Leading or lagging indicator?

The distinction is about timing relative to a decision. A lagging indicator measures an outcome once it has happened — quarterly revenue, annual churn, six-month retention. It is authoritative and it arrives too late to change anything. A leading indicator moves earlier and is believed to anticipate that outcome — activation rate, week-one engagement, trial-to-paid conversion — so it can inform a decision while the decision is still open.

The trade is always the same. Lagging indicators are what you actually care about and cannot steer by; leading indicators are actionable and their connection to what you care about is an assumption. That assumption is the whole substance of the pair, and it is exactly the claim a proxy metric makes — the two concepts overlap heavily, with the leading/lagging framing emphasising timing and the proxy framing emphasising substitution.

What makes a leading indicator genuinely leading rather than merely early is that it predicts the outcome under intervention, not just in historical correlation. Support ticket volume correlates with churn, but that may be because unhappy customers do both rather than because tickets cause departures — in which case suppressing tickets by making support harder to reach would move the leading indicator and worsen the lagging one. Every leading indicator carries this risk, and it is the reason the pairing needs periodic checking rather than one-off validation.

In practice a healthy measurement system runs both and uses them for different jobs. Leading indicators decide experiments and weekly operating reviews, because they move fast enough to be informative on that cadence. Lagging indicators validate the leading ones and settle the question of whether the programme is actually working — which requires deliberately holding some slow readout, whether that is a long-term holdout or simply revisiting shipped changes after a quarter.

The failure this structure prevents is a specific and common one: a year of experiments all reporting wins on leading indicators, with the lagging numbers flat. Every individual test was correctly run and correctly analysed. What went wrong is upstream of any of them — the assumed link never held, and because nothing was measuring the lagging side, nothing could say so. The only defence is measuring both and comparing them on purpose.

A B2B software company treats weekly active seats as its leading indicator for annual renewal, the lagging one. Renewal is 84% and can only be observed once a year. Over eighteen months they run experiments on seat activation and want to know whether the leading indicator is earning its authority.

A correlation of 0.63 looks like strong validation and is doing much less work than it appears to. The transfer rate of about 20% is the number that should drive forecasting, and it is only knowable because the company waited a year on seven cohorts and looked. The three non-transferring experiments are worth examining individually rather than averaging away — reviewing them, two had increased seat activation through admin bulk-invites, which adds seats that were never going to be used and moves the leading indicator without touching the thing it was supposed to predict. That is the shared-cause failure in its ordinary clothing. The practical response is not to abandon the leading indicator, which is still the only thing fast enough to steer by, but to narrow it: active seats that logged in twice or more, which is harder to manufacture and should transfer better. Then wait another year and check again.