AI 工具 SaaS 行业普遍面临三大增长瓶颈，业务侧虽已具备较完善的埋点数据采集能力，但缺少系统化的漏斗归因分析与用户分层洞察，产品迭代与运营决策缺乏数据支撑。

本项目目标：通过 SQL 驱动的数据分析，构建完整的转化漏斗模型，定位每一步流失卡点，输出可落地的产品与运营策略，并完成高保真原型改版设计稿。

2.1 数据集概览

2.2 五张核心数据表

① users（用户主表）

3.1 第 1 步：脏数据诊断（先查再洗）

在执行清洗前，先使用 SQL 全面诊断四类数据问题，避免盲目清洗导致信息丢失：

-- 1.1 重复主键检查
SELECT user_id, COUNT(*) AS cnt
FROM users GROUP BY user_id HAVING cnt > 1;

-- 1.2 空值检查
SELECT
  SUM(CASE WHEN user_id   IS NULL THEN 1 ELSE 0 END) AS null_uid,
  SUM(CASE WHEN mrr       IS NULL THEN 1 ELSE 0 END) AS null_mrr
FROM users;

-- 1.3 异常值与外键完整性检查
SELECT MIN(duration_sec), MAX(duration_sec), MIN(mrr), MAX(mrr)
FROM events, subscriptions;

3.2 第 2 步：去重处理（SQLite 兼容写法）

-- MySQL 写法（支持窗口函数）
CREATE TABLE cleaned_users AS
SELECT * FROM (
  SELECT *, ROW_NUMBER() OVER(PARTITION BY user_id ORDER BY signup_time) AS rn
  FROM users
) t WHERE rn = 1;

-- SQLite 兼容写法（子查询去重）
CREATE TABLE cleaned_users AS
SELECT u1.*
FROM users u1
WHERE u1.rowid = (
  SELECT MIN(u2.rowid) FROM users u2
  WHERE u2.user_id = u1.user_id
);

3.3 第 3-6 步：完整清洗 SQL

-- 空值处理：分类字段补默认值，数值字段补 0
UPDATE cleaned_users SET
  plan         = COALESCE(plan, 'free'),
  industry     = COALESCE(industry, '未知'),
  company_size = COALESCE(company_size, '1-10');

-- 异常值过滤：duration 超 24h / mrr 负数 / 漏斗标志位非 0-1
DELETE FROM cleaned_events WHERE duration_sec > 86400 OR duration_sec < 0;
UPDATE cleaned_users SET
  is_activate = CASE WHEN is_activate NOT IN (0,1) THEN 0 ELSE is_activate END;

-- 时间格式标准化 + 外键校验（删除孤儿记录）
UPDATE cleaned_users SET signup_time = strftime('%Y-%m-%d %H:%M:%S', signup_time);
DELETE FROM cleaned_events WHERE user_id NOT IN (SELECT user_id FROM cleaned_users);

3.4 清洗规则汇总表

4.1 五层转化漏斗建模

构建「注册 → 激活 → 核心功能 → 付费 → 7 日留存」五层转化漏斗，确保每一步用户都是上一步的子集：注册用户 ⊃ 激活用户 ⊃ 核心功能用户 ⊃ 付费用户 ⊃ 7日留存用户。

WITH funnel AS (
  SELECT
    COUNT(*) AS total_users,
    SUM(is_activate) AS activated,
    SUM(CASE WHEN is_activate = 1 AND core_used = 1 THEN 1 ELSE 0 END) AS core_used,
    SUM(CASE WHEN is_activate = 1 AND core_used = 1 AND is_pay = 1 THEN 1 ELSE 0 END) AS paid,
    SUM(CASE WHEN is_activate = 1 AND core_used = 1 AND is_pay = 1 AND retain_7d = 1 THEN 1 ELSE 0 END) AS retained
  FROM cleaned_users
)
SELECT
  ROUND(activated  * 100.0 / total_users, 2) AS '激活率(%)',
  ROUND(core_used  * 100.0 / activated,   2) AS '激活→核心转化率(%)',
  ROUND(paid       * 100.0 / core_used,   2) AS '核心→付费转化率(%)',
  ROUND(retained   * 100.0 / paid,        2) AS '付费7日留存率(%)'
FROM funnel;

4.2 新手引导完成度与流失相关性分析

验证假设：「新手引导完成度越高，用户流失率越低」。通过 SQL CASE WHEN 分箱对比量化引导完成度对留存的影响。

4.3 用户分层分析（S / A / B / C / D 五层）

依据核心功能使用频次（从 events 表统计 use_core_feature 次数）与付费状态，将用户划分为五层，为精细化运营提供分层依据。

SELECT
  CASE
    WHEN core_used = 0                           THEN 'D-冷启动'
    WHEN core_used = 1 AND is_pay = 1 AND freq >= 10 THEN 'S-高价值'
    WHEN core_used = 1 AND is_pay = 1 AND freq >= 3  THEN 'A-稳定付费'
    WHEN core_used = 1 AND is_pay = 0 AND freq >= 10 THEN 'B-高潜力'
    WHEN core_used = 1 AND is_pay = 0 AND freq >= 1  THEN 'C-待激活'
    ELSE 'D-冷启动'
  END AS user_tier,
  COUNT(*) AS users,
  ROUND(SUM(is_pay) * 100.0 / COUNT(*), 2) AS pay_rate,
  ROUND(SUM(retain_7d) * 100.0 / COUNT(*), 2) AS retain_rate
FROM (
  SELECT u.*, IFNULL(e.freq, 0) AS freq
  FROM cleaned_users u
  LEFT JOIN (
    SELECT user_id, COUNT(*) AS freq FROM cleaned_events
    WHERE event_name = 'use_core_feature' GROUP BY user_id
  ) e ON u.user_id = e.user_id
) t GROUP BY user_tier ORDER BY pay_rate DESC;

4.4 渠道质量分析

按渠道使用 SQL GROUP BY channel 聚合对比，为渠道预算分配与投放优化提供数据支撑。

7.1 产出物清单

7.2 能力沉淀链路

附录 A：关键指标速查

附录 B：来源声明