KPI tree
Definition
KPI tree
A KPI tree is a diagram that breaks one top line number into the measures that drive it, level by level. Each key performance indicator (KPI) below the top sits there because arithmetic links it to the one above, not because someone liked it.
Our entry on the key performance indicator (KPI) handles the single measure: what it is, and which handful deserve tracking. A tree defines none of them. It says which one moves which other one.
A key performance indicators template is the other child of that parent. It defines one measure properly, where a tree relates many. Pick the template if your problem is a vague definition, and the tree if nobody can explain why a number fell.
Trees go by other names. Driver tree, value driver tree, metric tree — the object is the same, and so is the rule that each level has to reconstruct the one above it.
Key takeaways
- A KPI tree decomposes one headline result into the drivers sitting beneath it.
- Every branch must hold arithmetically, so a change below shows up above.
- Trees locate a miss; they do not define the measures they contain.
- Three levels is usually enough, because depth costs more than it reveals.
How it works
A KPI tree starts with one number the business is judged on, then splits it into factors that multiply or add to it. You repeat the split until the bottom row names something a team can change this week.
| Level | What sits here | Contact centre example |
|---|---|---|
| Top | The outcome the business is judged on | Annual service margin |
| Second | The arithmetic factors of that outcome | Revenue per seat, cost per seat |
| Third | The operating drivers of each factor | Occupancy, attrition, rework rate |
| Fourth | What a team changes this week | Schedule adherence, coaching hours |
Build it downward, never upward. Start from the number your board asks about, split it once, and check the split is complete before going further, or you end up with a tidy diagram missing a third of the cost.
The arithmetic is the test. If the second row does not rebuild the top row, you have a list of related measures and not a tree — and the list will not tell you where a miss came from.
Completeness is the second test. If the branches at one level do not account for the whole of the level above, the tree will keep pointing at the wrong branch whenever something moves.
Public sector practice reaches the same place. In 2015 the US Government Accountability Office described a nutrition programme that built a model showing how programme outputs contribute to longer term outcomes, naming outside factors such as food prices.
The shape also appears in law. Section 1115 of Title 31 of the US Code tells each agency to set performance goals and then establish a balanced set of performance indicators measuring progress toward each one.
Goal on top, indicators beneath it. That is a two level tree written into statute, and it carries the same warning: a balanced set, not every number the system can produce.
A tree is not a balanced scorecard. The scorecard groups measures by perspective so leaders see a rounded view — the tree links them by arithmetic so an analyst can trace a cause.
Keep it small enough to read on one page. A diagram that needs scrolling has stopped being a map, and most branches past the third level belong in a separate tree of their own.
Examples
Three trees cover most of what you will meet in practice. A revenue tree, a cost tree and a quality tree, each built from the arithmetic of the number at the top. Branches change by industry; the method does not.
An ecommerce revenue tree splits revenue into sessions, conversion rate and average order value. Each of those splits again, and the performance metrics at the bottom are things a merchandiser or a media buyer owns outright.
A contact centre cost tree splits cost per contact into volume, handle time and loaded hourly rate. Volume is the branch most teams forget — and it is usually the biggest one.
Shared services teams build the same tree around throughput. The efficiency metrics beneath it cover touches per case, rework and automation rate, which is where most recoverable cost actually sits.
Vendor managers keep two trees side by side. One holds the contracted numbers; the other holds the internal metrics nobody sends the client, which is how you tell a real problem from a reporting artefact.
All four share one habit worth copying. The lowest row names an action, not a condition, because a branch nobody can act on is a fact rather than a driver.
Related terms
- Key Performance Indicator (KPI): the single measure a tree arranges; it defines, the tree relates.
- Performance metrics: the broad family of numbers that populate the lower branches.
- Efficiency Metrics: the cost side branches, measuring output against resource used.
- Descriptive analytics: the reporting that fills a tree with what already happened.
- Call drivers: the reasons behind contact volume, the deepest branch of a service tree.
FAQ
How deep should a KPI tree go?
Stop at the level where a named team can change the number inside a week. For most operations that is three levels, occasionally four.
What is the difference between a KPI tree and a driver tree?
Very little in practice. Driver tree is the commoner phrase in finance, while KPI tree is used where the branches are already defined measures with named owners.
Does every branch need to be arithmetic?
The top two levels should be. Lower down you can accept strong causal links, as long as you mark them as influence rather than identity.
Who builds and maintains the tree?
Usually the analytics or finance team, with each branch signed off by the manager who owns it. An unowned branch quietly stops being updated, and the tree starts lying.
Can a KPI tree replace a dashboard?
No. A dashboard shows the current value of chosen numbers, while the tree explains how those numbers relate, so most teams keep the tree behind the dashboard as a map.
Explore how outsourcing teams run their numbers across the Outsource Accelerator glossary and guides.
Independent
