顺子の杂货铺
生命不息，折腾不止，且行且珍惜~
        找回密码
当前位置：順子の杂货铺 > BigData > 正文

26-RFM分析

2026-01-18 分类：BigData / BLOG / Database / SQL 阅读(5) 评论(0)
DMIT VPS

RFM分析

RFM分析是用户价值分群的经典方法，通过Recency（最近）、Frequency（频率）、Monetary（金额）三个维度对用户进行分类。

📊 支持情况

数据库 MAX/MIN COUNT SUM NTILE WINDOW 备注
MySQL ✅ ✅ ✅ ✅ 8.0+ ✅ 完整支持
Oracle ✅ ✅ ✅ ✅ ✅ 完整支持
ClickHouse ✅ ✅ ✅ ✅ ✅ 完整支持
Hologres ✅ ✅ ✅ ✅ ✅ 完整支持
MaxCompute ✅ ✅ ✅ ✅ ✅ 完整支持
Hive ✅ ✅ ✅ ✅ ✅ 完整支持

1. RFM指标计算

基础RFM计算

-- MySQL - 计算RFM指标
WITH rfm AS (
    SELECT
        user_id,
        MAX(order_date) AS last_order_date,
        COUNT(*) AS frequency,
        SUM(amount) AS monetary
    FROM orders
    WHERE order_date >= DATE_SUB(CURDATE(), INTERVAL 1 YEAR)
    GROUP BY user_id
)
SELECT
    user_id,
    last_order_date,
    DATEDIFF(CURDATE(), last_order_date) AS recency_days,
    frequency,
    monetary,
    RANK() OVER (ORDER BY DATEDIFF(CURDATE(), last_order_date)) AS r_rank,
    RANK() OVER (ORDER BY frequency DESC) AS f_rank,
    RANK() OVER (ORDER BY monetary DESC) AS m_rank
FROM rfm;

计算RFM分值

-- MySQL - 使用NTILE分组
WITH rfm AS (
    SELECT
        user_id,
        DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
        COUNT(*) AS frequency,
        SUM(amount) AS monetary
    FROM orders
    GROUP BY user_id
),
rfm_score AS (
    SELECT
        user_id,
        recency,
        frequency,
        monetary,
        NTILE(5) OVER (ORDER BY recency DESC) AS r_score,  -- 越小越好，反向
        NTILE(5) OVER (ORDER BY frequency) AS f_score,     -- 越大越好
        NTILE(5) OVER (ORDER BY monetary) AS m_score       -- 越大越好
    FROM rfm
)
SELECT
    user_id,
    recency,
    frequency,
    monetary,
    r_score,
    f_score,
    m_score,
    CONCAT(r_score, f_score, m_score) AS rfm_combined
FROM rfm_score;

2. 用户分群

基于RFM的用户分群

-- MySQL - 用户价值分群
SELECT
    user_id,
    r_score,
    f_score,
    m_score,
    CASE
        WHEN r_score >= 4 AND f_score >= 4 AND m_score >= 4 THEN '重要价值用户'
        WHEN r_score >= 4 AND f_score < 4 AND m_score >= 4 THEN '重要发展用户'
        WHEN r_score < 4 AND f_score >= 4 AND m_score >= 4 THEN '重要保持用户'
        WHEN r_score < 4 AND f_score < 4 AND m_score >= 4 THEN '重要挽留用户'
        WHEN r_score >= 4 AND f_score >= 4 AND m_score < 4 THEN '一般价值用户'
        WHEN r_score >= 4 AND f_score < 4 AND m_score < 4 THEN '一般发展用户'
        WHEN r_score < 4 AND f_score >= 4 AND m_score < 4 THEN '一般保持用户'
        ELSE '一般挽留用户'
    END AS user_segment
FROM (
    SELECT
        user_id,
        NTILE(5) OVER (ORDER BY recency DESC) AS r_score,
        NTILE(5) OVER (ORDER BY frequency) AS f_score,
        NTILE(5) OVER (ORDER BY monetary) AS m_score
    FROM (
        SELECT
            user_id,
            DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
            COUNT(*) AS frequency,
            SUM(amount) AS monetary
        FROM orders
        GROUP BY user_id
    ) t
) t;

