What does Three-Light Cut Points conclude, in numbers, about “Red Lights Beyond Price: Stop Means Stop, Vulnerable Customers Go to Staff, and the Staffing Cap”?

Answer: Without a cap the optimal cut points are those of Chapter 2, 0.400 and 0.950, with a load of 22.1 units and a loss of 295.2 yuan per hundred conversations, which here includes the staff cost of the hard-rule conversations. With a cap of 20, the shadow price is 1.96 yuan, the green threshold falls from 0.950 to 0.938, the red threshold from 0.400 to 0.253, green rises to 0.572, and the loss grows only to 297.1 yuan. With a cap of 16, the shadow price is 11.05 yuan and the red threshold falls to zero, so that apart from the hard rules no conversation is handed over in full for low confidence; the green threshold falls to 0.881 and the loss is 320.0 yuan. With a cap of 12, the shadow price is 40.14 yuan, the green threshold 0.699, green covers 0.911, and the loss is 408.9 yuan.

Scope of the numbers: Private customer service has a third action, a bot draft checked by a person, so it needs a six-cell loss matrix with two cut points; under illustrative parameters red lights below a chance of being right of 0.4000 and green from 0.9500. Parameters are illustrative, all conversations are fictional, and all numbers come from the book's programs, which read and write no file and connect to no customer-service system.

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: 0.253, 0.400, 0.572, 0.699, 0.881, 0.911

caplamp_lop_higreenamberredloadcost/100
250.000.4000.9500.5140.4700.01722.1295.2
201.960.2530.9380.5720.4220.00520.0297.1
185.250.0060.9170.6460.3540.00018.0304.1
1611.050.0000.8810.7340.2660.00016.0320.0
1420.680.0000.8210.8230.1770.00014.0350.7
1240.140.0000.6990.9110.0890.00012.0408.9
9below the hard-rule load 10: add staff, never drop the rules
Show the full program
# Chapter 6. p62: cut points when human hours are capped.
# Model (illustrative): 10 % of chats are red by hard rule (stop
# requests and vulnerable customers, p61) and are never displaced. The
# other 90 % follow the p21 loss matrix (LW 50, CA 2, M 0.8, CR 8) on
# the p11 bot. Human load counts a take-over as 1 unit and an amber
# review as KAPPA = 0.25 unit, per 100 chats. A cap on load is met by
# a shadow price lam per unit (Lagrange): red costs CR + lam, amber
# CA + KAPPA lam; lam is found by bisection. Cost excludes lam.
import math

LW, CA, M, CR, KAPPA, HARD = 50.0, 2.0, 0.8, 8.0, 0.25, 0.10


def nodes(mu=3.0, sd=1.6, n=4001):
    zs = [mu + sd * (-8.0 + 16.0 * i / (n - 1)) for i in range(n)]
    ws = [math.exp(-0.5 * ((z - mu) / sd) ** 2) for z in zs]
    tot = math.fsum(ws)
    return [(1.0 / (1.0 + math.exp(-z)), w / tot) for z, w in zip(zs, ws)]


PW = nodes()


def solve(lam):
    hi = 1 - (CA + KAPPA * lam) / (M * LW)
    lo = 1 - (CR + lam - CA - KAPPA * lam) / ((1 - M) * LW)
    sh = {0: [], 1: [], 2: []}
    cost = []
    for p, w in PW:
        k = 0 if p >= hi else (1 if p >= lo else 2)
        sh[k].append(w)
        cost.append(w * ((1 - p) * LW, CA + (1 - p) * (1 - M) * LW,
                         CR)[k])
    g, a, r = (math.fsum(sh[k]) for k in range(3))
    load = 100 * (HARD + (1 - HARD) * (r + KAPPA * a))
    money = 100 * (HARD * CR + (1 - HARD) * math.fsum(cost))
    return max(lo, 0.0), hi, g, a, r, load, money


print("  cap    lam   p_lo   p_hi  green  amber  red   load  cost/100")
for cap in (25.0, 20.0, 18.0, 16.0, 14.0, 12.0, 9.0):
    if cap < 100 * HARD:
        print("%5.0f  below the hard-rule load %.0f: add staff, never"
              " drop the rules" % (cap, 100 * HARD))
        continue
    lo_l, hi_l = 0.0, 0.0
    if solve(0.0)[5] > cap:
        hi_l = 1.0
        while solve(hi_l)[5] > cap:
            hi_l *= 2
        for _ in range(60):
            mid = (lo_l + hi_l) / 2
            if solve(mid)[5] > cap:
                lo_l = mid
            else:
                hi_l = mid
    out = solve(hi_l)
    print("%5.0f  %5.2f  %5.3f  %5.3f  %5.3f  %5.3f  %5.3f  %5.1f  %8.1f"
          % ((cap, hi_l) + out))

Single source and citation

Fan, Y. (2026). Three Lights for Handing Over between Bot and Human in Private Customer Service: Misclassification Loss Matrices, Chow's Reject Rule and Optimal Cut Points on Calibrated Confidence (Version 1.0.0), Chapter 6: Red Lights Beyond Price: Stop Means Stop, Vulnerable Customers Go to Staff, and the Staffing Cap. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-266 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026266,
  author    = {Fan, Yafu},
  title     = {Three Lights for Handing Over between Bot and Human in Private Customer Service: Misclassification Loss Matrices, Chow's Reject Rule and Optimal Cut Points on Calibrated Confidence},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book