North Star Metric

A north star metric is one number. The work is never in the number. It is in the three or four inputs underneath it, because those are the only things a team can move on a Tuesday.

Teams that pick a north star and stop there have bought a poster. Teams that build the input tree have bought a roadmap.

What is a north star metric?

A north star metric is the single measure a product team agrees best captures the value customers get from the product. Sean Ellis popularised the term in growth circles, and the most widely used written treatment is Amplitude's North Star Playbook, which pairs the one metric with a small set of input metrics that drive it.

The example cited most often is Airbnb's nights booked. It is a good teaching example because it satisfies both sides at once: a guest gets a night somewhere, a host gets paid for it, and Airbnb earns a fee. One number, three parties served.

What makes a good north star metric?

Four tests. A candidate metric has to pass all four, and most fail the third.

1. It expresses value the customer received. Nights booked is a night somebody slept somewhere. Revenue is what you collected afterwards.
2. It leads rather than lags. You want a number that moves this month because of work done this month. Annual recurring revenue moves months after the thing that caused it.
3. The team can move it inside a quarter. A metric nobody can influence in the planning cycle becomes wallpaper by week six.
4. It is one number, and it breaks into inputs. If you cannot write it as roughly three multiplied terms, it is a theme rather than a metric.

How does each north star candidate get gamed?

Every metric a team can be measured on is a metric someone can move the wrong way. Charles Goodhart's observation, usually stated as "when a measure becomes a target, it ceases to be a good measure," is not a warning about bad people. It describes what happens to any number that decides bonuses. So pick the counter-metric at the same time as the metric.

A worked example: the input tree, with the arithmetic

Say a team-collaboration product picks weekly active paying teams as its north star. It currently sits at 4,000. The tree beneath it, written as multiplication:

new teams started x activation rate x month-one retention = new paying teams added, and the existing base carries forward at its own retention rate.

Current numbers: 1,200 teams start a trial each month. 34% of them activate, meaning three people post in a shared channel in the first week. 78% of activated teams are still paying a month later. So the monthly addition is 1,200 x 0.34 x 0.78 = 318 teams.

Now run the sensitivity, because this is the part that decides what goes on the roadmap.

• Spend to raise trial starts by 20%, from 1,200 to 1,440. New addition: 1,440 x 0.34 x 0.78 = 382. A gain of 64 teams, and it costs money every month you want it.
• Raise activation by 5 points, from 34% to 39%. New addition: 1,200 x 0.39 x 0.78 = 365. A gain of 47 teams, and it keeps paying after the work is done.
• Raise month-one retention by 5 points, from 78% to 83%. New addition: 1,200 x 0.34 x 0.83 = 339. Only 21 teams, so this looks like the weakest of the three.

Except it is not, and this is where a single-month view misleads. Retention compounds against the whole base. This month's cohort is one slice of it. Against a base of 4,000 teams, five points of monthly retention is worth vastly more over a year than five points of activation applied to 1,200 trials. A traffic gain of 20% is the biggest number in month one and the smallest by month twelve.

The tree is what surfaced that. The north star number on its own would have shown the same 4,000 either way.

Where do north star metrics break down?

1. Choosing a lagging outcome. Revenue, ARR and NPS all move well after the work that caused them. A team cannot run an experiment against a number with a two-month lag, so it stops trying.
2. No counter-metric. Every candidate in the table above has a cheap way to move it. Without a paired metric on the same chart, the cheap way wins, because it works.
3. One number for a two-sided market. Supply and demand do not share a north star. A marketplace that picks a demand-side metric will under-invest in supply for exactly as long as the metric holds.
4. One number for a multi-product company. By the time a company has four products with different business models, a single north star is either so abstract it guides nothing or it belongs to one product and quietly deprioritises the rest.
5. Changing it every two quarters. The value of a north star comes partly from the comparison across time. Replace it twice a year and you have a series of short unrelated charts.
6. Picking a metric the team cannot move. If the only lever on the number sits with sales or with a partner, the product team will watch it rather than work on it.
7. Confusing the north star with the company goal. The company goal can be revenue. The north star is the customer-value number the team believes causes that revenue. Keeping them separate is what lets you notice when the belief turns out to be wrong.

How do interviewers ask about north star metrics?

This is one of the most reliably asked question shapes in a PM loop, because the answer is hard to fake. Expect:

• "What would you pick as the north star for [our product]?" Give one metric, then immediately give the input tree and the counter-metric. Stopping at the metric is a half answer.
• "Your north star is up and your users are complaining. What do you do?" The Goodhart question. Work out which cheap lever moved, then propose the counter-metric you should have had.
• "Would you pick DAU or weekly retained users for this?" A test of whether you understand product cadence. A tool used once a week has no business being graded on daily actives.
• "How would you break that metric down?" The follow-up that separates candidates. Practise writing any metric as three multiplied terms until it is automatic.

The strongest answers name the metric in one sentence, spend most of the time on the tree, and finish with the number the metric could be gamed against. That order signals you have owned a metric rather than read a list of them.