Spend taxonomy: how many levels and categories do you need?

Published 25 August 2026 · 6 min read

A spend taxonomy is the category hierarchy you classify procurement transactions into — the structure that turns a payment ledger into a spend analysis. Two questions come up whenever a team designs one from scratch: how many levels deep should it go, and how many categories belong at each level. Both have defensible answers, and getting them wrong is the most common reason a taxonomy is quietly abandoned six months after it is built.

What is a spend taxonomy?

A hierarchy of procurement categories, usually two to four levels deep, that every transaction gets assigned to. It is sometimes called a spend category tree, a commodity structure, or a procurement category hierarchy. Standards like UNSPSC, NAICS and NACE are published examples; most organisations build their own, anchored to one of those for comparability.

It is the prerequisite for everything downstream. Without it you have a transaction log; with it you have a spend cube.

How many levels should a spend taxonomy have?

Three is the working standard: a top-level tower, a family beneath it, and a leaf that names the thing being bought.

Level 1 — tower. Board-level reporting. “Facilities”, “IT and Telecommunications”, “Logistics”.
Level 2 — family. Where a category manager owns a budget. “Cleaning equipment”, “Hand protection”.
Level 3 — leaf. The commodity itself. “Mops”, “Gloves”, “Boots”.

Four levels look more precise and behave worse. Every extra level multiplies the number of judgement calls a human has to make, and each one is a chance for two people to classify the same item differently. A taxonomy nobody applies consistently is worse than a coarser one everybody does.

The exception is a large direct-materials manufacturer where the engineering bill of materials already carries a deep hierarchy. If that structure exists and is maintained, use it. Do not invent one.

How many Level 1 categories?

Between ten and twenty. Below ten and the towers are so broad that “Facilities” contains a third of your spend and tells you nothing. Above roughly twenty-five and nobody can hold the list in their head, which means people stop navigating it and start guessing.

The common failure is inflation. A team starts with fourteen towers, then adds one every time an item does not obviously fit, and ends up with forty. At that point Level 1 has stopped being a summary and become a second Level 2.

When something does not fit, the right move is almost always to widen an existing tower's definition, not to create a new one.

How many Level 3 leaves?

This one scales with your data, and that is fine. A 10,000-line materials file might legitimately produce 800 to 1,200 leaves, because that is roughly how many distinct commodities a manufacturer buys.

The test is not the count. It is whether each leaf corresponds to a buying decision someone could act on. “Gloves” passes: a buyer can run one tender for gloves. “Blue nitrile gloves size medium powder-free” fails, because that is a specification, not a category — and it fragments a single negotiation into forty.

Aim for the level of granularity at which you would go to market. Boots are boots. Brooms are brooms.

Signs your taxonomy is wrong

A single leaf holds more than 5% of lines. Too coarse — it is hiding distinct commodities.
Hundreds of leaves hold one line each. Too fine, or your leaf names are really product descriptions.
An “Other” or “Miscellaneous” tower over 5%. The structure does not fit the business.
Two people classify the same item differently. The definitions are ambiguous, not the data.
Categories named after suppliers. A classic drift — you are now mapping vendors, not spend.

Anchor to a standard, do not adopt one wholesale

UNSPSC has tens of thousands of codes across four levels. Almost nobody needs that. Use it as a reference for naming and structure so your categories are recognisable to outsiders and comparable to benchmark data, then build the subset your spend actually touches.

A taxonomy is a management tool, not a filing standard. It should be the smallest structure that lets you make decisions.