Level 3 · What a customer has been worth

Lifetime value, looking back

Customer lifetime value (LTV) is the revenue (or margin) a customer brings in over their whole relationship with the business. The simplest honest version is historic LTV: take a group of customers, add up everything they have actually spent so far, and divide by how many there are. No forecasts, no assumptions.

The natural group is the acquisition cohort: everyone whose first booked order fell in the same year. Here revenue is net revenue (discounted line value minus refunds, no shipping), counted over everything the data holds, 2023 to 2025.

#   first_year  customers  net_revenue      ltv
# 0       2023        574   3564271.36  6209.53
# 1       2024        525   1401056.50  2668.68
# 2       2025        641    721955.12  1126.30

Why young cohorts always look worse

The 2023 cohort looks five times more valuable than the 2025 cohort. It isn’t, or at least this table can’t say so. A customer acquired in January 2023 had three years to keep ordering; one acquired in November 2025 had a few weeks. Historic LTV grows with cohort age, so comparing cohorts of different ages mostly measures time.

The fix is an equal window: revenue in each customer’s first 365 days after their first order. That gives $3,605.08 for 2023 and $2,142.10 for 2024. The 2025 cohort cannot be measured yet, because most of its customers have not had a full year. A real drop is still visible (2024 customers spent less in their first year), but it is about 40%, not the 57% the to-date numbers suggest.

And the first cohort is special

The 2023 number has its own bias. The data starts on 2023-01-01, so 152 of the 574 customers “acquired” in 2023 had signed up before that and were already regulars. Their first year in the data is not their first year as customers, which flatters the 2023 cohort.

Common mistakes

• Counting rows instead of customers. size counts orders. Use nunique on customer_id.
• Using gross sales. Refunds and discounts are money the store never kept. LTV in revenue terms should be net revenue, and the margin version (next level) is better still.
• One number for everyone. A single all-time LTV mixes old and young cohorts and changes every month even if nothing about the customers changes.

Interview note

“Our LTV is falling for newer cohorts, what do you check first?” The expected answer: whether the cohorts have been observed for the same length of time. Then compare an equal window, such as the first 12 months.