用户分群统计

-- MySQL - 各分群用户统计
SELECT
    user_segment,
    COUNT(*) AS user_count,
    ROUND(COUNT(*) * 100.0 / SUM(COUNT(*)) OVER (), 2) AS pct,
    ROUND(AVG(monetary), 2) AS avg_monetary,
    ROUND(AVG(frequency), 2) AS avg_frequency
FROM (
    SELECT
        user_id,
        monetary,
        frequency,
        CASE
            WHEN r_score >= 4 AND f_score >= 4 THEN '高价值'
            WHEN r_score >= 4 THEN '新用户'
            WHEN f_score >= 4 THEN '活跃用户'
            ELSE '沉默用户'
        END AS user_segment
    FROM (
        SELECT
            user_id,
            DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
            COUNT(*) AS frequency,
            SUM(amount) AS monetary,
            NTILE(5) OVER (ORDER BY DATEDIFF(CURDATE(), MAX(order_date)) DESC) AS r_score,
            NTILE(5) OVER (ORDER BY COUNT(*)) AS f_score
        FROM orders
        GROUP BY user_id
    ) t
) t
GROUP BY user_segment;

3. 实战案例

场景1：电商用户价值分层

-- MySQL
WITH rfm_base AS (
    SELECT
        user_id,
        DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
        COUNT(DISTINCT order_id) AS frequency,
        SUM(amount) AS monetary
    FROM orders
    WHERE order_date >= DATE_SUB(CURDATE(), INTERVAL 365 DAY)
    GROUP BY user_id
),
rfm_scores AS (
    SELECT
        user_id,
        recency,
        frequency,
        monetary,
        NTILE(5) OVER (ORDER BY recency DESC) AS r_score,
        NTILE(5) OVER (ORDER BY frequency DESC) AS f_score,
        NTILE(5) OVER (ORDER BY monetary DESC) AS m_score
    FROM rfm_base
)
SELECT
    r_score,
    f_score,
    COUNT(*) AS users,
    ROUND(SUM(monetary), 2) AS total_revenue,
    ROUND(AVG(monetary), 2) AS avg_revenue
FROM rfm_scores
GROUP BY r_score, f_score
ORDER BY r_score, f_score;

场景2：流失预警

-- MySQL - 识别高风险流失用户
SELECT
    user_id,
    recency,
    frequency,
    monetary,
    r_score,
    CASE
        WHEN r_score = 1 THEN '高流失风险'
        WHEN r_score = 2 THEN '较高风险'
        WHEN r_score = 3 THEN '中等风险'
        ELSE '低风险'
    END AS churn_risk,
    recommended_action
FROM (
    SELECT
        user_id,
        DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
        COUNT(*) AS frequency,
        SUM(amount) AS monetary,
        NTILE(5) OVER (ORDER BY DATEDIFF(CURDATE(), MAX(order_date))) AS r_score
    FROM orders
    GROUP BY user_id
) t
ORDER BY r_score;

场景3：用户生命周期价值

-- MySQL - CLV预测
WITH rfm AS (
    SELECT
        user_id,
        DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
        COUNT(*) AS frequency,
        SUM(amount) AS monetary
    FROM orders
    GROUP BY user_id
),
clv AS (
    SELECT
        user_id,
        monetary AS ltv,
        monetary / NULLIF(frequency, 0) AS avg_order_value,
        CASE
            WHEN recency <= 30 THEN '活跃'
            WHEN recency <= 90 THEN '沉睡'
            ELSE '流失'
        END AS status
    FROM rfm
)
SELECT
    status,
    COUNT(*) AS users,
    ROUND(AVG(ltv), 2) AS avg_ltv,
    ROUND(AVG(avg_order_value), 2) AS avg_aov
FROM clv
GROUP BY status;

