metric tree

Also called: driver tree, KPI tree, metrics tree, metric decomposition

A breakdown of a top metric into the input metrics that drive it, so each team can see which lever it moves.

A metric tree starts from one top metric (a north star or a revenue line) and splits it, level by level, into the input metrics that produce it, until each leaf is something a team can act on directly. It answers two questions: "what should my team move?" and "why did the top number change?".

There are two ways to split a node, and a good tree uses both:

Multiplicative: a rate chain, where the top equals the product of its factors. For subscription revenue:

$\text{New Plus revenue} = \text{MAU} \times \frac{\text{paywall viewers}}{\text{MAU}} \times \frac{\text{trials}}{\text{paywall viewers}} \times \frac{\text{paid}}{\text{trials}} \times \text{price}$

Additive: a sum of parts, such as revenue = monthly plans + yearly plans, or active groups = flats + couples + trips + other.

To build one, write the top metric's exact definition, split it so the parts multiply or add back exactly to the parent, stop when a leaf has a clear owner and can move within weeks, and note which leaves you can actually measure today. A tree whose parts don't reconcile with the top is a mind map, not a metric tree.

The tree earns its keep in diagnosis. When the top metric moves, compare each factor period over period: in a multiplicative tree the relative changes multiply, so you can see which factor explains the move instead of arguing about it. It also shows trade-offs: pushing one leaf (more paywall views) can pull down the next (a lower trial rate).

Example

Halves tracks new Plus revenue from first monthly payments (monthly plan, $5). Maya sees it fall from May to June while MAU grows, and walks down the tree:

Factor	May	June	Change
MAU	40,000	45,000	×1.125
Saw the paywall (share of MAU)	8% → 3,200	8% → 3,600	×1.00
Started a trial (share of viewers)	6.25% → 200	5% → 180	×0.80
Paid after the trial	40% → 80	40% → 72	×1.00
Price	$5	$5	×1.00
New Plus revenue	80 × $5 = $400	72 × $5 = $360	×0.90

Check: 1.125 × 1.00 × 0.80 × 1.00 × 1.00 = 0.90, i.e. revenue −10%, matching $360 / $400.

Everything else held or grew, so the whole drop sits in one leaf: the trial start rate fell by a fifth (5% / 6.25% = 0.80). That points the investigation at the paywall screen itself (a new design shipped on 3 June) rather than at marketing, which actually did its job.

Common mistakes

Parts that don't reconcile. If the factors don't multiply or add back to the top metric, the tree can't explain any change.
Leaves nobody owns. A leaf without a team or a lever is decoration; keep splitting or merge it.
Mixing denominators. "Trial rate" must be trials per paywall viewer at every level, not sometimes per MAU.
Ignoring links between leaves. Pushing more users to the paywall often lowers the trial rate; read neighboring factors together.
Too deep, too early. Twenty leaves you can't measure are worse than five you can; grow the tree as tracking improves.