把市场问题，拆成可以回答的数据问题Turning Market Questions into Answerable Ones
定量与定性结合，线上与线下并行：先确定口径与样本，再收集、交叉验证，最后给结论和局限。 Quantitative and qualitative, online and offline: define the methodology and sample first, then collect, triangulate, and state both conclusions and limits.
四种证据来源，互相补短板Four Sources of Evidence
单一方法都有偏差；我们优先用两种以上方法交叉印证同一结论。Any single method carries bias; we cross-check each conclusion with at least two.
定量研究Quantitative
抽样设计、结构化问卷、量化访谈；给出样本代表性、置信区间与误差说明。Sampling design, structured surveys and quantified interviews — with representativeness, intervals and error stated.
定性研究Qualitative
深访、焦点小组、专家访谈；用来生成假设、解释"为什么"，不假装代表总体。In-depth interviews, focus groups and expert calls — to generate hypotheses and explain why, never passed off as the population.
实地研究Fieldwork
门店与终端走访、货架与陈列观察、渠道访谈；还原真实价格、铺货与动销。Store and terminal visits, shelf and display audits, channel interviews — real prices, distribution and sell-through.
线上研究Online
公开披露、电商与社媒、搜索与舆情；补足广度、频次与趋势。Public disclosures, e-commerce and social data, search and sentiment — adding breadth, frequency and trend.
常用技术：分层与配额抽样、结构化问卷、深访、焦点小组、专家访谈（Delphi）、门店与货架观察、神秘顾客、桌面研究。 Typical tools: stratified and quota sampling, structured questionnaires, depth interviews, focus groups, expert interviews (Delphi), store and shelf audits, mystery shopping, desk research.
方法按问题选，不堆模型Methods Chosen to the Question, Not Stacked
先看样本与假设能不能支撑，再决定用什么方法；支撑不了，就降级为描述性结论并写明。First check whether the sample and assumptions can carry the method; if not, we step down to descriptive findings and say so.
统计方法Statistics
- 抽样与样本量Sampling & sample size分层、配额、整群抽样；按检验功效倒推最小样本，避免"跑完再说"。 Stratified, quota and cluster sampling; minimum sample derived from power, not decided after the fact.
- 误差与置信区间Error & intervals抽样误差、设计效应与加权；给区间，不给单一数字。Sampling error, design effect and weighting; ranges rather than single points.
- 假设检验Hypothesis testingt 检验、卡方、方差分析；多重比较做校正，避免"挑显著的讲"。t-tests, chi-square, ANOVA; multiple-comparison correction to avoid cherry-picking significance.
- 回归与离散选择Regression & choice modelsOLS、Logit/Probit、联合分析（Conjoint）；先看共线性与稳健性。OLS, logit/probit, conjoint — collinearity and robustness checked first.
- 降维与分群Reduction & clustering因子分析与聚类，用于找结构；结果要能解释、能落地。Factor analysis and clustering to find structure — results must be explainable and usable.
- 时间序列Time series趋势分解、季节调整、自相关；预测区间与回测一并给出。Decomposition, seasonal adjustment, autocorrelation; forecast intervals and backtests included.
分析方法Analysis
- 描述与基准对比Descriptive & benchmarking分布、集中度、分位；先回答"是多少"，再谈"为什么"。Distribution, concentration, percentiles — what before why.
- 归因与贡献拆解Attribution量价拆分、结构贡献、驱动因素；拆到不能再拆为止。Volume-price, mix and driver decomposition until it cannot be split further.
- 漏斗与队列Funnel & cohort转化路径、留存曲线、同群对比；区分"时间效应"与"策略效应"。Conversion paths, retention curves and cohort comparison — separating time effects from strategy effects.
- 分群与画像SegmentationRFM、聚类、规则分层；每群要对应一个可执行动作。RFM, clustering, rule-based tiers — each segment tied to an action.
- 情景与敏感性Scenarios & sensitivity基准/乐观/悲观三档，标注关键假设与临界点。Base, upside and downside cases with key assumptions and tipping points.
- 文本与情感Text & sentiment主题归并、情感倾向、热度与来源权重；人工抽检校核。Topic grouping, sentiment, heat and source weighting — human spot-checks.
原则：样本不足、假设不成立时，不出显著性结论；模型只用来解释结构，不用来制造确定性。 Principle: with thin samples or unmet assumptions we do not report significance; models explain structure, they do not manufacture certainty.
