Customer lifetime value

Judge acquisition on the first order and you will systematically underpay for customers who come back and overpay for those who do not. Lifetime value is the correction: value the relationship, not the transaction.

The distinction almost nobody states. Historic CLV is arithmetic — what existing customers have actually spent to date. It is a fact, and it is backward-looking and biased toward people who have been around long enough to accumulate spend. Predicted CLV is a model estimating what a customer will be worth, and it is a forecast with error bars, increasingly produced by machine learning.

Both get called “LTV” and put in the same column. One is a measurement, the other is a prediction, and decisions that would be sound on the first can be badly wrong on the second.

Where it goes wrong in practice.

Survivorship. Averaging the value of current customers excludes everyone who already left, which inflates the figure — often severely for a young business where most cohorts have not had time to churn.

Revenue instead of margin. A high-revenue customer with heavy discounting, returns or support cost may be worth less than a quieter one. Lifetime value should be value.

Applying an average to an individual. The distribution is usually very skewed, so the average describes almost nobody. Paying the average to acquire a specific customer is a bet on them being typical when most are not.

The horizon is a choice. Value over one year and value over five are different numbers, and the longer one is more flattering and less certain. State the window, or the figure means nothing.