Driver trees

A target like “four more points of gross margin” is broken down, branch by branch, into the everyday numbers one person can read and move.

4-minute readUpdated October 2026

A big number, broken into small ones

“Raise gross margin four points” is a goal nobody can work on on a Monday. A driver tree breaks that number into the smaller numbers it is made of, branch by branch, until every end point is something one person can read from a system and move by their own work.

Every target in the plan sits on top of one of these trees. The sample has five trees, one per objective, and they produce 38 measurable numbers. The plan keeps the ones that matter and assigns each an owner.

The tree is drawn for one business, one period and one definition, fixed before anything is drawn. The root is the engine input the objective moves; the trees hang beneath the valuation itself, so the whole report is one connected structure from the value at sale down to, for example, the share of discounts that carry no reason code.

Four of the sample’s trees

Tap a tree. Branches below the second level open when you tap them. Labels marked “in the plan” are the measures the team will track; “not measured yet” means the first job is to start measuring it.

Why each tree has a different shape

The shape is not a matter of taste. The kind of number at the top decides it, so two analysts given the same number draw the same tree.

• A per-unit numberPrice per seat is built up from its parts, each expressed per seat: list price, less discounts, plus add-ons.per-unit build-up
• A rate or a marginGross margin is one minus the share each cost element takes. Every cost line becomes a branch.loss form
• A loss rateChurn is split by cause, and the causes add up exactly. Losses with no recorded cause get their own branch.causes plus an unrecorded bucket
• A balanceSeats are last period’s seats, plus every way seats arrive, less every way they leave.roll-forward

When the parts combine exactly (a sum, a product, a ratio), the link is arithmetic and the tree must reconcile to the recorded number. When they don’t, the link is causal and carries a direction and a grade. In the sample, every causal link is a management hypothesis (marked h+ or h−) rather than a measured relationship; the report says so on the page.

• Link precedence. Arithmetic where an exact formula passes the split and identity tests; causal only where none does; a proxy only to end a construct that no system records
• Leaves. Every branch ends at one of five kinds: tactical (read from a system of record, moved by one role, observable daily to monthly), exogenous (matters but nobody inside can move it), state (inherited from earlier periods), parameter (a number the business sets, such as a quota or a rate card) or a reference to a node drawn elsewhere
• One driver, one place. A driver appears once, at its first occurrence, and is referenced after; never as a near-duplicate under another parent
• Depth. Most leaves sit two to four levels down, none deeper than five; a deeper branch becomes a tree of its own
• Slices are not children. Splitting by customer, region or channel shows where a change sits, not why, so it is a view on a node, never a branch

See it on a real report, or see it on yours.Download a sample reportSchedule demo