渠道获客 ROI 面试问答：从概念到业务决策

我会先把这个问题定义为：前端流量充足，但后端转化不足。不能马上判断渠道差，而要先确认数据是否真实，再用漏斗定位掉点。

我的排查顺序是：先确认口径和数据延迟，再拆公式判断是成本上升还是收益下降，然后拆漏斗和渠道结构，最后检查外部环境与产品承接。

这题考的不是“敢不敢反对业务”，而是能不能把冲突转成可验证的经营决策。低 ROI 渠道不一定不能投，但必须讲清楚增量收益、增量亏损、机会成本和止损线。

这题要回答“报表服务什么决策”。我会把日常数据监控拆成六类：获客日报、ROI 监控、转化漏斗、渠道质量、分层分析和异常预警。

我会按离最终经营结果的远近排序。只在题目给出的几个指标里，我会排为：成交率、客单收益、申请率、注册率、点击率。

这里的 SQL（结构化查询语言） 不只是写查询，更重要的是先定义口径：分析哪个渠道、按点击时间还是成交时间、按用户去重还是按行为次数、归因窗口多长、首次点击还是末次点击归因。

WITH channel_users AS (
    SELECT
        user_id,
        MIN(click_time) AS first_click_time,
        SUM(cost) AS total_cost
    FROM channel_click_log
    WHERE channel_id = 'target_channel'
      AND click_time >= CURRENT_DATE - INTERVAL '30 day'
      AND click_time < CURRENT_DATE
      AND user_id IS NOT NULL
    GROUP BY user_id
),
funnel AS (
    SELECT
        cu.user_id,
        cu.total_cost,
        MAX(CASE WHEN e.event_name = 'register' THEN 1 ELSE 0 END) AS is_register,
        MAX(CASE WHEN o.apply_time IS NOT NULL THEN 1 ELSE 0 END) AS is_apply,
        MAX(CASE WHEN o.approve_time IS NOT NULL THEN 1 ELSE 0 END) AS is_approve,
        MAX(CASE WHEN o.deal_time IS NOT NULL THEN 1 ELSE 0 END) AS is_deal
    FROM channel_users cu
    LEFT JOIN user_event_log e
      ON cu.user_id = e.user_id
     AND e.event_time >= cu.first_click_time
     AND e.event_time < cu.first_click_time + INTERVAL '30 day'
    LEFT JOIN loan_order o
      ON cu.user_id = o.user_id
     AND o.apply_time >= cu.first_click_time
     AND o.apply_time < cu.first_click_time + INTERVAL '30 day'
    GROUP BY cu.user_id, cu.total_cost
)
SELECT
    COUNT(*) AS click_users,
    SUM(is_register) AS register_users,
    SUM(is_apply) AS apply_users,
    SUM(is_approve) AS approve_users,
    SUM(is_deal) AS deal_users,
    ROUND(SUM(is_register) * 1.0 / NULLIF(COUNT(*), 0), 4) AS register_rate,
    ROUND(SUM(is_apply) * 1.0 / NULLIF(SUM(is_register), 0), 4) AS apply_rate,
    ROUND(SUM(is_approve) * 1.0 / NULLIF(SUM(is_apply), 0), 4) AS approve_rate,
    ROUND(SUM(is_deal) * 1.0 / NULLIF(SUM(is_approve), 0), 4) AS deal_rate,
    SUM(total_cost) AS total_cost,
    ROUND(SUM(total_cost) * 1.0 / NULLIF(SUM(is_deal), 0), 2) AS cost_per_deal
FROM funnel;

Python（脚本分析语言） 在这个岗位里的价值，是把 SQL 查出的基础数据继续做批量清洗、指标加工、自动化报表和异常预警。它不需要复杂炫技，重点是稳定、可复用、能节省人工。

import pandas as pd

df = pd.read_csv("channel_daily.csv")

df["channel"] = df["channel"].map(channel_name_map).fillna(df["channel"])
df["stat_date"] = pd.to_datetime(df["stat_date"])
df = df.drop_duplicates(["stat_date", "channel", "user_id"])

summary = (
    df.groupby(["stat_date", "channel"], as_index=False)
      .agg(
          cost=("cost", "sum"),
          click_users=("user_id", "nunique"),
          apply_users=("is_apply", "sum"),
          deal_users=("is_deal", "sum"),
          net_revenue=("net_revenue", "sum"),
      )
)

summary["roi"] = summary["net_revenue"] / summary["cost"].replace(0, pd.NA)
summary["cost_per_deal"] = summary["cost"] / summary["deal_users"].replace(0, pd.NA)
summary["roi_7d_avg"] = summary.groupby("channel")["roi"].transform(
    lambda s: s.rolling(7, min_periods=3).mean()
)
summary["is_roi_alert"] = summary["roi"] < summary["roi_7d_avg"] * 0.7

这题考对账能力和合作沟通能力。渠道数据争议通常不是简单的“谁错了”，而是统计口径、归因规则、去重规则、时间窗口和有效性规则不同。

这题最适合体现经营视角。短期亏损不一定差，关键看长期回收能力、用户质量、风险成本和新增预算的边际 ROI。