关于「首发测价落地:四类商品的手册、测价伦理与四件套蒸馏规格」,《冷启动测价与贝叶斯调价》给出的定量结论是什么?
结论:四类商品的测价走四条不同的路:诊断报告能在线上并排测;牌卡把测价搬到预售期;课程按期测、每期一个价;精品不测流量,靠问卷、谈判记录与事先存档的降价阶梯。三种「最优」会分离,转化率最高的价格常常不是收入最高的,收入最高的又常常不是毛利最高的,高成本商品尤其如此。测价伦理可以写成程序检查:分流比例、划线价的 30 天规则、事先公布的价格保护,以及每一句稀缺话术都能追溯到一个数据字段。最后一节把这些连同前六章的计算一起,交给定价试算器。
数据口径:价格展示与分群定价的法规只写一般框架,不是法律或税务意见。 书中以标注框区分已发表的证据、可运行的计算、原典与作者的假说,每条假说都写明什么结果会推翻它;原典考辨取《管子·国蓄》的轻重敛散与「准平」。
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: 30
| card deck: unit cost 31, 900 visitors/week, RM 1.20 per paid visitor, units 180 | |
| price conv rev/visitor margin/visitor break-even_conv | |
| 79 | 0.0795 6.284 3.818 0.0250 |
| 99 | 0.0638 6.315 4.337 0.0176 |
| 119 | 0.0459 5.457 4.036 0.0136 |
| 139 | 0.0304 4.219 3.278 0.0111 |
| best by conversion 79, by revenue 99, by margin 99 | |
| 30% margin gap | 2561 visitors per arm, 11.4 weeks with four arms, 562 units sold (test outsells stock) |
| 10% margin gap | 23041 visitors per arm, 102.4 weeks with four arms, 5058 units sold (test outsells stock) |
| online course: unit cost 45, 700 visitors/week, RM 2.50 per paid visitor, units 60 | |
| price conv rev/visitor margin/visitor break-even_conv | |
| 390 | 0.0259 10.115 8.947 0.0072 |
| 490 | 0.0217 10.636 9.660 0.0056 |
| 590 | 0.0172 10.120 9.348 0.0046 |
| 690 | 0.0130 8.976 8.391 0.0039 |
| best by conversion 390, by revenue 490, by margin 490 | |
| 30% margin gap | 7861 visitors per arm, 44.9 weeks with four arms, 612 units sold (test outsells stock) |
| 10% margin gap | 70747 visitors per arm, 404.3 weeks with four arms, 5504 units sold (test outsells stock) |
| diagnostic report: unit cost 4, 2500 visitors/week, RM 0.60 per paid visitor, units no cap | |
| price conv rev/visitor margin/visitor break-even_conv | |
| 29 | 0.1152 3.340 2.880 0.0240 |
| 39 | 0.0803 3.131 2.810 0.0171 |
| 49 | 0.0490 2.401 2.205 0.0133 |
| 59 | 0.0277 1.633 1.523 0.0109 |
| best by conversion 29, by revenue 29, by margin 29 | |
| 30% margin gap | 1340 visitors per arm, 2.1 weeks with four arms, 365 units sold (stock holds) |
| 10% margin gap | 12059 visitors per arm, 19.3 weeks with four arms, 3282 units sold (stock holds) |
| boutique piece: unit cost 1700, 300 visitors/week, RM 3.00 per paid visitor, units 4 |
查看完整验算程序
# Chapter 7. p70: Four launch categories, four price rungs each: which objective wins, and can traffic decide?
# All four businesses are fictional. Conversion per visitor is q(p) = a * P(WTP >= p), with willingness to
# pay lognormal (median m, log-sd s): an assumed curve, not a measured one. For each rung: conversion,
# revenue and gross margin per visitor, and break-even conversion against paid traffic (cost per visitor /
# unit margin). Then the visitors per arm needed to detect a 30% or a 10% difference in margin per visitor
# at the best rung (two-sided 5%, power 80%, normal approximation, per-visitor variance (p - c)^2 q (1 - q) in
# both arms), the weeks that takes with four arms, and the units the test itself would sell.
import math
from statistics import NormalDist
Z = NormalDist().inv_cdf(0.975) + NormalDist().inv_cdf(0.80)
MDES = (0.30, 0.10)
CATS = [
# name, rungs, unit cost, a, median WTP, log-sd, visitors/week, cost per paid visitor, units (0 = no cap)
("card deck", [79, 99, 119, 139], 31, 0.090, 120, 0.35, 900, 1.20, 180),
("online course", [390, 490, 590, 690], 45, 0.030, 640, 0.45, 700, 2.50, 60),
("diagnostic report", [29, 39, 49, 59], 4, 0.140, 42, 0.40, 2500, 0.60, 0),
("boutique piece", [2400, 2800, 3200, 3600], 1700, 0.012, 2900, 0.20, 300, 3.00, 4),
]
def q(p, a, m, s):
return a * (1.0 - NormalDist().cdf(math.log(p / m) / s))
if __name__ == "__main__":
for name, rungs, c, a, m, s, vis, cpv, units in CATS:
print("{}: unit cost {}, {} visitors/week, RM {:.2f} per paid visitor, units {}".format(
name, c, vis, cpv, units if units else "no cap"))
print(" price conv rev/visitor margin/visitor break-even_conv")
rows = []
for p in rungs:
qq = q(p, a, m, s)
rows.append((p, qq, p * qq, (p - c) * qq))
print(" {:5d} {:.4f} {:11.3f} {:14.3f} {:15.4f}".format(
p, qq, p * qq, (p - c) * qq, cpv / (p - c)))
by_conv = max(rows, key=lambda r: r[1])[0]
by_rev = max(rows, key=lambda r: r[2])[0]
b = max(rows, key=lambda r: r[3])
var = 2.0 * (b[0] - c) ** 2 * b[1] * (1.0 - b[1])
print(" best by conversion {}, by revenue {}, by margin {}".format(by_conv, by_rev, b[0]))
for mde in MDES:
n = math.ceil(Z * Z * var / (mde * b[3]) ** 2)
sold = math.fsum(n * r[1] for r in rows)
flag = "test outsells stock" if units and sold > units else "stock holds"
print(" {:.0%} margin gap: {} visitors per arm, {:.1f} weeks with four arms, "
"{:.0f} units sold ({})"
.format(mde, n, 4 * n / vis, sold, flag))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《零历史数据冷启动测价、范氏价格敏感度边界与贝叶斯自适应调价母典》第 7 章「首发测价落地:四类商品的手册、测价伦理与四件套蒸馏规格」(版本 0.1.0). 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-133 · v0.1.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026133,
author = {Fan, Yafu},
title = {Zero-History Cold-Start Price Testing, Van Westendorp Sensitivity Bounds and Bayesian Adaptive Repricing},
year = {2026},
publisher = {5D Library},
edition = {0.1.0},
note = {DOI pending}
}