关于「首响时延的衰减曲线:先分层,再加人」,《点击直达对话的成交与归因》给出的定量结论是什么?
结论:已发表的研究给了两个比值:一小时内联系,资格率接近一小时后的 7 倍,是一天后的 60 多倍。拟合成带下限的指数衰减后,时间常数约 29.2 分钟,前半小时掉得最快,过了两小时贴着下限走;时间常数随格点假设在 29.2 到 51.1 分钟之间浮动。排队模拟说明,在对话量超过人手时,热的先答比加人更先起作用:一名接待员改排法,每天成交从 4.10 笔升到 12.44 笔。
数据口径:书中的转化率、时延、回复率、样本量与金额均为示意参数,全部计算由书中程序生成;本书是非个人化经营建议,不构成法律或营销合规意见,不承诺任何收入;只写有同意记录的触达,语音条只由真人本人录制,收款只用持牌支付机构的公开功能,紧迫感只来自真实库存与真实截止,推荐奖励只给直接推荐人一层。 第一、二章推导摩擦损耗减半的条件:示意参数下直跳后首句发出率须达 0.650;直跳让对话数涨到 2.500 倍,高意向占比从 0.571 跌到 0.327,每段接待成本 4.94 单位时两条路径打平。
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: 29.2, 51.1, 60
| audit firms 2,241, within 1 h 0.37, never 0.23, mean 42 h | ||
| fit with t1 = 0.5 h, t2 = 1.5 h: tau = 29.2 min | ||
| minutes | rel. to 24 h | rel. to 5 min |
| 5 | 139.8 | 1.000 |
| 15 | 99.6 | 0.712 |
| 30 | 60.0 | 0.429 |
| 60 | 22.1 | 0.158 |
| 120 | 3.7 | 0.027 |
| 240 | 1.0 | 0.007 |
| sensitivity of tau to the bin midpoints (minutes) | ||
| t1 = 0.25 h | t2 = 1.50 h | tau = 36.5 |
| t1 = 0.25 h | t2 = 2.00 h | tau = 51.1 |
| t1 = 0.50 h | t2 = 1.50 h | tau = 29.2 |
| t1 = 0.50 h | t2 = 2.00 h | tau = 43.8 |
查看完整验算程序
# Chapter 2. p21: a decay-plus-floor curve fitted to two published ratios.
# Oldroyd, McElheran and Elkington (2011) audited 2,241 US firms: 0.37
# answered a web lead within an hour, 0.24 took over 24 hours, 0.23 never
# answered; mean time among responders 42 hours. Firms trying within an
# hour were nearly 7 times as likely to qualify the lead as firms trying
# an hour later, and over 60 times as likely as firms waiting 24 hours.
# Model: q(t) = a * exp(-t / tau) + 1 (floor = 24-hour level). The bin
# midpoints t1 and t2 are this book's assumption, not the article's.
import math
AUDIT = {"firms": 2241, "within_1h": 0.37, "over_24h": 0.24,
"never": 0.23, "mean_hours": 42}
R_HOUR, R_DAY, T_DAY = 7.0, 60.0, 24.0
def fit(t1, t2):
def a_of(tau):
return (R_DAY - 1) / (math.exp(-t1 / tau)
- R_DAY * math.exp(-T_DAY / tau))
def gap(tau):
a = a_of(tau)
q1 = a * math.exp(-t1 / tau) + 1
q2 = a * math.exp(-t2 / tau) + 1
return q1 / q2 - R_HOUR
lo, hi = 0.05, 5.0
for _ in range(200):
mid = (lo + hi) / 2
if gap(mid) > 0:
lo = mid
else:
hi = mid
tau = (lo + hi) / 2
return tau, a_of(tau)
print("audit firms %s, within 1 h %.2f, never %.2f, mean %d h" % (
format(AUDIT["firms"], ","), AUDIT["within_1h"], AUDIT["never"],
AUDIT["mean_hours"]))
tau, a = fit(0.5, 1.5)
print("fit with t1 = 0.5 h, t2 = 1.5 h: tau = %.1f min" % (tau * 60))
q = lambda t: a * math.exp(-t / tau) + 1
print("minutes rel. to 24 h rel. to 5 min")
for mins in (5, 15, 30, 60, 120, 240):
t = mins / 60
print("%7d %13.1f %14.3f" % (mins, q(t), q(t) / q(5 / 60)))
print("sensitivity of tau to the bin midpoints (minutes)")
for t1 in (0.25, 0.5):
for t2 in (1.5, 2.0):
print(" t1 = %.2f h t2 = %.2f h tau = %.1f" % (
t1, t2, fit(t1, t2)[0] * 60))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《东南亚即时通讯成交漏斗、点击直达对话归因与语音权威收口母典》第 2 章「首响时延的衰减曲线:先分层,再加人」(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-366 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026366,
author = {Fan, Yafu},
title = {Click-to-Messenger Conversion Funnels, Chat Attribution and Voice-Note Closing in Southeast Asian Commerce},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}