常见八类问题Eight Recurring Questions
按"要不要进、跟谁打、卖给谁、怎么卖、卖得怎样"组织，分别落在行业、公司、产品、财务与金融数据五条主线上；具体设计随问题调整。Organised as whether to enter, whom you compete with, whom you sell to, how you sell, and how it performs — the design follows the question.
市场进入与规模Market Entry & Sizing
自上而下与自下而上两套测算，差异超过阈值就回到假设逐项核对。Top-down and bottom-up sizing; when they diverge beyond a threshold, we go back through the assumptions line by line.
竞争格局与份额Landscape & Share
集中度、梯队与替代关系；份额按口径分别给区间，不报单一数字。Concentration, tiers and substitution; share given as ranges per definition, not a single number.
用户需求与分层Needs & Segmentation
需求优先级、决策路径与未满足点；分层之后要能对应到动作。Need priorities, decision paths and unmet demand — segments only matter if they map to actions.
品牌与传播Brand & Communication
认知、提及与转化漏斗；区分"知道"与"会选择"。Awareness, recall and the conversion funnel — separating knowing from choosing.
产品与定价测试Product & Pricing Tests
概念测试、价格敏感度、组合与包装；给区间与情景，不给"最优价"。Concept tests, price sensitivity, assortment and packaging — ranges and scenarios, never a single "best price".
渠道与终端Channels & Retail
渠道结构、利润分配与终端执行；走访记录与价格扫描作为原始证据。Channel structure, margin split and in-store execution — visit notes and price scans as raw evidence.
满意度与流失Satisfaction & Churn
满意度与推荐意愿、流失原因归因；样本不足时明确标注不可外推。Satisfaction, recommendation and churn reasons — flagged as non-generalisable when the sample is thin.
行业与政策扫描Sector & Policy Scan
政策、技术与供需变化的定期跟踪，输出时间轴与影响路径。Periodic tracking of policy, technology and supply-demand shifts — timelines and transmission paths.
四种形态，按决策节奏选Four Formats, Chosen to the Decision Cadence
研究报告Research Report
结论先行，附方法、样本、数据来源与局限。Conclusion first, with method, sample, sources and limits.
数据看板Dashboard
把关键指标沉淀为可持续更新的视图。Key metrics kept in a view that keeps updating.
定期简报Periodic Brief
按周或按月跟踪，口径长期一致，趋势可比。Weekly or monthly tracking on a stable methodology, so trends compare.
汇报与工作坊Briefing & Workshop
把结论讲成决策语言，现场答疑并形成行动项。Findings translated into decision language, with Q&A and action items.
五步，每步都有交付物Five Steps, Each With an Output
界定问题Frame
明确决策用途与可接受的误差。Clarify the decision and acceptable error.
设计口径Design
定义指标口径、数据来源与抽样或样本框。Define metrics, sources and the sampling frame.
采集与核验Collect & verify
多源采集、交叉印证，过程留痕。Multi-source collection, cross-checked and logged.
分析建模Analyse
对照、归因与情景测算，结论与数据一一对应。Comparison, attribution and scenarios — every claim tied to evidence.
交付跟踪Deliver
结论、局限与后续跟踪计划。Conclusions, limits and a follow-up plan.
先说清楚什么不能保证What We Do Not Promise
- 样本Sample写清样本框、规模与招募方式；样本不足时不外推。Frame, size and recruitment stated; no extrapolation on thin samples.
- 误差Error定量给区间，定性不给百分比。Ranges for quantitative work; no percentages from qualitative.
- 来源Sources公开数据、访谈记录、价格扫描分别标注，可回溯。Public data, interview notes and price scans labelled separately and traceable.
- 隐私Privacy受访者信息脱敏，仅用于本项研究。Respondent data anonymised and used only for the study.
- 局限Limits结论有边界；与预期不符时，我们报告不符，而不是换一种说法。Conclusions have boundaries; when findings contradict expectations, we report that rather than rephrasing.
说明：线上与线下的市场调研能力仍在建设中，现阶段按项目承接，范围、样本与周期先行沟通。 Note: our online and offline research capability is still being built; projects are taken case by case, with scope, sample and timeline agreed first.
先聊问题，再谈方案Start With the Question, Not the Proposal
把你的决策场景说清楚：要回答什么、什么时候要、能接受多大误差。我们据此判断要不要做、怎么做。Tell us the decision, the deadline and the error you can live with — we will say whether and how to research it.
查看服务方案View Services