What does Dunbar Layers and Customer Half-Lives conclude, in numbers, about “The Decay Half-Life of Private-Channel Customers: Why a Cohort's Half-Life Grows”?

Answer: Chapter 5 shows that differences between customers make a cohort's half-life grow with time, with survivors' conditional half-life reaching 1061.4 days, and uses the BG/NBD formula to judge whether a silent customer is still active.

Scope of the numbers: Its layer floors, half-lives, weights and stop probabilities are illustrative; all calculations are generated by its programs; it is non-individualized business guidance, and its compliance rules yield to the law in force locally. Marked boxes separate published evidence, runnable Python calculations (19 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 1061.4

cohort half-life from day 0 : 55.1 days
exponential fitted on 30 d: half-life 46.4 days
age aS(a) LomaxS(a) exp30half-life of survivors
01.0001.00055.1
300.6390.63996.5
900.3890.261179.2
1800.2560.068303.3
3650.1570.004558.3
7300.0940.0001061.4
Monte Carlo S(180), 20000 customers : 0.258
mean rate of survivors falls from 0.0200 to 0.0020 per day

Key formula

\[S(t) = \left(\frac{b}{b + t}\right)^{k},\]
Show the full program
# Chapter 5. p51: a cohort's half-life grows with age, nobody's has to
# Model: each customer goes silent at an exponential rate lam; across
# customers lam ~ Gamma(shape k = 0.8, rate b = 40 days), illustrative.
# The cohort then survives as S(t) = (b / (b + t)) ** k (a Lomax
# curve). Survivors at age a have lam ~ Gamma(k, b + a), so their
# conditional half-life is (b + a) * (2 ** (1 / k) - 1). A single
# exponential fitted to the first 30 days is shown for contrast, and a
# Monte Carlo of 20000 customers (seed 20261113) checks S(180).
import math
import random

K, B = 0.8, 40.0


def surv(t):
    return (B / (B + t)) ** K


def main():
    c = 2.0 ** (1.0 / K) - 1.0
    print("cohort half-life from day 0 : {:.1f} days".format(B * c))
    lam30 = -math.log(surv(30.0)) / 30.0
    print("exponential fitted on 30 d  : half-life {:.1f} days".format(
        math.log(2.0) / lam30))
    print("age a   S(a) Lomax  S(a) exp30  half-life of survivors")
    for a in (0, 30, 90, 180, 365, 730):
        print("{:5d} {:11.3f} {:11.3f} {:16.1f}".format(
            a, surv(a), math.exp(-lam30 * a), (B + a) * c))
    rng = random.Random(20261113)
    alive = 0
    for _ in range(20000):
        lam = rng.gammavariate(K, 1.0 / B)
        alive += 1 if rng.expovariate(lam) > 180.0 else 0
    print("Monte Carlo S(180), 20000 customers : {:.3f}".format(
        alive / 20000))
    print("mean rate of survivors falls from {:.4f} to {:.4f} per day"
          .format(K / B, K / (B + 365.0)))


main()

Single source and citation

Fan, Y. (2026). Dunbar's Layered Numbers, Relationship-Maintenance Time Budgets and the Decay Half-Life of Private-Channel Customers (Version 1.0.0), Chapter 5: The Decay Half-Life of Private-Channel Customers: Why a Cohort's Half-Life Grows. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-231 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026231,
  author    = {Fan, Yafu},
  title     = {Dunbar's Layered Numbers, Relationship-Maintenance Time Budgets and the Decay Half-Life of Private-Channel Customers},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book