How to build a procurement spend cube (the pragmatic way)

Published 23 May 2026 · 6 min read

Spend cubes are simple to describe — classified procurement spend pivoted by supplier, category, and time — and surprisingly painful to build. This post walks through the practical steps that work, from raw ERP export to a cube your CFO can actually use.

The five components you need

A transaction file. 12-36 months of supplier payments, with at minimum: supplier name, line description, amount, date.
Supplier normalisation. Mapping duplicate spellings (“AWS” / “Amazon Web Services”) and parent / subsidiary relationships.
A 3-level taxonomy. Level 1 (universal towers like IT, HR, Logistics), Level 2 (category families), Level 3 (leaf sub-categories that map to UNSPSC).
Classification of every row against the taxonomy.
A pivot tool. Excel for small datasets; Power BI / Tableau / Looker for anything serious.

Where it usually goes wrong

Steps 3 and 4 are where 80% of spend-cube projects stall. Most teams either spend weeks manually classifying transactions in Excel (which doesn't scale) or commit to an enterprise platform (which takes months to implement). Neither suits a finance team that needs the answer for a board paper next week.

The pragmatic build

Export. Pull the last 12-36 months from your ERP / accounting system. CSV or XLSX. One row per invoice line or PO line.
Quick normalisation. Run a fuzzy-match pass on supplier names — group on uppercase + collapsed whitespace to start. Don't over-engineer at this stage; you can refine after the first cube is up.
Generate a taxonomy from your data. Either hand-craft one (slow, but you know your business) or use a tool that proposes one anchored to UNSPSC / SIC / NAICS / NACE. The fixed Level 1 matters: it's what makes your cube comparable with peer companies and industry benchmarks.
Classify a sample first. 95% confidence + 10% margin gives a sample of around 100 rows for most datasets. Validate accuracy on the sample before you classify the whole dataset.
Apply to the full dataset. Once you trust the sample, classify everything.
Load into a pivot. Three sheets: Summary (one row per transaction, with Level 1/2/3 assigned), Transactions (full raw + classification metadata), Taxonomy (the validated category tree). Drop the Summary sheet into your BI tool, build the supplier × category × time pivot, and you're done.

What questions a good spend cube answers

Top 10 suppliers by category — consolidation candidates surface immediately.
Spend by category over time — budget tracking and trend analysis.
Tail spend per business unit — where unmanaged spend is hiding.
Year-on-year savings tracking by category.
Supplier diversity reporting (regional, minority-owned, SME) when supplier metadata is attached.

How long should this take?

Manual approach (analyst, Excel): 2-6 weeks per 100k rows. Enterprise platform: 8-16 week implementation. Structera: under an hour for the upload and review of a sample, then minutes for the full-dataset run.

If you want to skip the manual classification step, create a free account — sample classification is free and you can validate accuracy on your own data before paying anything.