关于「归因叠加:站内外重复计数、去重规则与 ROAS 之和」,《暗投、归因与双轨抽佣》给出的定量结论是什么?

结论:合成周里,三个后台认领的订单合计是实际订单的 1.112 倍,真实增量只占实际订单的 0.312。去重把总数降到 8,556 张,消除了重复计数,却仍高于真实增量 3,549 张一倍多,末次触点给站外广告的功劳是其真实增量的 3.16 倍。换成钱,后台 ROAS 之和是 4.81,混合 ROAS 4.33,真实增量 ROAS 只有 1.35,低于盈亏线 2.884;后台说这一周赚 RM 100,176,实验说亏 RM 79,835。本书的核心命题在这个世界里被检验:后台之和不是真实回报。

数据口径:在一个所有真相都已知的合成周里,三个后台认领的订单合计是实际订单的 1.112 倍,真正由广告带来的增量只占实际订单的 0.312;后台 ROAS 之和是 4.81,真实增量 ROAS 只有 1.35,低于盈亏平衡线 2.884。 核对原文之后,本书做了蓝图修正:付费向自然流量的外溢在随机实验里有正有负,本书把它写成可正可负的净效应;沙普利值适合分功劳,不适合定预算;用户级留出组对多数东南亚品牌太大,重点改为全马十六个地区的地区实验;平台费率无法逐项核实,全部作为示意输入;马来西亚电子交易条例已由 2024 年新条例取代。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 1.35, 100176, 2.884, 4.33, 4.81

channel          spend  reported   iROAS  rep profit true profit
in-app ad       90,000      4.12    1.14      38,562     -54,511
creator link    19,847     10.00    3.58      48,962       4,773
off-site ad     40,000      3.80    0.71      12,652     -30,097
total spend 149,847; actual GMV 648,774; sum of attributed GMV 721,151
summed platform ROAS 4.81 | blended ROAS 4.33 | true incremental ROAS 1.35
all-ads increment check: 202,507 GMV from 3,559 orders
profit the dashboards imply 100,176; profit the experiment implies -79,835
break-even ROAS 2.884: summed ROAS clears it by 1.93, true ROAS misses it by 1.54
查看完整验算程序
# Chapter 4. p43: "the sum of platform-reported ROAS is the true return" - tested on the synthetic
# week of p41.  Claimed and causal orders are p41's output; spends are illustrative (RM a week).
# The creator channel costs a 0.10 commission on the GMV it claims.  m = 0.3467 (p21).
import math

NET, M = 56.90, 0.3467
ACTUAL, NO_ADS = 11402, 7843
CH = [("in-app ad", 6517, 1799, 90000.0), ("creator link", 3488, 1248, None),
      ("off-site ad", 2669, 502, 40000.0)]

rows = []
for name, claimed, causal, spend in CH:
    if spend is None:
        spend = 0.10 * claimed * NET
    rows.append((name, claimed * NET, causal * NET, spend))
print("{:<13}{:>9}{:>10}{:>8}{:>12}{:>12}".format("channel", "spend", "reported", "iROAS",
                                                 "rep profit", "true profit"))
for name, rep, inc, s in rows:
    print("{:<13}{:>9,.0f}{:>10.2f}{:>8.2f}{:>12,.0f}{:>12,.0f}".format(
        name, s, rep / s, inc / s, M * rep - s, M * inc - s))
spend = math.fsum(r[3] for r in rows)
rep_gmv = math.fsum(r[1] for r in rows)
inc_gmv = math.fsum(r[2] for r in rows)
print("total spend {:,.0f}; actual GMV {:,.0f}; sum of attributed GMV {:,.0f}".format(
    spend, ACTUAL * NET, rep_gmv))
print("summed platform ROAS {:.2f} | blended ROAS {:.2f} | true incremental ROAS {:.2f}".format(
    rep_gmv / spend, ACTUAL * NET / spend, inc_gmv / spend))
print("all-ads increment check: {:,.0f} GMV from {:,} orders".format((ACTUAL - NO_ADS) * NET,
                                                                   ACTUAL - NO_ADS))
print("profit the dashboards imply {:,.0f}; profit the experiment implies {:,.0f}".format(
    M * rep_gmv - spend, M * inc_gmv - spend))
print("break-even ROAS {:.3f}: summed ROAS clears it by {:.2f}, true ROAS misses it by {:.2f}".format(
    1.0 / M, rep_gmv / spend - 1.0 / M, 1.0 / M - inc_gmv / spend))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《东南亚社交电商暗投带货联动、站内外归因叠加与双轨抽佣增长母典》第 4 章「归因叠加:站内外重复计数、去重规则与 ROAS 之和」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-351 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026351,
  author    = {Fan, Yafu},
  title     = {Southeast Asian Social-Commerce Whitelist Amplification, Cross-Channel Attribution and Dual-Track Commission Growth},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论