Activation is necessary, but it's not predictive
Most PLG dashboards treat activation — the user hitting some defined first-value milestone — as the headline metric, and it's an important gate. But activation alone doesn't predict who sticks around three months later. It just predicts who got far enough to have a chance. The metrics that actually forecast retention measure depth and repetition of usage after that first milestone. Not the milestone itself.
Three metrics worth tracking past activation
- Time-to-second-value — how long it takes a user to get value a second time, not just once. A single "aha moment" is easy to manufacture in onboarding. A repeated one is a much stronger signal of habit forming.
- Breadth of feature adoption — users who adopt two or three connected features in their first weeks tend to retain better than users who use one feature intensely, because breadth makes the product harder to fully replace.
- Team or account expansion — in products with a collaborative or multi-seat piece, whether a user invites teammates in their first month is often a stronger early retention signal than any usage metric measured on that user alone.
Why these metrics get missed
Activation is easy to define and easy to report quickly, which is exactly why it dominates early-stage dashboards — teams need something to show before they have enough retained cohorts to measure retention itself. The deeper metrics take patience. You need months of cohort data before time-to-second-value or feature-breadth correlations show up, and by then the team's often already built its whole reporting habit around activation and never revisits the question.
Building them into the dictionary early, even before you can measure them
The practical fix is defining these deeper metrics in the KPI dictionary as soon as the product has any usage data at all, even before there's enough history for a trustworthy number. That way the instrumentation exists by the time the cohort data's ready, instead of discovering six months in that the event you needed was never logged.
A word of caution on over-indexing on any single leading indicator
The moment a metric like time-to-second-value shows a strong correlation with retention, it's tempting to crown it the new North Star and optimize hard toward it. Be careful with that jump. A correlation in historical cohort data describes what retained users had in common — not necessarily what causes retention. Push every user toward hitting a leading indicator faster, and sometimes the correlation weakens, because you've optimized for the symptom instead of whatever the metric was originally a proxy for.
Safer to treat these deeper metrics as diagnostic — useful for understanding which cohorts are healthy and why — rather than promoting them straight to Primary-tier targets the whole team gets compensated on.
And revisit the correlation itself periodically. A leading indicator that predicted retention well for one cohort can weaken as the product, the audience, or the onboarding flow changes. Treat the relationship as something to keep checking, not a fact you establish once and trust forever.
About the author
Zach Edelstein
Founder, KPI Compass
Zach has spent the last decade in data analytics, working both inside large media agencies and in-house at enterprise companies. He built KPI Compass because he kept hitting the same walls every BI team eventually hits — messy definitions, benchmarks nobody can verify, dashboards that quietly drift from reality. He's especially into where AI actually helps with this work, and he's still actively evolving KPI Compass to keep up with how fast the data landscape moves.
KPI Compass on LinkedIn