Customer Analytics

Not all customers are equal. In almost every business, a small group delivers a large share of the value. Find them, keep them happy, and you've done more for the business than any fancy model could.

• Not all customers are equal
• A few deliver most of the value
• Find them and keep them

Value tiers

We estimate each customer's monthly value, as visits times spend, then split them into four equal tiers. Run it. Look at the share column. The top quarter of customers bring in about half of the money. And look at their churn. They're the least likely to leave. Even so, losing one customer from the top tier hurts far more than losing one from the bottom. So where should the loyalty budget go? Now you can answer.

import pandas as pd

df = pd.read_csv("customers.csv").drop_duplicates()

df["value"] = df["visits_per_month"] * df["avg_spend"]
df["tier"] = pd.qcut(df["value"], 4, labels=["Low", "Mid", "High", "Top"])

s = df.groupby("tier", observed=True).agg(customers=("value", "size"),
                                          monthly=("value", "sum"), churn=("churned", "mean"))
s["share"] = (s["monthly"] / s["monthly"].sum()).round(3)
print(s.round(2).to_string())

      customers    monthly  churn  share
tier
Low        1500   19486.23   0.38   0.06
Mid        1500   44509.27   0.33   0.15
High       1500   76442.46   0.24   0.25
Top        1500  161155.27   0.14   0.53

Cohort analysis

A cohort is a group that started at the same time. Comparing cohorts removes a lot of confusion. We compare churn by signup year, with and without a loyalty card. Run it. The card holders leave less in all three years, so this pattern isn't a one off. If a metric gets better overall, cohorts can tell you if that's a real improvement, or just a change in the customer mix.

import pandas as pd

df = pd.read_csv("customers.csv", parse_dates=["signup"]).drop_duplicates()
df["cohort"] = df["signup"].dt.year

table = df.groupby(["cohort", "loyalty"])["churned"].mean().unstack().round(3)
table.columns = ["No card", "Card"]
print(table.to_string())

        No card   Card
cohort
2023      0.323  0.195
2024      0.291  0.201
2025      0.315  0.212

What is a customer worth?

Customer lifetime value turns all of this into one number. Monthly profit, divided by churn. Run it. See how sensitive it is. Cutting churn from six percent to four raises each customer's value by half. So keeping customers is often worth more than winning new ones. Now the founder knows how much to spend on winning a customer.

monthly_value, margin, churn = 40.0, 0.68, 0.06

lifetime = 1 / churn
ltv = monthly_value * margin * lifetime
print(f"Expected lifetime: {lifetime:.1f} months")
print(f"Customer value: ${ltv:,.0f}")

for c in [0.06, 0.05, 0.04]:
    print(f"Churn {c:.0%}: value ${monthly_value * margin / c:,.0f}")

Expected lifetime: 16.7 months
Customer value: $453
Churn 6%: value $453
Churn 5%: value $544
Churn 4%: value $680

Summary