CSE Roadmap

Common tier · about 3 hours · 12 parts

Part 5 of 12 · 2 min read

The heavy tail decides which price structure actually survives contact with real customers

A price list that looks fine on the average customer can still lose money on the customers who use the product the most.

Think of an all-you-can-eat buffet priced around the average diner. It works fine until the restaurant notices a small number of regulars are eating three times the average portion, every single day.

In plain words: customer usage is heavy-tailed, not uniform, so a price structure has to be checked against the whole distribution of customers — especially the heaviest users — not just against an average that a handful of customers blow straight past.

A worked example. Across 200 simulated customers with heavy-tailed usage (median 412 tasks a month, p90 2,460, max 7,150), four price structures were each tuned to the same 60 percent blended margin: per-seat pricing put 10.0 percent of customers at a negative margin; per-task pricing put only 1.5 percent at negative margin, but those customers paid very unevenly ($0.87 to $1.18 per resolved task across p10 to p90); per-resolved-task pricing put 5.5 percent at negative margin but charged every customer the same $0.99 per resolved task; a platform fee plus overage put 4.5 percent at negative margin. No structure gets every customer to a positive margin — each one just moves where the risk lands.

The key lines. From code/mini-pricing/economics.py:

# from code/mini-pricing/economics.py
def customers(n=200, seed=0):
    rng = random.Random(seed)
    out = []
    for i in range(n):
        seats = max(1, round(5 * math.exp(rng.gauss(0, 0.6))))
        tasks = max(10, round(seats * 80 * math.exp(rng.gauss(0, 1.0))))

Read it as: tasks are drawn from a lognormal distribution (exp(gauss(...))). This produces exactly the long right tail that makes a handful of customers use far more than the median. A price list checked only against the median customer never sees this tail coming.

Heavy tail

A heavy tail is a usage distribution where a few customers use many times the median amount — common enough in real products that a price structure has to be checked against it explicitly, not assumed away.

Example: the simulated p90 customer here (2,460 tasks) uses roughly six times the median customer (412 tasks) — exactly the kind of customer who can turn a comfortable flat price into a loss.