Customer success metrics that matter

The test for including a metric is specific: name the decision it changes and the person who makes that decision. "Track NPS" fails the test. "Track NPS quarterly so the CS lead can trigger outreach to detractors within 48 hours" passes.

Applying this ruthlessly usually cuts a dashboard by half. What remains is smaller, gets looked at, and produces action — which is the entire point. A comprehensive dashboard that nobody reads is worse than a sparse one that drives weekly decisions, because it consumes the credibility that a useful one would need.

Gross revenue retention excludes expansion and can never exceed 100%. It answers: of the revenue we had, how much did we keep. This is the cleanest accountability metric for retention work because a few large upsells cannot flatter it.

Net revenue retention includes expansion and can exceed 100%. Above 100% means existing customers more than replace all losses, which makes growth structurally less dependent on acquisition spend. It is the metric investors weight most heavily, and for good reason — it compounds.

The gap between them is itself a useful number. GRR of 88% with NRR of 112% means expansion contributes 24 points. A wide gap over a weak GRR means you are outrunning a leak, which is more fragile than the NRR figure alone suggests.

Logo retention counts accounts rather than dollars. Track it alongside revenue retention: divergence tells you whether your losses are concentrated in large or small accounts, which changes where you deploy effort.

Health scores are prioritization devices, not predictions to be reported. Their value is lead time, so they must update continuously and decompose to the evidence behind each movement. A health score nobody has validated against actual churn is an untested hypothesis with false authority.

The leading indicators worth instrumenting directly: stakeholder stability, engaged contacts per account, engagement recency and reciprocity, unresolved commitments, and value realization. These predict better than usage in most B2B contexts, and they are the ones most commonly missing because they live in conversations rather than event streams.

Time-to-first-value deserves its own line. It is both a leading indicator of retention and a direct target for improvement, which is rare.

NPS tracks relationship sentiment over time and is a reasonable trigger for outreach. It is a poor account-level churn predictor, largely because the person filling in the survey is frequently not the person who decides on renewal. Treat it as one input to a health score, never as the score.

CSAT measures a specific interaction and is transactional. Ask it immediately after the interaction, keep it to one question, and remember that scale design and top-box definition change the number enough to make external benchmarks meaningless.

Customer effort score tends to predict repeat behaviour better than either, because reducing friction correlates with loyalty more reliably than exceeding expectations does. It is also the most actionable of the three, since a high-effort result points at a specific process.

Always publish response rate alongside any survey metric. A rising score with a falling response rate is usually a warning, not a win — disengaged accounts are the least likely to answer.

Accounts per CSM, revenue under management per CSM, and cost of retention as a share of revenue are all legitimate operating metrics. They become harmful when treated as targets to maximize rather than constraints to respect: raising book size without changing the touch model simply converts proactive work into reactive work, and the retention consequence arrives two quarters later where nobody attributes it.

Track the ratio of proactive to reactive time if you can. It is hard to measure and unusually diagnostic — when reactive work consumes the week, the work that actually drives retention stops happening, and no other metric shows this until churn does.