customer-research

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

When this skill is activated, always start your first response with the 🧢 emoji.

激活此技能后，首次回复需以🧢表情开头。

Customer Research

客户研究

Customer research is the systematic practice of understanding who your customers are, what they need, and how they behave. It combines qualitative methods (interviews, open-ended surveys) with quantitative methods (NPS, structured surveys, behavioral analytics) and synthesis techniques (persona building, segmentation, journey mapping). This skill equips an agent to design research instruments, analyze collected data, and produce actionable artifacts like personas, insight reports, and research-backed recommendations.

客户研究是系统性了解客户身份、需求及行为的实践。它结合了定性研究方法（访谈、开放式问卷）、定量研究方法（NPS、结构化问卷、行为分析）以及综合分析技术（用户画像构建、客户细分、旅程地图绘制）。此技能可让Agent设计研究工具、分析收集到的数据，并产出可落地的成果，如用户画像、洞察报告及基于研究的建议。

When to use this skill

何时使用此技能

Trigger this skill when the user:

Wants to design a customer survey or questionnaire
Needs an interview guide, script, or recruiting screener
Asks to analyze or interpret NPS (Net Promoter Score) data
Wants to set up or interpret behavioral analytics (funnels, cohorts, retention)
Needs to build, refine, or validate user personas
Asks about customer segmentation or Jobs To Be Done (JTBD) frameworks
Wants to synthesize qualitative data (affinity mapping, thematic analysis)
Needs a research plan or study design for a product initiative

Do NOT trigger this skill for:

Market sizing, competitive analysis, or pricing strategy (market research, not customer research)
A/B testing or experimentation design (product experimentation, not research)

当用户有以下需求时触发此技能：

想要设计客户调研问卷
需要访谈指南、脚本或招募筛选问卷
要求分析或解读NPS（净推荐值）数据
想要设置或解读行为分析数据（转化漏斗、用户群组、留存率）
需要构建、优化或验证用户画像
咨询客户细分或Jobs To Be Done（JTBD）框架
想要整合定性研究数据（亲和图分析、主题分析）
需要为产品项目制定研究计划或研究设计

请勿在以下场景触发此技能：

市场规模测算、竞品分析或定价策略（属于市场研究，而非客户研究）
A/B测试或实验设计（属于产品实验，而非研究）

Key principles

核心原则

Research question first - Every research activity starts with a clear question. "What do we want to learn?" comes before "What method should we use?" A survey without a research question produces data without insight.
Triangulate methods - Never rely on a single source. Combine qualitative (interviews, open-ended responses) with quantitative (surveys, analytics) to validate findings. What people say they do and what they actually do often diverge.
Bias awareness - Every method introduces bias. Surveys have response bias and question-order effects. Interviews have interviewer bias and social desirability. Analytics miss intent and context. Name the bias, design around it, caveat findings.
Sample matters more than size - A well-recruited sample of 8 interview participants produces better insight than a poorly targeted survey of 1,000. Define the target population, screen rigorously, aim for representation over volume.
Actionability over thoroughness - Research that does not change a decision is wasted effort. Every deliverable should answer: "What should we do differently based on this?" If the answer is nothing, the research question was wrong.

先明确研究问题 - 所有研究活动都始于清晰的问题。“我们想要了解什么？”的优先级高于“我们应使用什么方法？”。没有研究问题的问卷只会产生无洞察的数据。
多方法交叉验证 - 绝不依赖单一数据源。结合定性（访谈、开放式回复）与定量（问卷、分析数据）方法验证研究结果。人们口中的行为与实际行为往往存在差异。
警惕研究偏差 - 每种方法都会引入偏差。问卷存在回复偏差和问题顺序效应；访谈存在访谈者偏差和社会期望偏差；行为分析无法体现用户意图与背景。要明确偏差类型，在设计时规避，并在结论中注明局限性。
样本质量优于数量 - 8名精准招募的访谈参与者能比1000名目标不符的问卷受访者提供更有价值的洞察。明确定标人群，严格筛选样本，优先代表性而非数量。
落地性优于全面性 - 无法影响决策的研究都是无用功。每份交付成果都应回答：“基于此，我们应做出哪些改变？”如果答案是无，说明研究问题本身存在错误。

Core concepts

核心概念

Research methods spectrum - Methods range from qualitative (rich, small-n, exploratory) to quantitative (structured, large-n, confirmatory). Qualitative methods (interviews, diary studies, contextual inquiry) generate hypotheses. Quantitative methods (surveys, analytics, NPS) test them. The best research programs cycle between the two.

