Driver tree

You've met the issue tree already: a qualitative, MECE decomposition of a problem statement into the possible reasons behind it ("why might margin be falling: pricing, cost, mix, volume, one-offs"). The driver tree is a different tool that works alongside it. It's a quantitative, algebraic decomposition of a metric into the arithmetic components that literally produce its value. An issue tree answers "what are the possible reasons". A driver tree answers "of the actual number, which piece moved, and by how much." You need both. The issue tree tells you where to look, and the driver tree is what you build once you've picked a branch worth measuring.

What are the two types of driver tree?

Driver trees come in two mathematical flavors. Knowing which one you're building matters a great deal, because it decides whether your later arithmetic will be clean or will hide a trap:

• Additive decomposition: a metric built from parts that simply sum, e.g. Profit = Revenue − Cost, or Cost = Fixed cost + Variable cost. Additive trees are forgiving. The change in the total is always the exact sum of the changes in the parts, with nothing left over. If revenue rises by 100 and cost rises by 40, profit rises by 60. There's no residual and no ambiguity.
• Multiplicative decomposition: a metric built from parts that multiply, e.g. Revenue = Price × Volume, or Volume = Market size × Market share. Multiplicative trees take more craft, and they hide the trap. Unlike addition, multiplication doesn't split a change cleanly into independent pieces. The bridge section works this out by hand, because it's the most valuable piece of arithmetic you'll learn here.

A realistic driver tree usually nests both forms. Profit is additive at the top (Profit = Revenue − Cost) but multiplicative one level down, where Revenue = Price × Volume and Variable cost = Cost per unit × Volume:

Profit
 ├─ Revenue = Price × Volume
 │             Volume = Market size × Market share
 └─ Cost
     ├─ Variable cost = Cost per unit × Volume
     └─ Fixed cost

How do you build a driver tree?

Building the tree is mechanical once you know the business. Start from the metric the client cares about and ask "what, multiplied or added together, literally produces this number?" Keep breaking each child node down the same way until you reach something you can measure separately, like a price you can look up or a headcount you can count. Stop one or two levels past where your hypothesis lives. A tree that goes five levels deeper than your kill test needs is itself a small act of ocean-boiling, and we'll come back to that in boiling the ocean.