关于「暗投放大:付费向自然流量的外溢、大促混淆与边际回报」,《暗投、归因与双轨抽佣》给出的定量结论是什么?

结论:暗投的净外溢 \(\rho\) 由截流与外溢两股力量决定,可正可负。示意模型里,截流三成且无外溢时真实回报只有 2.240,低于盈亏线;中等外溢时回到 3.508;截流小而外溢强时高达 5.416。大促期间用前后对比测外溢,在没有外溢的世界里读出 +0.037,在真实 \(\rho\) 为 +0.102 的世界里读出 +0.340;双重差分分别读出 -0.201 与 +0.104,接近真值。饱和曲线上,平均 ROAS 过线时边际 ROAS 可能早已跌破;最佳投放额是每天 RM 1,273,而按平均 ROAS 加预算会一路加到 RM 2,806。

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: -0.201, 0.037, 0.102, 0.104, 0.340

g0.0: true rho -0.200 | before-after +0.037 | diff-in-diff -0.201 (sd 0.059)
share of campaigns where before-after reads rho > 0.20.000
g0.4: true rho +0.102 | before-after +0.340 | diff-in-diff +0.104 (sd 0.062)
share of campaigns where before-after reads rho > 0.20.992
查看完整验算程序
# Chapter 3. p32: why a before-and-after reading of "organic lift" is confounded by a mega-sale.
# One treated SKU is whitelisted for 21 days (days 28..48); a mega-sale lifts organic demand of
# EVERY SKU by 1.6 on days 35..41.  Eight untreated sibling SKUs serve as controls.  Cannibalisation
# c = 0.2.  Spillover g is either 0 (none) or 0.4.  2,000 simulated campaigns, fixed seed.
import math
import random

rng = random.Random(351)
PAID, LAM, BASE, DAYS, START, STOP = 33.7, 0.90, 40.0, 49, 28, 49
C, N_CTRL, RUNS = 0.2, 8, 2000


def demand(d):
    return BASE * (1.6 if 35 <= d <= 41 else 1.0)


def mean_path(g, treated):
    a, out = 0.0, []
    for d in range(DAYS):
        paid = PAID if (treated and START <= d < STOP) else 0.0
        a = LAM * a + (1.0 - LAM) * paid / BASE
        out.append(demand(d) * (1.0 + g * a) - C * paid)
    return out


def draw(mu):
    return max(0.0, round(rng.gauss(mu, math.sqrt(max(mu, 1.0)))))


def window(xs, lo, hi):
    return math.fsum(xs[lo:hi]) / (hi - lo)


paid_n = PAID * (STOP - START)
for g in (0.0, 0.4):
    t_mu, c_mu = mean_path(g, True), mean_path(g, False)
    true_rho = math.fsum(t_mu[d] - c_mu[d] for d in range(START, STOP)) / paid_n
    naive, did = [], []
    for _ in range(RUNS):
        t = [draw(m) for m in t_mu]
        ctrl = [[draw(m) for m in c_mu] for _ in range(N_CTRL)]
        cbar = [math.fsum(row[d] for row in ctrl) / N_CTRL for d in range(DAYS)]
        dt = window(t, START, STOP) - window(t, 0, START)
        dc = window(cbar, START, STOP) - window(cbar, 0, START)
        naive.append(dt * (STOP - START) / paid_n)
        did.append((dt - dc) * (STOP - START) / paid_n)
    mn, md = math.fsum(naive) / RUNS, math.fsum(did) / RUNS
    sd = math.sqrt(math.fsum((x - md) ** 2 for x in did) / (RUNS - 1))
    hi_share = sum(1 for x in naive if x > 0.2) / RUNS
    print("g = {:.1f}: true rho {:+.3f} | before-after {:+.3f} | diff-in-diff {:+.3f} (sd {:.3f})".format(
        g, true_rho, mn, md, sd))
    print("        share of campaigns where before-after reads rho > 0.2: {:.3f}".format(hi_share))

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 3 章「暗投放大:付费向自然流量的外溢、大促混淆与边际回报」(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