Voice of Customer (VoC) - The aggregate understanding of customer needs, expectations, and pain points across all channels - support tickets, survey verbatims, interview transcripts, reviews, social mentions. VoC is an ongoing program, not a one-time project.

Jobs To Be Done (JTBD) - A framework that reframes needs as "jobs" customers hire products to do. Format: "When [situation], I want to [motivation], so I can [outcome]." This prevents feature-driven thinking and keeps research anchored to outcomes.

Research operations (ResearchOps) - The infrastructure layer: participant recruitment panels, consent and privacy workflows, data repositories, insight libraries. Without ResearchOps, each study starts from scratch and insights get lost between teams.

研究方法谱系 - 研究方法从定性（信息丰富、小样本、探索性）到定量（结构化、大样本、验证性）不等。定性方法（访谈、日记研究、情境调研）用于生成假设，定量方法（问卷、行为分析、NPS）用于验证假设。优秀的研究项目会在两种方法间循环推进。

Voice of Customer（VoC，客户心声） - 整合所有渠道的客户需求、期望与痛点的综合认知，包括支持工单、问卷回复、访谈记录、评论及社交提及。VoC是持续进行的项目，而非一次性任务。

Jobs To Be Done（JTBD，用户待办任务） - 一种将用户需求重构为“客户雇佣产品完成的任务”的框架。格式为：“当[场景]时，我想要[动机]，以便[结果]”。这能避免以功能为导向的思维，让研究始终聚焦用户成果。

ResearchOps（研究运营） - 研究的基础设施层：参与者招募面板、知情同意与隐私流程、数据存储库、洞察库。没有ResearchOps，每项研究都要从零开始，且洞察会在团队间流失。

Common tasks

常见任务

Design a customer survey

设计客户调研问卷

Start with the research question - what decision will this survey inform? Structure:

Screener questions (1-3) - Filter out non-target respondents early
Warm-up questions (1-2) - Easy, non-threatening questions to build engagement
Core questions (5-10) - The questions that answer the research question
Demographics (2-4) - At the end, not the beginning (reduces drop-off)

Key rules: one concept per question, avoid leading language, use 5-point Likert scales for attitudes, randomize option order, limit open-ended questions to 2-3, target 5-7 minutes completion time (12-15 questions max).

See

references/surveys.md

for question type catalog, scale design, and distribution.

从研究问题入手——这份问卷将为哪项决策提供依据？结构如下：

筛选问题（1-3个）- 提前排除非目标受访者
热身问题（1-2个）- 简单无压力的问题，提升参与度
核心问题（5-10个）- 直接回答研究问题的关键问题
人口统计问题（2-4个）- 放在末尾而非开头（降低流失率）

关键规则：一个问题对应一个概念，避免诱导性语言，态度类问题使用5级李克特量表，随机排列选项顺序，开放式问题限制在2-3个，目标完成时间为5-7分钟（最多12-15个问题）。

可查看

references/surveys.md

获取问题类型目录、量表设计及分发指南。

Create an interview guide

撰写访谈指南

Structure a 45-60 minute semi-structured interview in five blocks:

Introduction (5 min) - Purpose, consent, expectations
Context (10 min) - Role, workflow, environment
Core exploration (25 min) - Open-ended deep-dive on the research topic
Reactions (10 min) - Show prototypes or concepts if applicable
Wrap-up (5 min) - "Anything else?", next steps, thanks

Technique rules: ask "how" and "why" not "do you"; use "tell me about a time when..." for behavioral recall; use the 5-second silence technique after answers; never suggest answers or finish sentences; record verbatim quotes.

See

references/interviews.md

for the full protocol and analysis framework.

将45-60分钟的半结构化访谈分为五个模块：

开场介绍（5分钟）- 研究目的、知情同意、访谈预期
背景了解（10分钟）- 用户角色、工作流程、使用环境
核心探索（25分钟）- 针对研究主题的开放式深度访谈
反馈收集（10分钟）- 如有需要，展示原型或概念并收集反馈
收尾总结（5分钟）- “还有其他补充吗？”、后续安排、致谢

技巧规则：多问“如何”和“为什么”，而非“你是否”；用“请描述一次你……的经历”引导行为回忆；在用户回答后使用5秒沉默技巧；绝不暗示答案或帮用户完成句子；记录完整的原话引用。

可查看

references/interviews.md

获取完整流程及分析框架。

Conduct NPS deep-dive analysis

进行NPS深度分析

NPS asks: "How likely are you to recommend [product]?" on a 0-10 scale. Promoters (9-10), Passives (7-8), Detractors (0-6). NPS = %Promoters - %Detractors.

