North Star Metrics That Actually Predict PMF

The most predictive North Star metrics measure a state change in the customer's world, not a behaviour inside your product. 'Messages sent' measures behaviour. 'Response time under 2 hours on critical customer issues' measures a change in how the customer's business actually operates. If your product claims to speed up response time, the second version is the one worth tracking.

What separates a real signal from a useful-but-wrong metric is retention correlation. Segment your customers by whether they hit your candidate North Star threshold, then compare their retention curves. If the curves diverge significantly, you've found something real. If they barely move, the metric is tracking activity that doesn't represent enough value to create a genuine dependency.

The metrics that predict retention best tend to have a natural ceiling — a point where the customer has received the core value. 'Jobs reviewed per week' has a cap based on actual hiring volume. 'Total messages sent' doesn't. Bounded metrics keep an honest score. Unbounded ones can climb steadily while customer value quietly erodes.

Monthly active users: measures who opened the app, not who received value. A user who logs in to check a notification and leaves in 30 seconds counts the same as a user who spends two hours completing a core workflow. The metric says nothing about whether the product is doing real work.

Feature adoption rate: tells you which features people click on, not whether those features improved the outcome the product promises. A feature can have 90% adoption and still not be the reason the customer renews. The adoption metric and the retention driver can be completely different features.

Session length: longer sessions can mean the product is delivering value or that it is confusing and users are spending time figuring out what to do. Without additional context, session length is uninterpretable as a value metric.

Start from the core value promise. What specific outcome does your product deliver, stated in the customer's terms? Not 'better analytics' but 'the ability to identify which customer segment has the highest 90-day LTV before allocating marketing budget to it.' The North Star should measure how often and how completely that outcome is delivered.

Then validate it against retention data. Pull your cohorts, segment by North Star metric achievement (above threshold vs. below), and compare retention curves. If the curves diverge meaningfully, you have found a predictive metric. If they don't, the metric is not connected to the value the product actually delivers.

Run this analysis quarterly. As the product evolves and as you understand the customer better, the right North Star metric often shifts. A company that started measuring 'reports generated' might find, after 18 months, that 'decisions made based on a report' is actually what correlates with retention — and shift accordingly.