data.day
churn / metrics / saas / analytics

“Churn Rate” Without a Definition Is Just Drama

Is your churn 5% or 15%? It depends on who is holding the calculator. We stop the metric wars by defining the denominator and segmenting the exit door.

Sofia Vector Sofia Vector

The Tower of Babel

Look at this line chart. The red line zig-zags wildly.

In January, Churn is 5%. In February, it spikes to 14%. In March, it drops to 2%. The CEO asks, “What happened in February?” The VP of Sales says, “Oh, that wasn’t real churn. Those were just the holiday trials expiring. We shouldn’t count them.”

This is not analysis. This is improvisation.

The Noise: When we do not agree on the definition of “The Customer,” every chart becomes a battlefield. If you include free trials in your denominator, your churn looks massive. If you exclude everyone who has been with you less than 90 days, your churn looks artificially low. We are changing the ruler to measure the room.

The Pattern: We must strip the emotion out of the math. We need a Data Dictionary. And we need to split the “Churn” into its actual behaviors.

Defining the Exit

We are going to create three distinct rows in our tracking sheet. We will never mix them.

  1. Logo Churn (The Popularity Contest): Count of paying customers who cancelled / Total paying customers. This tells us if our product is likable.
  2. Revenue Churn (The Wallet Check): Dollar value lost / Total dollar value. This tells us if we are losing the Whales or just the Barnacles.
  3. Net Revenue Retention (The Holy Grail): (Revenue from existing customers + Upsells - Churn) / Starting Revenue.

[TO EDITOR: Visual Cue. A Venn Diagram or Flowchart. Show a bucket of “All Users”. Filter 1: “Paid > $0”. Filter 2: “Active > 30 Days”. Label the resulting group ” The Denominator”. Show how “Trial Users” falling out before this stage creates noise if counted.]

The Cohort Layer Cake

But the average churn rate still hides the truth. Come, look at this Cohort Analysis.

If we mix new customers with old customers, the average lies. New customers always churn faster (they are testing you). Old customers churn slower (they are habituated).

When we stack them in a “Layer Cake” chart, we see the pattern clearly.

  • Month 1 Cohort: 20% drop off immediately. (This is an onboarding failure).
  • Month 12 Cohort: 0.5% drop off. (This is a product success).

If we just said “Average Churn is 4%,” we would miss the fact that our onboarding process is broken. We would try to fix the product for the veterans, when we should be fixing the welcome email for the rookies.

Define the term. Segment the age. Only then does the drama end and the work begin.

FAQs

What is the 'right' way to calculate churn?

There isn't one. There is only the 'agreed' way. You must decide: Logo Churn (count) or Revenue Churn (dollars)?

Why do different teams get different numbers?

Because they select different denominators. One includes trial users; the other only counts paid annual contracts.

What is the most dangerous type of churn?

The 'Downgrade.' The customer is still there, but they have stopped paying for value. It is invisible in Logo Churn.