Go beyond the top-line score:

Segment by cohort - NPS by tenure, plan tier, use case, geography
Analyze the follow-up - The open-ended "why" is where the insight lives
Track trends - Monthly/quarterly trends matter more than any single score
Cross-reference behavior - Do Promoters refer? Do Detractors churn?
Close the loop - Contact Detractors within 48 hours; understand Passive blockers

See

references/nps-analysis.md

for scoring methodology, benchmarks, and coding.

NPS的问题为：“你向他人推荐[产品]的可能性有多大？”评分范围0-10分。推荐者（9-10分）、被动者（7-8分）、贬损者（0-6分）。NPS计算公式为：推荐者占比 - 贬损者占比。

不要仅停留在表面分数：

按群组细分 - 按使用时长、套餐等级、使用场景、地域划分NPS
分析跟进问题 - 开放式的“为什么”才是洞察的来源
追踪趋势 - 月度/季度趋势比单次分数更重要
关联行为数据 - 推荐者是否真的会推荐？贬损者是否会流失？
闭环跟进 - 在48小时内联系贬损者；了解被动者的顾虑

可查看

references/nps-analysis.md

获取评分方法、行业基准及编码指南。

Analyze behavioral analytics

分析行为数据

Define key behavioral metrics for a product:

Activation - What action signals a user "gets it"? (e.g., created first project)
Engagement - What does healthy usage look like? (DAU/MAU ratio, session frequency)
Retention - Cohort retention curves: Day 1, Day 7, Day 30 benchmarks
Funnel analysis - Map the critical path and measure drop-off at each step
Feature adoption - Which features correlate with retention? (correlation, not causation)

Behavioral analytics answers "what" and "how much" but never "why." Always pair with qualitative methods to interpret observed patterns.

See

references/behavioral-analytics.md

for metrics frameworks and cohort analysis.

为产品定义关键行为指标：

激活 - 哪个动作标志用户“理解产品价值”？（例如：创建首个项目）
参与度 - 健康的使用状态是什么样的？（日活/月活比、会话频率）
留存率 - 群组留存曲线：第1天、第7天、第30天基准
漏斗分析 - 绘制关键路径并测量各环节的流失率
功能 adoption - 哪些功能与留存率相关？（仅为相关性，非因果关系）

行为数据分析回答“是什么”和“有多少”，但永远无法回答“为什么”。需始终结合定性方法解读观察到的模式。

可查看

references/behavioral-analytics.md

获取指标框架及群组分析指南。

Build user personas

构建用户画像

Personas are archetypes synthesized from real data - not fictional characters from a workshop. Process:

Gather data - Combine interview transcripts, survey responses, analytics segments
Identify patterns - Affinity mapping to cluster behaviors, goals, pain points
Define dimensions - Choose 2-3 differentiating axes (e.g., skill vs. frequency)
Draft personas (3-5 max) - Each includes: name/role, key goals, pain points, behavioral patterns, real verbatim quotes, JTBD statement
Validate - Test personas against held-out data; refine until predictive

Personas without behavioral data are stereotypes. Always ground them in observation.

See

references/personas.md

for the persona template, affinity mapping guide, and validation checklist.

用户画像是基于真实数据合成的典型用户原型，而非 workshop 中的虚构角色。流程如下：

收集数据 - 整合访谈记录、问卷回复、分析数据细分群体
识别模式 - 用亲和图分析聚类行为、目标、痛点
定义维度 - 选择2-3个差异化维度（例如：技能水平 vs 使用频率）
绘制画像草稿（最多3-5个）- 每个画像包含：姓名/角色、核心目标、痛点、行为模式、真实原话引用、JTBD陈述
验证 - 用留存数据测试画像；优化至具备预测性

缺乏行为数据的用户画像只是刻板印象。需始终基于观察数据构建画像。

可查看

references/personas.md

获取画像模板、亲和图分析指南及验证清单。

Synthesize qualitative research data

整合定性研究数据

After collecting interview transcripts or open-ended survey responses:

Code the data - Tag recurring themes with descriptive codes
Affinity map - Group related codes into clusters; name each cluster
Identify patterns - Frequency (how often) and intensity (how strongly felt)
Build insight statements - "[Observation] because [reason], which means [implication for product]"
Prioritize - Rank by frequency, severity, and business alignment
Report - Executive summary, methodology, 3-5 key findings, recommendations

收集到访谈记录或开放式问卷回复后：

数据编码 - 用描述性标签标记重复出现的主题
亲和图聚类 - 将相关标签分组，并为每组命名
识别模式 - 统计频率（出现次数）和强度（感受强烈程度）
撰写洞察陈述 - “[观察结果]，因为[原因]，这意味着[对产品的影响]”
优先级排序 - 按频率、严重程度及业务契合度排序
撰写报告 - 执行摘要、研究方法、3-5个核心发现、建议