场景4：RFM交叉分析

-- MySQL - RFM热力图数据
SELECT
    r_score,
    f_score,
    COUNT(*) AS user_count,
    ROUND(AVG(monetary), 2) AS avg_monetary
FROM (
    SELECT
        user_id,
        NTILE(5) OVER (ORDER BY recency DESC) AS r_score,
        NTILE(5) OVER (ORDER BY frequency DESC) AS f_score,
        SUM(amount) AS monetary
    FROM (
        SELECT
            user_id,
            DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
            SUM(amount) AS monetary
        FROM orders
        GROUP BY user_id
    ) t
) t
GROUP BY r_score, f_score
ORDER BY r_score, f_score;

4. 用户生命周期管理

用户状态标记

-- MySQL - 用户生命周期状态
WITH latest_order AS (
    SELECT
        user_id,
        MAX(order_date) AS last_order,
        COUNT(*) AS order_count,
        SUM(amount) AS total_amount
    FROM orders
    GROUP BY user_id
),
user_status AS (
    SELECT
        user_id,
        last_order,
        DATEDIFF(CURDATE(), last_order) AS days_since_last,
        order_count,
        total_amount,
        CASE
            WHEN days_since_last <= 30 THEN '活跃'
            WHEN days_since_last <= 90 THEN '沉睡'
            WHEN days_since_last <= 180 THEN '不活跃'
            ELSE '流失'
        END AS status
    FROM latest_order
)
SELECT
    status,
    COUNT(*) AS users,
    ROUND(AVG(total_amount), 2) AS avg_spent
FROM user_status
GROUP BY status;

差异化运营策略

-- MySQL - 生成运营建议
SELECT
    user_id,
    user_segment,
    CASE
        WHEN user_segment = '重要价值用户' THEN '提供VIP专属服务，保持高价值'
        WHEN user_segment = '重要发展用户' THEN '推送高价值商品，提升消费频次'
        WHEN user_segment = '重要保持用户' THEN '发送召回优惠，唤醒活跃'
        WHEN user_segment = '重要挽留用户' THEN '紧急召回，分析流失原因'
        WHEN user_segment = '一般价值用户' THEN '培养消费习惯，引导升级'
        WHEN user_segment = '一般发展用户' THEN '首单优惠，促进复购'
        WHEN user_segment = '一般保持用户' THEN '定期触达，维护关系'
        ELSE '成本较高，降低运营投入'
    END AS recommended_action
FROM (
    SELECT
        user_id,
        CASE
            WHEN r_score >= 4 AND f_score >= 4 AND m_score >= 4 THEN '重要价值用户'
            WHEN r_score >= 4 AND f_score >= 4 THEN '重要发展用户'
            WHEN r_score < 4 AND f_score >= 4 THEN '重要保持用户'
            WHEN r_score < 4 AND f_score < 4 AND m_score >= 4 THEN '重要挽留用户'
            ELSE '一般用户'
        END AS user_segment,
        r_score, f_score, m_score
    FROM (
        SELECT
            user_id,
            NTILE(5) OVER (ORDER BY DATEDIFF(CURDATE(), MAX(order_date)) DESC) AS r_score,
            NTILE(5) OVER (ORDER BY COUNT(*)) AS f_score,
            NTILE(5) OVER (ORDER BY SUM(amount)) AS m_score
        FROM orders
        GROUP BY user_id
    ) t
) t;

⚠️ 注意事项

1. 分群数量选择

-- NTILE(5) 将用户分为5组
-- 可以根据业务需求调整为3、4、10等

-- 使用百分位数
SELECT
    PERCENTILE_CONT(0.2) WITHIN GROUP (ORDER BY recency) AS p20
FROM ...

