North-star metrics and metric trees

One metric that proxies real user value, broken into the drivers that move it — so you see trouble before the outcome does.

The idea

A north-star metric is the single number a team steers by. The trap is choosing one that’s easy to move but doesn’t mean much — a vanity volume like sign-ups, or revenue, which trails the value that earns it.

The good ones proxy durable user value and are hard to game. But value metrics tend to lag. So you decompose the north star into a tree of driver metrics: leading indicators upstream that move first, lagging outcomes downstream that confirm later.

See it work

You run the homepage for a streaming service. Pick a north star, build its input tree, then play two quarters — a hidden problem is waiting.

How it works

Choose the north star first, then decompose it downward into inputs you can actually influence.

NORTH STAR:  retained watchers
             weekly viewers who come back and finish something
             (proxies durable value; hard to game; but it LAGS)

  leading inputs  (move first, weeks early, you can steer them)
    - recommendation hit rate   did we suggest something they watched
    - completion rate           do they finish what they start
    - time to first play        how fast a session finds a watch

  lagging outcomes  (confirm later; results, not levers)
    - subscriptions, monthly revenue, realized churn

DEFENSIBLE vs GAMEABLE
  gameable    DAU up via push spam; hours up via autoplay  -> value flat
  defensible  retained watchers rises only if people truly get value

The tree is a hypothesis: if these inputs improve, the north star should follow. That is what makes it useful — a leading input dropping is an early warning weeks before the lagging outcome moves.

When to use it

Reach for it when	Trade-off to remember
Aligning a team on one metric and the levers under it	A tree is a hypothesis about causation — it can be wrong, so revisit it
Product-sense and metrics interviews (“what would you measure?”)	One number can’t capture everything; pair it with a guardrail or two
Catching value problems early, before revenue or churn react	Leading inputs are noisier; a single week’s dip may be nothing

Watch out for

• Vanity volume as the north star. DAU and sign-ups rise with a push-notification blast that adds no value. If a growth hack can move it without helping users, it’s gameable.
• Revenue as the north star. Revenue is an outcome, not a lever, and it lags. Steer by the user value that earns it; keep revenue as a guardrail.
• A tree of only lagging outcomes. Subscriptions and churn confirm the problem after the quarter’s already lost. You need leading inputs for early warning.
• Leading indicators off the mechanism that breaks. Watching playback errors won’t catch a recommendations problem. The input has to sit on the value that’s actually at risk.
• Goodhart’s law. Once a metric becomes a target, people optimize the number, not the value. Autoplay inflates “hours watched” while satisfaction quietly falls.

Worked example

Asked to pick a north star for a streaming homepage, the weak answer is DAU — easy to move, easy to game. A stronger answer: retained watchers, people who come back and finish something, because that only rises when the product genuinely delivers. Then decompose it: recommendation hit rate and completion rate are leading inputs; subscriptions and churn are lagging outcomes. In the simulation, a recommendation regression in week 5 shows up in completion rate within a week or two — but retained watchers doesn’t visibly drop until week 16, and DAU never flinches. Naming the defensible north star and the leading inputs that give you early warning is exactly the product-sense signal an interviewer is listening for.

Check yourself

Powering your career growth.