Reading Japan’s SaaS churn benchmark: check the distribution

Published Jul 24, 2026 Updated Oct 17, 2026

If you are setting a churn target for your Japan operation from a benchmark that says “Japanese SaaS churns at about 3% a month,” you are likely setting it against the wrong number.

That 3% is a mean, and it is pulled up by a small number of high-churn companies.

Nearly half the companies in the same dataset sit at 1% or lower. Read the mean as the market, and your Japan team looks like it is failing when it is roughly at par.

The number that circulates

The figure comes from CloudCircus’s Fullstar “Customer Success Survey 2025” (fielded August 2025, n=200, customer success and support staff at B2B information and communications companies in Japan; note that this is a vendor survey).

The headline is a mean monthly churn of 3.01%.

But the same survey publishes the distribution, and it is heavily skewed: 45% of companies report 1% or lower, while about 9% report 10% or higher.

Those few high-churn companies drag the average up.

The typical company sits well below the mean, at a third of it or less.

So “Japan churns at 3%” describes almost none of the companies in the data. It describes the gap between the typical company and a handful of struggling ones.

Why this matters when you report to HQ

Two failures follow from using the mean as if it were typical.

First, you set the wrong target.

If HQ benchmarks your Japan team against a 3% “market rate,” a team at 1% looks mediocre, and a team genuinely at 3% looks acceptable when roughly half the market sits at a third of that or less.

The averaged number hides both the good and the bad.

Second, you lose HQ’s trust in the whole report.

The moment a headline number stops matching what people see on the ground, they stop trusting the dashboard it came from.

After that, even your accurate numbers get discounted in decisions.

A dashboard exists to surface what needs intervention, not to produce a tidy single figure; a distrusted figure cannot do that job.

What to report instead

Split your own churn rate; do not look for a median in it. Your monthly churn is churned customers divided by customers at the start of the month: a single ratio, so it has no median. Report it split two ways instead: by logo and by revenue, and large accounts separately from the rest. How to lay those side by side is in Japan’s NRR looks healthy on your dashboard. Ask for the logo churn.. Use median and distribution only for the industry benchmark, and for metrics that have one value per record, such as deal size or days in stage.
Show large churned accounts as a separate list, not blended in. They are exactly what your customer success team should act on. Folding them into one rate hides the intervention target.
Name the source, year, and sample every time. When you cite the Japan benchmark, say “Fullstar 2025, n=200, vendor survey.” HQ can weigh a vendor survey of 200 companies appropriately, and a sourced number survives scrutiny that an unsourced one does not.
Do not compare Japan and home-market retention as if the numbers are equivalent. US benchmarks such as Benchmarkit 2025 report a median net revenue retention around 101%, and Japan’s Fullstar reports a mean NRR of 102.1%, but the samples, methods, and mean-versus-median basis all differ. They look close; they are not directly comparable. Present each with its own definition and let HQ draw the line.

Check the population, denominator and distribution before using a churn benchmark. Read the industry figure by its distribution; your own churn rate is a single ratio, so split it by logo and revenue, and large accounts from the rest, instead of looking for a median.

Common mistakes

Treating every metric as skewed. Symmetric metrics are fine to average. Deal size and stage duration are the ones that reliably skew; check the distribution per metric. Churn and conversion rates are ratios, so split them rather than taking a median.
Reporting the median but hiding the distribution. The median is a representative value, not a tool for making the outliers disappear.
Trusting the dashboard’s default. Most tools aggregate with a mean by default. For skewed metrics, do not ship that default to HQ.

Reading the median is the first half; attaching a decision rule to it is the second. See write the threshold before adding another dashboard.

Other notes on the same problem are collected under Revenue data.