2. 时间窗口

-- 建议使用1年的数据作为分析周期
WHERE order_date >= DATE_SUB(CURDATE(), INTERVAL 1 YEAR)

-- 对于新用户，可能需要单独处理

3. 金额处理

-- 考虑使用中位数或平均值减少异常值影响
-- 使用分位数过滤极端值

📝 练习题

-- 建表
CREATE TABLE orders_rfm (
    order_id INT,
    user_id INT,
    amount DECIMAL(10,2),
    order_date DATE
);

INSERT INTO orders_rfm VALUES
(1, 101, 500, '2024-01-14'),
(2, 101, 300, '2024-01-10'),
(3, 102, 800, '2024-01-08'),
(4, 102, 200, '2024-01-05'),
(5, 103, 1000, '2023-06-01'),
(6, 103, 1200, '2023-06-15'),
(7, 104, 150, '2024-01-14');

要求：

  1. 计算每个用户的RFM指标
  2. 给每个用户的RFM打分（1-5分）
  3. 对用户进行价值分群
  4. 统计各分群的用户数和总金额
-- 参考答案（MySQL）
-- 1. RFM指标
SELECT
    user_id,
    DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
    COUNT(*) AS frequency,
    SUM(amount) AS monetary
FROM orders_rfm
GROUP BY user_id;

-- 2. RFM打分
WITH rfm AS (
    SELECT user_id, DATEDIFF(CURDATE(), MAX(order_date)) AS recency,
           COUNT(*) AS frequency, SUM(amount) AS monetary
    FROM orders_rfm GROUP BY user_id
)
SELECT
    user_id,
    recency,
    frequency,
    monetary,
    NTILE(5) OVER (ORDER BY recency DESC) AS r_score,  -- 越近越好
    NTILE(5) OVER (ORDER BY frequency) AS f_score,     -- 越频越好
    NTILE(5) OVER (ORDER BY monetary) AS m_score       -- 越贵越好
FROM rfm;

-- 3. 用户分群
SELECT
    user_id,
    r_score, f_score, m_score,
    CASE
        WHEN r_score >= 4 AND f_score >= 4 AND m_score >= 4 THEN '重要价值用户'
        WHEN r_score >= 4 AND f_score >= 4 THEN '重要发展用户'
        WHEN r_score < 4 AND f_score >= 4 THEN '重要保持用户'
        WHEN r_score < 4 AND f_score < 4 AND m_score >= 4 THEN '重要挽留用户'
        ELSE '一般用户'
    END AS user_segment
FROM (
    SELECT user_id,
           NTILE(5) OVER (ORDER BY recency DESC) AS r_score,
           NTILE(5) OVER (ORDER BY frequency) AS f_score,
           NTILE(5) OVER (ORDER BY monetary) AS m_score
    FROM (SELECT user_id FROM orders_rfm GROUP BY user_id) t
) t;

-- 4. 分群统计
SELECT
    user_segment,
    COUNT(*) AS user_count,
    SUM(monetary) AS total_amount
FROM (
    SELECT user_id, r_score, f_score, m_score,
           CASE WHEN r_score >= 4 AND f_score >= 4 AND m_score >= 4 THEN '重要价值用户'
                ELSE '一般用户' END AS user_segment
    FROM (
        SELECT user_id,
               NTILE(5) OVER (ORDER BY DATEDIFF(CURDATE(), MAX(order_date)) DESC) AS r_score,
               NTILE(5) OVER (ORDER BY COUNT(*)) AS f_score,
               NTILE(5) OVER (ORDER BY SUM(amount)) AS m_score
        FROM orders_rfm GROUP BY user_id
    ) t
) t
GROUP BY user_segment;
赞(0)
未经允许不得转载：順子の杂货铺 » 26-RFM分析
标签：
Avatar photo

shunzi

一半忧伤，一半明媚！喜欢看书，听歌，写字，打游戏，交朋友！
搬瓦工VPS

相关推荐

评论 抢沙发

Avatar photo

取消

  • 昵称 (必填)
  • 邮箱 (必填)
  • 网址
吐血推荐

New

热门文章

热门标签

sql (48)mysql (22)oracle (21)python (15)Hive (13)MaxCompute (13)Hologres (13)ClickHouse (12)rm (6)password (5)port (5)install (4)linux (4)log (4)join (3)pos (3)google (3)vi (3)tab (3)redis (3)deepin (3)htaccess (2)master (2)centos (2)kangle (2)easypanel (2)list (2)django (2)plsql (2)docker (2)

网站统计

  • 日志总数：88
  • 评论总数：55
  • 标签总数：125
  • 最后更新：2026-01-18

分享创造快乐

联系我们联系我们