Write a research plan

撰写研究计划

For any new research initiative, produce a one-page research plan:

Background - What prompted this research? (2-3 sentences)
Research questions - 2-4 specific questions to answer
Method - Which method(s) and why; sample size and criteria
Timeline - Recruit, conduct, analyze, report milestones
Deliverables - What artifacts will be produced (personas, report, recommendations)
Stakeholders - Who needs the findings and in what format

针对任何新的研究项目，需产出一页纸的研究计划：

背景 - 开展此研究的动因？（2-3句话）
研究问题 - 2-4个具体的待回答问题
研究方法 - 使用的方法及原因；样本量与筛选标准
时间线 - 招募、执行、分析、报告的关键节点
交付成果 - 将产出的成果（用户画像、报告、建议）
利益相关方 - 谁需要研究结果，以及所需的呈现格式

Anti-patterns / common mistakes

反模式 / 常见错误

Mistake	Why it's wrong	What to do instead
Starting with the solution ("Do you want feature X?")	Confirmation bias - users agree to please you	Start with the problem space; let solutions emerge from patterns
Surveying without a research question	Produces data without insight; analysis becomes fishing	Define the decision the survey informs before writing questions
Using NPS as the only customer metric	NPS measures sentiment, not behavior; it is lagging and blunt	Combine NPS with behavioral metrics, CSAT, and qualitative feedback
Recruiting only power users	Survivor bias - misses churned and non-adopters	Recruit across segments including lapsed and churned users
Creating personas from assumptions	Personas without data reinforce existing biases	Ground every persona attribute in observed research data
Asking leading questions	"Don't you think X is frustrating?" always gets agreement	Use neutral, open-ended phrasing: "Tell me about your experience with X"
Ignoring small sample findings	5 interviews surfacing the same pain point is a strong signal	Qualitative validity comes from pattern saturation, not sample size

常见错误	问题所在	正确做法
从解决方案入手（例如：“你想要功能X吗？”）	存在确认偏差——用户会为了迎合你而同意	从问题领域入手；让解决方案从模式中自然浮现
无研究问题就开展问卷调研	产生无洞察的数据；分析变成无的放矢	在设计问题前，先明确问卷将支持哪项决策
将NPS作为唯一的客户指标	NPS测量的是用户情绪，而非行为；属于滞后且模糊的指标	结合NPS与行为指标、CSAT及定性反馈
仅招募核心用户	存在幸存者偏差——忽略了流失用户与未采用者	跨细分群体招募，包括流失用户与非活跃用户
基于假设创建用户画像	无数据支撑的用户画像会强化现有偏见	每个画像属性都需基于观察到的研究数据
提出诱导性问题	例如“你不觉得X很令人沮丧吗？”总会得到肯定回答	使用中立、开放式的表述：“请描述你使用X的体验”
忽略小样本的研究发现	5次访谈都提到的痛点是强烈的信号	定性研究的有效性来自模式饱和，而非样本量

References

参考资料

For detailed methodology on specific research techniques, read the relevant file from

references/

```
references/surveys.md
```
- Question types, scale design, sampling, distribution. Load when designing or reviewing a survey.
```
references/interviews.md
```
- Full interview protocol, recruiting, consent, thematic analysis. Load when planning or analyzing interviews.
```
references/nps-analysis.md
```
- Scoring methodology, benchmarks, verbatim coding, closed-loop process. Load when analyzing NPS data.
```
references/behavioral-analytics.md
```
- Metrics frameworks (AARRR, North Star), cohort analysis, funnel design. Load when setting up or interpreting analytics.
```
references/personas.md
```
- Persona template, affinity mapping, validation checklist, worked example. Load when building or refining personas.

Only load a references file if the current task requires it.

如需了解特定研究技术的详细方法，请查看

references/

目录下的对应文件：

```
references/surveys.md
```
- 问题类型、量表设计、抽样、分发设计或审核问卷时加载此文件。
```
references/interviews.md
```
- 完整访谈流程、招募、知情同意、主题分析规划或分析访谈时加载此文件。
```
references/nps-analysis.md
```
- 评分方法、行业基准、原话编码、闭环流程分析NPS数据时加载此文件。
```
references/behavioral-analytics.md
```
- 指标框架（AARRR、北极星指标）、群组分析、漏斗设计设置或解读分析数据时加载此文件。
```
references/personas.md
```
- 用户画像模板、亲和图分析指南、验证清单构建或优化用户画像时加载此文件。

仅在当前任务需要时加载参考文件。