关于「一百条素材能找出几条赢家:样本量、错误发现率与广告回报的信噪比」,《素材疲劳与正交混剪》给出的定量结论是什么?
结论:第五章计算样本量与错误发现率:测出一成的点击率提升,每条素材需要 163,095 次曝光。
数据口径:买量圈流行的说法是:扩量期素材衰退得快,因为观众重复曝光使点击率按可估计的半衰期衰减;把短视频拆成开头钩子、痛点、证据、行动召唤四个模块,用正交表组合,十几个镜头就能自动合成上百条低相关的测试素材,突破产能瓶颈。 书中参数为示意,全部计算由书中程序生成。
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: 163095
| relative lift | CTR a=0.05 | CTR a=0.0005 | buy a=0.05 |
| 0.05 | 637,010 | 1,516,297 | 12,865,565 |
| 0.10 | 163,095 | 388,221 | 3,294,798 |
| 0.20 | 42,693 | 101,625 | 862,901 |
| 0.30 | 19,827 | 47,195 | 400,934 |
| 100 creatives + control at lift 0.10, CTR: 16,472,595 impressions | |||
| same on purchases with Bonferroni: 792,114,417 impressions |
查看完整验算程序
# Chapter 5. p51: impressions per creative needed to see a CTR lift
# Two-sided two-proportion z test, power 0.80. Baselines are
# illustrative: CTR 0.010 per impression, purchase rate 0.0005 per
# impression. Per-test alpha 0.05, and 0.05 / 100 when a hundred
# creatives are each compared with the control (Bonferroni bound).
import math
import statistics
Z = statistics.NormalDist()
def n_per_arm(p1, lift, alpha, power=0.80):
p2 = p1 * (1.0 + lift)
pb = (p1 + p2) / 2.0
za = Z.inv_cdf(1.0 - alpha / 2.0)
zb = Z.inv_cdf(power)
num = za * math.sqrt(2 * pb * (1 - pb)) + \
zb * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))
return math.ceil(num ** 2 / (p2 - p1) ** 2)
print("relative lift CTR a=0.05 CTR a=0.0005 buy a=0.05")
for lift in (0.05, 0.10, 0.20, 0.30):
print("{:>13.2f} {:>11,} {:>13,} {:>11,}".format(
lift, n_per_arm(0.010, lift, 0.05), n_per_arm(0.010, lift, 0.0005),
n_per_arm(0.0005, lift, 0.05)))
n10 = n_per_arm(0.010, 0.10, 0.05)
print("100 creatives + control at lift 0.10, CTR: {:,} impressions".format(
101 * n10))
nb = n_per_arm(0.0005, 0.10, 0.0005)
print("same on purchases with Bonferroni: {:,} impressions".format(
101 * nb))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《投放素材频次疲劳半衰期、分镜基因正交重组与百条广告全自动混剪母典》第 5 章「一百条素材能找出几条赢家:样本量、错误发现率与广告回报的信噪比」(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-400 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026400,
author = {Fan, Yafu},
title = {Ad-Creative Fatigue Half-Lives, Orthogonal Storyboard Recombination and Automated Permutation Rendering},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}