Journal · 9 December 2025
Reading a survival curve without a data-science degree
A survival curve in Customer Lifetime Value Analytics is a picture of how long relationships last inside a window you chose. It is not a prophecy. Operators get into trouble when they read the right-hand tail as if the future had already arrived.
Censoring, said plainly
Some customers in your extract simply have not had enough calendar time to churn or to buy again. They are still “alive” because the clock stopped, not because they love you. If you treat them as proven loyalists, the curve stays artificially high. Mark them. Any decent sheet can hold a “last seen date” and an “extract end date.”
The window is a decision
Twelve months is not morally better than six. A meal kit with weekly delivery can learn something in six. A furniture retailer pretending to see a five-year lifetime in eighteen months of data is performing. In Metric Cloudhub classrooms we ask: what decision will this curve change this quarter? Then we pick a window that can actually inform that decision.
Where to stop drawing
When the number of remaining buyers in a cohort is small, the line becomes gossip. We use a simple classroom rule: stop interpreting once fewer than a few dozen buyers remain, or once you leave the window you declared. Do not extend a Thai holiday-affected curve into next Songkran and call it science.
If notation still feels hostile, sit the “Survival without a PhD” module in the flagship programme. Bring a printed curve. We will write on it.