Ask HN: How do people get to a single number for churn? | Hacker News Reader

Ask HN: How do people get to a single number for churn?

I often see churn quoted as a single number. How do people "flatten" churn like this? In my experience churn is very high in the first month, then falls as time goes on.

My current (consumer) startup has a churn of around 25% the first month, then dropping to almost nothing after a few months. Using a single churn number doesn't even come close to approximating real behavior.

Seeing single number churns quoted as benchmarks makes them fairly useless. How do people reconcile this?

View on HN

It depends on your business, but in the end it is just an average.

If you are a saas business then you may find churn is a case of customers trying the product and then disappearing (often forever). This appears to be the situation you are describing.

With eCommerce it can be completely different, people order once and then take months to re-order. Maybe you sell gifts and people don't come back until next Christmas! For a business like this, churn is simply a matter of 'did they order again within the time period'

If you are in the first group then flattening churn to one number is totally unhelpful. What would you do with that anyway? With business metrics it is important to ask yourself why you are measuring, and what you are going to do with the output. You need to consider the likely behavior for a person trying your service.

Imagine you are 2 years into your service and you decide to make a change which may impact retention, then measuring the 1 month and 2 month drop off may be a much better metric than some 12 month average. You need to measure things that make sense in your business. Taken this way it is a bit like regression testing, did my change break the company.

Perhaps you were thinking of benchmarking churn against some market average? Unless you have fabulous market research with almost identical competitors this seems pointless.

You can use probability models to get better churn metrics. One we used in business school was the beta-geometric which gives a discrete model for heterogeneous churns (the population uses a beta distribution to describe probability of churning). http://www.brucehardie.com/papers/021/sbg_2006-05-30.pdf

You’re absolutely right to question the numbers. Without knowing where you’re seeing these single churn numbers, it’s important to also know the context and business model before being able to use it for comparing.

Reply on news.ycombinator.com