The Two Metrics That Actually Predict B2B Churn

I have audited churn at more than a dozen B2B SaaS companies in the last three years. Every one of them had a customer health score. Almost none of those scores actually predicted churn earlier than a much simpler pair of signals.

Signal one: weekly active admins

The single best predictor of B2B churn I have ever found is the number of distinct admin users who logged in during the trailing four weeks. Not total active users. Not seats. Admins.

The logic is straightforward. Admins are the people who can renew, expand, or cancel. If they have stopped logging in, the account is being run on autopilot, and autopilot is one reorg away from cancellation. When weekly active admins drops by half, the renewal is in trouble even if usage from end users looks fine.

Most CS teams I work with do not track this metric at all. They track 'active users' as a single number and miss the signal entirely.

Signal two: time since the last new use case

The second signal is harder to instrument but more powerful. It is the time since the customer last adopted a meaningfully new use case in your product.

Accounts that keep finding new reasons to use you have low churn. Accounts that have settled into a stable, narrow usage pattern have high churn, even when that pattern looks healthy on the surface. The reason is simple: a customer using one feature is one budget cycle away from a cheaper alternative. A customer using six features is too embedded to move.

You can approximate this with feature adoption breadth over time. The teams that do it well also tag specific workflows and watch for new workflow starts, not just feature usage.

Why composite scores fail

Most health scores are weighted averages of ten or fifteen signals. They feel rigorous. They are mostly noise. The signals are usually correlated with each other, the weights are guessed, and the score moves smoothly even when something has actually broken.

A dashboard with two specific metrics, each tied to a clear intervention, is worth more than a composite score that nobody on the CS team can explain.

What to do about it

The practical move is to start tracking these two signals explicitly, set thresholds, and route alerts to the account owner. You do not need to throw out your health score. You do need to stop treating it as the primary signal.

I have watched teams cut churn by twenty percent within two quarters of making this change. Not because the underlying product improved, but because the CS team finally had a signal they could act on early enough to matter.