Build the metric tree, not a lonely number

A north star you cannot touch directly

The property that makes a north star worth steering by is the same property that makes it useless on its own: you cannot move it directly. It is an outcome. You cannot sit down on Monday and crank up nights booked or value delivered per account by an act of will. If you could, by the test from chapter two, it would not be a good north star in the first place.

So a north star written on a wall with nothing beneath it is not a system, it is a wish. The wall says delivered value should go up, everyone nods, and nobody knows what to do tomorrow morning. The star only becomes operational when you build the tree of inputs beneath it, the levers you can actually pull that, together, move the outcome you cannot pull. This is the same shape as finding your one growth bottleneck: the star tells you the system is winning or losing, the tree tells you which input is the constraint right now.

Three to five inputs you can move this week

The rule of thumb on the tree is specific: pair the north star with a set of three to five complementary input metrics that you can directly influence. Three to five, not fifteen, because the point of a north star was focus and a sprawling input list quietly destroys it. Directly influenceable, because these are the things your team or your agents act on, the half of the system you can reach.

A clean way to find your three to five is to decompose the star along breadth, depth, frequency, and efficiency. Breadth is how many accounts or users reach the value. Depth is how much value each one receives. Frequency is how often they come back to receive it. Efficiency is how cleanly they get there, fewer steps, less friction, which in a productivity-game B2B product is often the truest lever of all. Take the self-serve SaaS example: the star "trial accounts reaching three active users in week one" decomposes into breadth (how many trials start), depth (invites sent per trial), frequency (return visits in the first seven days), and efficiency (steps from signup to first core action). Pick the three to five of these that genuinely drive your particular star and you have a tree, not a number.

The tree is what your agents optimise toward

For a lean operator running growth with AI agents, this tree is not a nicety, it is the interface. You do not point an agent at delivered value and walk away; the agent cannot move an outcome directly any more than you can. You point each agent and each weekly experiment at an input, breadth or depth or frequency or efficiency, and the inputs compound up into the star. This is exactly the loop described in run growth experiments: one experiment moves one input and you watch the star respond. The north star tells you whether the whole system is winning. The input tree tells every agent what to do on Monday. Without the tree, the star is a slogan your agents cannot act on; with it, the star becomes the single thing the whole machine is quietly optimising toward.