关于「三问的信息价值:题数、完成率与停问点」,《资质三问、同天连看与独家委托》给出的定量结论是什么?
结论:题数的最优点由两条曲线决定:准确率随题数递减地上升,完成率随题数几何地下降。示意参数下,每条线索的价值在三题处达到峰值 129.21,此后每题都在亏;但留存比例一变,最优题数就在二到五之间移动。误判代价决定阈值,阈值决定分层。序贯检验告诉我们,前两题一致时可以立刻停问,答案打架时补充题也推不动结论,应当改由电话面谈。
数据口径:第一句被证伪:每多问一题,完成率下降,准确率的边际增益递减;示意参数下,现金、偿债比与时间表三问使每条线索的期望净价值达到峰值 129.21,但最优题数随每题留存比例在二到五题之间移动,三问是区间结论而不是定律。 第二句只对了一半:锚定在房价判断中真实存在,诺斯克拉夫特与尼尔的数字隐含锚定权重 0.4850;诱饵效应在真实可感的选择里通常不出现,示意账本中诱饵要胜过真比较房,效应须超过 0.1200。
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: 129.21
| PI 0.20 gain 1,000 false slot 350 nurture 0.30 | |
| k completion accuracy value/complete value/lead marginal | |
| 0 | 1.0000 0.8000 60.00 60.00 |
| 1 | 0.9600 0.8000 116.00 111.36 +51.36 |
| 2 | 0.9216 0.8880 135.60 124.97 +13.61 |
| 3 | 0.8847 0.8679 146.04 129.21 +4.24 |
| 4 | 0.8493 0.8924 148.51 126.14 -3.07 |
| 5 | 0.8154 0.8918 150.57 122.77 -3.37 |
| 6 | 0.7828 0.8951 150.97 118.17 -4.60 |
| 7 | 0.7514 0.8958 151.25 113.66 -4.52 |
| best number of questions | 3 (value/lead 129.21) |
| sensitivity to the keep-rate per extra question: | |
| keep 0.99: best k | 5 value/lead 138.85 |
| keep 0.98: best k | 3 value/lead 134.65 |
| keep 0.97: best k | 3 value/lead 131.92 |
| keep 0.96: best k | 3 value/lead 129.21 |
| keep 0.94: best k | 3 value/lead 123.88 |
| keep 0.92: best k | 2 value/lead 119.76 |
查看完整验算程序
# Chapter 1. p11: expected net value of a lead form with k questions.
# Model: a fraction PI of submitted leads are buyers who can complete a
# purchase within six months. Each yes/no question has a sensitivity
# (share of real buyers answering "ready") and a specificity (share of
# others answering "not ready"); answers are conditionally independent.
# For each answer pattern the broker either offers a same-day viewing
# slot (gain GAIN from a buyer, loss COST_FP from a non-buyer) or puts
# the lead on a slow nurture track (a buyer is worth NURTURE * GAIN).
# A lead who abandons the form is worth nothing. Every extra question
# keeps only a share DROP of those still filling in the form.
# All parameters are illustrative, not measurements of any real site.
import itertools
import math
PI = 0.20 # share of genuine buyers among submissions
GAIN = 1000.0 # value units won by a same-day slot to a buyer
COST_FP = 350.0 # value units lost on a slot given to a non-buyer
NURTURE = 0.30 # share of GAIN recovered on the nurture track
C1, DROP = 0.96, 0.96 # completion with one question; keep-rate/extra
QUESTIONS = [ # (label, sensitivity, specificity)
("cash", 0.80, 0.80), ("dsr", 0.80, 0.75), ("timeline", 0.78, 0.75),
("budget", 0.65, 0.60), ("area", 0.60, 0.58), ("type", 0.58, 0.55),
("family", 0.55, 0.53)]
def completion(k, drop):
return 1.0 if k == 0 else C1 * drop ** (k - 1)
def decision_value(k):
"""Value per completed lead and decision accuracy, exact."""
vals, acc = [], []
for ans in itertools.product((1, 0), repeat=k):
p_b, p_n = PI, 1.0 - PI
for (lab, se, sp), a in zip(QUESTIONS[:k], ans):
p_b *= se if a else 1.0 - se
p_n *= 1.0 - sp if a else sp
slot = p_b * GAIN - p_n * COST_FP
slow = p_b * NURTURE * GAIN
vals.append(max(slot, slow))
acc.append(p_b if slot > slow else p_n)
return math.fsum(vals), math.fsum(acc)
def curve(drop):
return [completion(k, drop) * decision_value(k)[0] for k in range(8)]
if __name__ == "__main__":
print("PI {:.2f} gain {:,.0f} false slot {:,.0f} nurture {:.2f}"
.format(PI, GAIN, COST_FP, NURTURE))
print(" k completion accuracy value/complete value/lead marginal")
vs = curve(DROP)
for k in range(8):
v, a = decision_value(k)
mg = "" if k == 0 else "{:+8.2f}".format(vs[k] - vs[k - 1])
print("{:2d} {:10.4f} {:8.4f} {:14.2f} {:10.2f} {}".format(
k, completion(k, DROP), a, v, vs[k], mg))
print("best number of questions: {} (value/lead {:.2f})".format(
vs.index(max(vs)), max(vs)))
print("sensitivity to the keep-rate per extra question:")
for drop in (0.99, 0.98, 0.97, 0.96, 0.94, 0.92):
c = curve(drop)
print(" keep {:.2f}: best k = {} value/lead {:.2f}".format(
drop, c.index(max(c)), max(c)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《房产高意向客资质三问秒筛、多盘同天连看逼单与房东独家委托母典》第 1 章「三问的信息价值:题数、完成率与停问点」(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-355 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026355,
author = {Fan, Yafu},
title = {Three-Question Property Lead Qualification, Same-Day Multi-Viewing Closing Kinetics and Exclusive Landlord Mandates},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}