Teams often ask for a forecast of revenue or churn, when what they actually need is a forecast of something earlier in the causal chain — because revenue and churn are lagging indicators. By the time they move, whatever caused the movement already happened weeks or months earlier. The place to spend forecasting effort is on the leading indicators that predict the lagging ones, while there's still time to act.
Telling the two apart
- Lagging indicators confirm what already happened — revenue realized, churn that occurred, a quarter that's closed. These should be measured accurately, not forecasted; the forecast opportunity has passed by the time they're known.
- Leading indicators move before the outcome does — declining usage frequency ahead of churn, rising support-ticket sentiment negativity ahead of a complaint spike, slowing time-to-first-value ahead of a cohort's poor retention.
- The gap between them is your action window — the earlier a leading indicator moves relative to the lagging outcome, the more time there is to intervene before the outcome is locked in.
Finding your own leading indicators
They're specific to your business, not a generic template — the exercise is correlating historical leading candidates (engagement drops, support sentiment, usage pattern shifts) against the lagging outcome you actually care about, and keeping the ones with real predictive lead time, not just correlation.
The right question isn't 'can we forecast churn more accurately.' It's 'what moves two months before churn does, and are we watching it.' The second question, answered well, makes the first one almost unnecessary.
Where forecasting effort should actually go
Spend the model-building budget on the leading indicators with real lead time and real intervention options attached — a forecast nobody can act on, however accurate, is a more expensive way of finding out what a lagging metric would have told you anyway.