CommerceLens
Random Variables & Uncertainty

Campaign Sampling Without Replacement

Model Simulation

A campaign pool contains N customers, K of them high-value. We randomly select n customers without replacement for a limited-slot campaign. What is the probability that exactly k of the selected customers are high-value?

Parameters
Defaults: pool of 500 customers, 72 high-value, sampling 40. K/N is seeded from the dataset’s actual observed high-value customer share, 14.5%.
Pool size (N)500
High-value customers in pool (K)72 (14.4% of pool)
Sample size (n)40
Target high-value selected (k)6

Live-Computed Probabilities

P(X = 6), exact
18.07%
Expected high-value in sample
5.76
SD(X)
2.13
Hypergeometric formulas
P(X = k) = C(K, k) · C(N−K, n−k) / C(N, n) E[X] = n·(K/N) Var(X) = n·(K/N)·((N−K)/N)·((N−n)/(N−1))
PMF of X = high-value customers selected
Sampling n = 40 without replacement from a pool of 500

Hypergeometric vs. the Naive Binomial Approximation

Hypergeometric: P(X = 6)
18.069%
Binomial approx (p = K/N): P(X = 6)
17.315%
Absolute gap
0.753%

As a sanity check: at this page’s default parameters (N = 500, K = 72, n = 40, k = 6) the precomputed reference values are P(X = k) = 0.1807 (hypergeometric, exact) vs. 0.1732 (binomial approximation) — matching what the live calculator above shows when the sliders are reset to those values.

The binomial approximation assumes sampling with replacement — a constant probability of drawing a high-value customer on every single draw. Sampling without replacement, as a real campaign selection does, is different: every customer selected changes the composition of the remaining pool, so the probability of the next draw being high-value shifts slightly after each pick. The hypergeometric distribution accounts for this; the binomial does not. The gap between the two shrinks as N grows large relative to n (currently n is 8.0% of the pool) — with a huge pool and a small sample, removing one customer barely changes the remaining odds, and the binomial approximation becomes reasonable.