查其时间戳与定义是否匹配问题范围。高置信的系统记录仍可能与问题无关；多份一致访谈也可能共享同一信息缺口。证据冲突时，应保留冲突并安排能够解决分歧的检验，而不是对意见取平均。发生重要变化后还应为工作表创建新版本，让复核者看到假设何时被提出、质疑、支持或淘汰。

Turn a candidate root cause into a corrective-action plan把候选根因转化为纠正措施计划

A root-cause statement is useful only when it supports a control that interrupts the demonstrated mechanism. Write the candidate cause in a testable form: the condition, the pathway by which it produced the problem, the operating context, and the evidence supporting the link. Then generate more than one response. A quick containment action may protect current users, while a corrective action changes the process and a preventive action reduces the chance that the same mechanism appears elsewhere. Keep those purposes separate in the record.

只有当根因陈述能够支持一项中断已证明机制的控制时，它才真正有用。把候选原因写成可检验形式：说明条件、它导致问题的路径、运行环境以及支持该联系的证据；随后提出多个响应方案。快速遏制可以保护当前用户，纠正措施用于改变现有流程，预防措施则降低同一机制在其他位置出现的概率。记录中应清楚区分这三种目的。

Prefer controls that are reliable under normal workload and do not depend on perfect memory. In the pipeline example, a workload class and enforced concurrency limit act directly on resource contention; a checklist update supports future review; training alone is weaker because it does not constrain the system. Before rollout, assess whether the action transfers risk, delays other work, creates access problems, or introduces a new failure mode. Pilot proportionately when possible, preserve the baseline, and define both an effectiveness target and a rollback or reopening condition.

优先选择在正常工作负荷下仍可靠、且不依赖完美记忆的控制。在数据管道示例中，工作负载分类和强制并发限制直接作用于资源争用；检查表更新用于支持后续评审；单独培训较弱，因为它没有约束系统。推广前应评估措施是否会转移风险、延误其他工作、制造访问问题或引入新的故障模式。条件允许时按风险比例开展试点，保留基线，并同时定义有效性目标以及回滚或重新开启条件。

Do not select an action only because it is easy to count. The number of people trained, tickets closed, or documents published measures activity, not risk reduction. Pair implementation measures with an outcome tied to the original problem and a control-performance measure that can reveal degradation before recurrence. Keep the observation window long enough to cover the conditions under which the failure originally appeared.

不要仅因某项措施容易计数就选择它。受训人数、关闭工单数或发布文档数衡量的是活动，而不是风险降低。应把实施指标与关联原问题的结果指标、以及能在复发前揭示控制退化的控制性能指标配对；观察窗口还应足够长，覆盖故障最初出现时的运行条件。

Common 5 Whys mistakes, limitations, and safeguards五问法的常见错误、局限与防护措施

• Asking exactly five times: this can stop too early or dig beyond useful control. Use evidence and actionability as the stopping rule.固定追问五次：可能过早停止，也可能深入到无法有效控制的层面。应以证据与可行动性作为停止规则。
• Blaming a person: “operator error” rarely explains why the action was possible, likely, undetected, or repeated. Examine task design, information, controls, workload, interfaces, and supervision without erasing individual agency.归责个人：“操作员错误”通常无法解释该行为为何可能发生、容易发生、未被发现或重复发生。应检查任务设计、信息、控制、工作负荷、界面和监督，同时不抹去个人责任。
• Confusing correlation with cause: a condition that appeared before the event may be a marker rather than a mechanism. Seek a counterfactual, comparison, controlled change, reproduction, or other appropriate test.把相关当因果：事件前出现的条件可能只是标志而非机制。应寻找反事实、对照、受控变更、复现或其他适当检验。
• Forcing one chain: complex failures commonly combine a trigger, enabling conditions, failed barriers, and detection gaps. Branch the analysis or switch methods.强迫使用单链：复杂故障通常结合触发因素、促成条件、屏障失效和检测缺口。应保留分支或更换方法。
• Closing on action completion: training delivered, code deployed, or policy published does not prove risk reduction. Predefine outcome, control, and side-effect measures.措施完成即结案：完成培训、部署代码或发布政策，并不能证明风险下降。应预先定义结果指标、控制指标和副作用指标。

The method is especially vulnerable to hindsight bias, incomplete records, facilitator framing, organizational pressure, and different teams producing different chains. High-consequence findings need proportionate independent review and any required professional, regulatory, clinical, engineering, or safety process.

该方法尤其容易受到后见之明偏差、记录不完整、主持人框架、组织压力以及不同团队得出不同链条的影响。高后果结论需要相称的独立复核，并遵守适用的专业、监管、临床、工程或安全程序。

How to verify a 5 Whys root cause and corrective action如何验证五问法根因与纠正措施

Document a reopening rule before declaring success: recurrence, contradictory evidence, failure to meet the target, or an unacceptable side effect should trigger review. When you need evidence traceability, assignments, version history, and follow-up records, compare the available entries in the InfiniSynapse tools directory without assuming a general analysis workspace is dedicated RCA software.

在宣布成功前，应记录重新开启规则：出现复发、矛盾证据、未达到目标或不可接受的副作用时必须复核。需要证据追踪、任务分配、版本历史和跟进记录时，可比较 InfiniSynapse 工具目录中的可用入口，但不要把通用分析工作区假定为专用 RCA 软件。

Frequently asked questions about 5 Whys analysis5 Whys 分析常见问题

Official sources and verification notes官方来源与验证说明

• U.S. Centers for Medicare & Medicaid Services: Five Whys Tool for Root Cause Analysis—official guidance on problem statements, team facilitation, continued questioning, validation, and when to consider a broader method.
• Oregon OSHA: Root Cause Analysis workbook—official training material covering fact collection, event sequences, systems-oriented causes, alternative analysis tools, control strength, recommended improvements, and follow-up.

• 美国医疗保险和医疗补助服务中心：Five Whys 根因分析工具——关于问题陈述、团队主持、持续追问、根因验证以及何时考虑更广泛分析方法的官方指南。
• Oregon OSHA：根因分析工作簿——涵盖事实收集、事件顺序、系统性原因、替代分析工具、控制强度、改进建议与跟进的官方培训材料。

These sources support the method, cautions, and verification principles used in this guide. They do not establish one universal Five Whys procedure for every industry or prove that a proposed cause is correct. Recheck the exact regulations, standards, organizational procedures, evidence requirements, and approval responsibilities that apply to the incident being investigated. Examples and worksheets on this page are practical learning aids; adapt them to the relevant engineering, clinical, safety, quality, privacy, security, or other professional review process, and retain qualified independent oversight when the potential consequences are significant.

这些来源支持本指南采用的方法、注意事项与验证原则，但并未为所有行业建立一套通用的五问法程序，也不能自动证明候选原因正确。请针对正在调查的事件，重新核对适用的法规、标准、组织程序、证据要求与审批责任。本页示例和工作表属于实用学习辅助材料；实际使用时，应根据相关工程、临床、安全、质量、隐私、信息安全或其他专业复核流程进行调整，并在潜在后果重大时保留合格的独立监督。