KPI Trees

Building a tree is mechanical once you know the two moves, and almost everyone gets one of them wrong the first time.

Start at the top with a single goal metric — the number the business genuinely cares about this year. Then ask: what does this number multiply or add up from? Not “what affects it”, which lets anything in, but what it is arithmetically composed of. Revenue splits into new customers and revenue per customer. Those split again. Keep going until you reach something a team can move directly.

The two joins you are allowed are multiplication and addition. Multiplication is for rates and volumes chained together: visitors × conversion = customers. Addition is for parts of a whole: new revenue + expansion revenue = total revenue. If you cannot describe a join as one of those two, the branch does not belong there yet.

The common mistake is including things that influence the parent without composing it — brand awareness, customer satisfaction, team morale. These are real and they matter, and they are not tree nodes, because you cannot check them. The test to apply at every level: if I improve every child by 10%, can I calculate exactly what happens to the parent? If not, the level is decoration.

Keep it shallow, too. Three or four levels is usually enough to reach something actionable, and a tree with nine levels is a diagram nobody will open twice.

A finished tree gives you twenty or thirty numbers. The obvious next question is which one to work on — and “the one that looks worst” is a surprisingly bad answer.

Three things decide the value of a lever, and you need all three together. Size: how much room is there to improve? A conversion rate at 2% has more headroom than one at 61%. Reach: how much of the business does this node touch? Improving something that applies to every customer beats the same improvement on a fifth of them. Effort: what would it cost to move it?

Reach is the one people forget, and it is often decisive. Lifting conversion from 2% to 3% and lifting average order value by 10% can look like similar wins on a slide. If conversion touches every customer and order value only applies to the fifth who buy the add-on, the first is worth roughly five times the second for the same work — and you can only see that when both numbers sit in the same structure with their volumes attached.

This is the real payoff of building the tree. Not the diagram, but the fact that it makes two proposals comparable that were previously just two opinions. Someone can still disagree, but now they have to disagree with an arithmetic statement rather than with your enthusiasm.

One caution: the biggest lever is not always the right one this quarter. A node with huge potential and a twelve-month build may lose to a smaller one you can move in three weeks — especially if you need evidence that the tree is right before anyone will fund the big bet.

Two failures turn a KPI tree from a decision tool into wall decoration, and both are easy to spot once you know the shape of them.

The first is the vanity metric: a number that only ever goes up. “Total registered users, all-time” is the classic. It rose every week for two years at one company, including the six months when they were losing more accounts than they gained, because a cumulative count cannot fall. A metric that cannot go down cannot carry information — it is a counter, not an indicator. The fix is to make every node capable of moving in both directions: active users this month rather than users ever.

The second is the broken branch: a level whose children do not actually produce the parent. Sometimes this is a soft node like “engagement” wedged into an arithmetic chain. Sometimes it is subtler — two branches that overlap, so improving both double-counts the same customers and the tree promises a gain that never arrives.

Both failures share a symptom worth watching for: the tree stops being consulted. If a team built one in January and nobody has opened it by March, it is almost always because someone tried to use it to settle an argument, found the arithmetic did not hold, and quietly went back to opinions.

A tree is only worth building if you intend to check it against reality. When a branch improves and the parent does not move, that is not an annoyance — it is the tree telling you your model of the business is wrong, which is the most valuable thing it will ever say.