The purpose of a product team is to drive customer outcomes and business outcomes. The problem is that the customer outcome we care about most is often the hardest thing to measure, which in turn creates a problem when choosing a North Star metric.
A North Star metric is the closest, practical, observable measure of the customer value your current product strategy is trying to create, and it should be a leading indicator of future business outcomes.
Suppose a product helps managers make better decisions. “Better decisions” is close to the real customer value. Try measuring it, though, and you may need months to know whether a decision was good (not practical). Even then, the judgment can be subjective, and you still do not know what the manager would have done without the product (not observable).
“Reports viewed” has the opposite problem. It is easy to measure, but a long way from the value (not closest). A more useful North Star might be “Weekly Insight-to-Action Workflows Completed”: the number of completed workflows each week that start with a question, produce an insight, and lead to a trackable action. Though it is still a proxy for better decisions, it captures the workflow through which the product creates value.
A North Star is useful because you can spot it. If the outcome you care about is more like a black hole, measure its observable effects. An activation behavior can be one such effect. If retention takes six months to observe, but something a user does in the first week represents that they saw value and strongly predicts retention, this behavior is more useful for making product decisions.
Whether to count workflows or users is a product strategy choice. Early on, the strategy may focus on getting a few power users to complete the workflow frequently, so “Weekly Insight-to-Action Workflows Completed” may be useful. Later if the strategy shifts toward broader adoption, “Weekly Users Completing an Insight-to-Action Workflow” may be more useful.
Amplitude ties the North Star to customer value, product strategy, and future business outcomes, and recognizes that outcomes that cannot be observed directly may require proxies. Getting closer to the true customer outcome usually makes a metric more meaningful, but measurement has a cost and lead time. My rule of thumb is to move toward the true customer outcome until getting closer is no longer worth the effort or time required to measure it.
Count, rate, or ratio?
The choice depends partly on whether you actually want more of the thing being counted. Nights booked works well as a count because another booking represents more customer value. Support queries are different. A customer-support chatbot cannot increase support volume and should not even try to. I would probably measure the percentage of eligible queries successfully resolved without human escalation, then use customer satisfaction or recontact rate as a guardrail.
Define and test
Name the metric and define it tightly so that when different people calculate it, they get the same number. “Engaged accounts” could mean anything. “Accounts with at least three users completing the core workflow in each of the previous four weeks” is harder to misinterpret. A metric is operational when its numerator, denominator, population, and time window are clear.
To identify candidates for the North Star, look at your “must-have” customers, using Sean Ellis’s term for customers who would be “very disappointed” if they could no longer use the product. Then, look for their “aha moment”, workflows where the product’s value becomes clear to these customers. (Revenue is usually not a useful North Star metric because it is downstream of this value realization.)
Then test the assumption behind the metric. Do customers that score well on the North Star metric actually achieve the outcome you care about? Revisit the metric as the product strategy changes. A product trying to establish product-market fit may care about a different measure of customer value than a mature product focused on retention or expansion. Make sure the metric has enough volume and represents typical user behavior. For example, “thumbs up/down” feedback is often sparse and skewed toward negative responses.
Ask one last set of questions: Could our team materially improve this metric? Could we materially improve the metric while degrading the customer’s experience? If yes, either the metric needs work or you need a guardrail metric.
PS: Check out more articles on building products. I write to pay it forward and to clarify my thinking.