5D Library › Order-Bump Friction Elasticity, One-Click Upsell Kinetics and Cross-Domain Recommendation Matrices · 中文

Chapter 2. Second 1 and Day 7: the Two Peaks of the Upsell

Decaying interest and full credit for what was paid

After payment succeeds, a buyer’s interest in “going further” decays with time. The program writes acceptance as a(t)=a0e−t/τ+a∞a(t) = a_0 e^{-t/\tau} + a_\infty (illustrative a0=0.10a_0 = 0.10, τ=2\tau = 2 days, a∞=0.03a_\infty = 0.03); a one-click upsell in the first second needs no card re-entry, while an invitation on day 7 carries a friction factor of 0.8. The upgrade is an annual pass at RM 398, with the RM 98 already paid credited in full, so the difference is RM 300.

Calculation

# Chapter 2. p03: One-click upsell at second 1 versus an offer on day 7.
# A buyer of a RM 98 report is offered the annual pass (RM 398) with the
# RM 98 already paid credited, so the step-up price is RM 300. Interest
# in going further decays after purchase: acceptance(t) = a0 exp(-t/tau)
# + a_inf; one-click at t = 0 needs no card re-entry, a later offer has
# friction (a factor f < 1). A buyer who declined at t = 0 accepts later
# with a reduced chance (factor g). Illustrative parameters below.
from math import exp

A0, AINF, TAU = 0.10, 0.03, 2.0      # tau in days

def acc(t, friction=1.0):
    return (A0 * exp(-t / TAU) + AINF) * friction

if __name__ == "__main__":
    step = 300.0
    now = acc(0)
    day7 = acc(7, friction=0.8)
    print("   acceptance: second 1 {:.4f}; day 7 {:.4f}".format(now, day7))
    for g in (1.0, 0.7, 0.4):
        both = now + (1 - now) * day7 * g
        print("   decline at second 1 dampens day 7 by {:.1f}:"
              " both offers {:.4f}".format(g, both))
        print("     upgrade revenue per 100 buyers: second 1 only RM {:,.0f},"
              .format(100 * step * now))
        print("     day 7 only RM {:,.0f}, both RM {:,.0f}".format(
            100 * step * day7, 100 * step * both))
    print("   without the RM 98 credit (step RM 398, acceptance x 0.75):")
    print("     second 1: RM {:,.0f} per 100 buyers".format(
        100 * 398 * now * 0.75))
#    acceptance: second 1 0.1300; day 7 0.0264
#    decline at second 1 dampens day 7 by 1.0: both offers 0.1530
#      upgrade revenue per 100 buyers: second 1 only RM 3,900,
#      day 7 only RM 792, both RM 4,589
#    decline at second 1 dampens day 7 by 0.7: both offers 0.1461
#      upgrade revenue per 100 buyers: second 1 only RM 3,900,
#      day 7 only RM 792, both RM 4,383
#    decline at second 1 dampens day 7 by 0.4: both offers 0.1392
#      upgrade revenue per 100 buyers: second 1 only RM 3,900,
#      day 7 only RM 792, both RM 4,176
#    without the RM 98 credit (step RM 398, acceptance x 0.75):
#      second 1: RM 3,880 per 100 buyers

Acceptance is 0.1300 in the first second and 0.0264 on day 7. Per 100 buyers, the first-second upsell brings RM 3,900 and a day-7 invitation alone RM 792. Making both offers, if a refusal in the first second does not affect day 7, gives RM 4,589 in all; if a refusal cuts day-7 acceptance to seven or four tenths, RM 4,383 or RM 4,176. The first second is the main peak and day 7 a smaller second peak; making both offers beats making one, but the second peak’s value depends on the shadow the first refusal leaves. Full credit has value in itself: without it (paying RM 398, acceptance at three quarters) 100 buyers bring only RM 3,880.

Acceptance of an upgrade decays with time after payment (p03).

Regret and refunds

Calculation

# Chapter 2. p04: Upsells and buyer's remorse - refunds in the ledger.
# An upsell accepted in the moment of purchase is more often regretted:
# refund probability r0 for upgrades taken at second 1, r7 for those taken
# on day 7 after the buyer has used the product. A refund of an upgrade
# also refunds the credited main purchase with probability q (the buyer
# asks for everything back). Net upgrade revenue per acceptance.
def net(step, main, r, q):
    return step * (1 - r) - main * r * q

if __name__ == "__main__":
    step, main = 300.0, 98.0
    for label, r in (("second 1", 0.12), ("day 7", 0.05)):
        for q in (0.0, 0.5):
            print("   {:8s} refund {:.2f}, whole-order share {:.1f}:"
                  " net RM {:6.2f}".format(label, r, q,
                                           net(step, main, r, q)))
    # acceptance rates from p03: 0.1300 at second 1, 0.0264 on day 7
    now = 0.1300 * net(step, main, 0.12, 0.5)
    later = 0.0264 * net(step, main, 0.05, 0.5)
    print("   per buyer: second 1 RM {:.2f}; day 7 RM {:.2f}".format(
        now, later))
    print("   second 1 still wins while its acceptance exceeds day 7's by"
          " {:.2f} x".format(net(step, main, 0.05, 0.5)
                             / net(step, main, 0.12, 0.5)))
#    second 1 refund 0.12, whole-order share 0.0: net RM 264.00
#    second 1 refund 0.12, whole-order share 0.5: net RM 258.12
#    day 7    refund 0.05, whole-order share 0.0: net RM 285.00
#    day 7    refund 0.05, whole-order share 0.5: net RM 282.55
#    per buyer: second 1 RM 33.56; day 7 RM 7.46
#    second 1 still wins while its acceptance exceeds day 7's by 1.09 x

Upgrades taken on impulse are regretted more often. Let upgrades taken in the first second be refunded at 0.12 and those taken on day 7 at 0.05, with half of those refunding an upgrade also refunding the main purchase. Net revenue per acceptance is RM 258.12 and RM 282.55. Per buyer, the first second yields RM 33.56 and day 7 RM 7.46: as long as first-second acceptance exceeds day 7’s by a factor of 1.09 the first second stays better, and under the illustrative parameters it is nearly five times higher.

Hypothesis · not established

The author’s hypothesis (not yet tested): for reading reports, a day-7 upsell built on features the buyer has actually used has a significantly lower refund rate than a first-second upsell. What would show it wrong: if, in a randomised comparison, upgrades offered at the two times show no significant difference in refund rates, the hypothesis fails.

Summary

Acceptance of an upsell decays quickly after payment; the first second is the main peak (0.1300 under illustrative parameters) and day 7 a second peak (0.0264). Crediting the amount paid in full makes the difference easier to accept. Counting regret refunds, the first second still wins, but day-7 upsells are refunded less and suit a second invitation.

Fan, Y. (2026). Order-Bump Friction Elasticity, One-Click Upsell Kinetics and Cross-Domain Recommendation Matrices: Zero-Reentry Impulse Attachment, Conditional Mutual Information and Cart-Value Multipliers [结账顺手加购弹性、一键升单动力学与跨领域推荐矩阵] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090207 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).