Handling "Censored Data": Accounts That Haven't Converted Yet

In data science, there is a concept called Survival Analysis. It was originally designed (grimly) to measure how long patients survive after a diagnosis. One of its key concepts is Right-Censoring.

"Censored Data" refers to subjects who haven't experienced the event yet, but haven't dropped out either. The clock is still ticking.

The New User Problem

In SaaS metrics, this is the "New User Problem," and it destroys the accuracy of most TTV dashboards.

Imagine 100 users signed up today.

10 sprinters activate in 5 minutes.
90 marathon runners are still working on it. they will likely activate next week.

If you calculate "Average Time To Value" right now, you only have data for the 10 sprinters. Your dashboard says: "TTV: 5 minutes!"

You high-five your team. But you are hallucinating. You are ignoring the 90 people who are currently dragging the average up. Modeling only the winners biases your data toward speed.

The Lag in Truth

This creates a paradox: Recent data is the most interesting, but the least accurate. Historical data (from last year) is 100% accurate because everyone has either won or lost by now. But it's irrelevant.

How to Handle Censoring in Tivalio

To solve this, you need to define a Validity Window (or Time Threshold).

We treat a user as "valid" for the aggregate calculation only when:

Case A: They have successfully activated.
Case B: They have failed to activate within a reasonable timeframe (e.g., 30 days).

If a user is only 1 hour old, they are in limbo. They are Censored. We exclude them from the P90 calculation because their story isn't finished.

The "Global P90" Rule of Thumb

How long should you wait before calling a cohort "done"? A good rule is to look at your historical Global P90.

If historically, 90% of successful users activate within 7 days, then you should generally treat the last 7 days of data as "Volatile."

"January 1-7 TTV": Finalized.
"January 25 (Today) TTV": Pending.

Managing Executive Expectations

This is the hardest part for a Product Manager. Your CEO wants "Real Time Data." You have to explain:

"We can count the number of activations in real-time. But we cannot measure the average speed of the cohort until the cohort has matured."

If you report TTV too early, you will always report an overly optimistic number, and you will be confused when it "gets worse" over the next month as the slow users finally cross the finish line.