Driver trees, the method
Any number you want to move breaks down into everyday measures that one person on your team actually controls. A driver tree is how.
Why big numbers get broken down
Nobody can work on “increase the value” on a Monday morning. Somebody can work on “call every past-due account within ten days.” A driver tree is the chain that connects the two, so every everyday measure on the plan traces up to a dollar of value.
The sample report carries five of them, one under each input the plan moves: price per seat, gross margin, overhead, seats, and customers lost each year.
The kind of number picks the shape
A tree isn’t drawn freehand. The kind of number at the top decides how it splits, so two analysts building the same tree get the same structure.
Churn, margin and conversion rates split into the causes of the loss, and the causes add up exactly. Gross margin becomes one minus each cost line’s share of revenue.
churn = Σ losses by cause, including “no recorded reason”
Price per seat or cost per seat is built up from its parts, each stated per the same seat.
cost per seat = minutes + numbers + 911 + platform + hosting
Seats, customers or headcount roll forward: what you started with, plus what came in, minus what went out.
seats = opening seats + seats added − seats lost
Revenue and costs split into a base times a rate, or a sum of their parts.
service revenue = seats billed × average price per seat
A worked example: customers lost each year
Here is the top of the sample’s churn tree, two levels down. Each leaf names the role that moves it and how often it is read.
- Customers lost each yearrate · split by cause, the parts add up exactly
- Lost to a competitor’s offer
- Customers out of contractAccount management · monthly
- Price above the marketOwner · monthly
- Cancellations savedAccount management · monthly
- Lost to a service failure
- Call quality below standardNetwork operations · daily
- Outage minutesNetwork operations · monthly
- Time to resolve a ticketSupport · weekly
- Lost through a partner
- Partners contacted in 90 daysChannel manager · monthly
- Rival providers’ partner incentivesoutside your control · monthly
- Business closed or moved
- Small-business closure rateoutside your control · quarterly
- Stopped payingsuspensions × (1 − cure rate)
- Autopay enrollmentBilling · monthly
- Past-due accounts called within 10 daysBilling · weekly
- Left with no recorded reasonthe measurement gap, named
- Disconnects closed without a reason codeBilling · monthly
- Exit calls completedAccount management · monthly
- Lost to a competitor’s offer
How to read it. The dark box is the number the plan moves. Each box under it is one cause; the causes add up to the whole. The bottom row of each branch is a measure someone owns.
Three kinds of link
= exact arithmetic→ cause, with a direction≈ a measurable stand-in
An exact link is used wherever a formula exists. A causal link is used only where none does, and it carries a sign, a rank, a lag and a grade for its evidence: measured from your own records, estimated from a benchmark, or asserted from judgment. A stand-in is used only to end a branch that can’t be read from any system, like “platform embeddedness”, measured by the share of accounts using two or more add-on features.
Leaves are one of five kinds: tactical (one role moves it, read from a system of record), outside your control, inherited from earlier periods, a parameter someone sets, or a reference to a measure already expanded elsewhere.
Rules that keep a tree honest
One driver, one place
A measure enters the tree once. Where it matters to two branches, the second is a reference back to the first, so its effect is never counted twice.
Leftovers are named
If the causes don’t explain the whole, the remainder is its own branch with its size printed. A remainder above a tenth of the total is a measurement gap, and the tree isn’t accepted as explaining anything until it shrinks.
Slices aren’t causes
Splitting churn by region or by product says where it happens, not why. Those views hang beside a box; they are never its children.
Measure before you manage
A measure the plan depends on but nobody records gets an instrumentation task with a date. Until it reports, its status is “not measured yet”, never “on track”.
∂V / ∂ℓ = (∂V / ∂x) · (∂x / ∂ℓ)
the value of moving a leaf = the value of moving the input it feeds × how much the leaf moves that input
- Sample tree size. The churn tree has 1 root, 13 intermediate measures, 20 tactical leaves, 3 outside-control leaves and 4 references, three levels deep
- Evidence. The causal links are hypotheses with their evidence grade shown, not measured elasticities; the report says so
- Depth. Most leaves land two to four levels down and none deeper than five; a branch that would go deeper becomes a tree of its own
See it on a real report, or see it on yours.Download a sample reportSchedule demo