SStructera

How Structera works

Last updated: 20 May 2026

Structera turns a raw transaction export into a clean, classified spend dataset in five steps. Every step is free up to and including the sample review — you only pay when you decide to apply the validated taxonomy to the full dataset.

  1. Step 1 — Upload transactions. CSV or XLSX (up to 25 MB). Supports supplier, material, and service data modes. FREE
  2. Step 2 — Generate a 3-level taxonomy. Anchored to UNSPSC, SIC, NAICS, and NACE. Editable inline. FREE
  3. Step 3 — Classify a sample. ~100 statistically representative rows, tagged automatically in seconds. FREE
  4. Step 4 — Review & refine. Approve, override, or correct any row before you pay. FREE
  5. Step 5 — Apply & export. Pay once, classify the whole dataset, download the Excel workbook (Summary, Taxonomy, and the procurement-consultant insight sheets). PAID

Step 1 — Upload your transactions

Export your procurement data as CSV or XLSX. Three data modes are supported: suppliers (one row per invoice or purchase order line, keyed on supplier name), materials (one row per material or item purchased), and services (one row per service line). Each mode accepts an optional description, amount, and date column. Templates are downloadable inside the upload step so you can see the exact column shape.

Uploads are capped at 25 MB per file, which comfortably covers datasets of around 100,000 rows. Bigger datasets can be split, or you can ask us about a higher limit.

Step 2 — Generate a 3-level taxonomy

18 universal Level 1 procurement towers anchored to UNSPSC, SIC, NAICS and NACE, with an example IT and Software branch expanding to Cloud infrastructure (Level 2) and Cloud compute services (Level 3)

Structera reads every transaction in your dataset (not just a sample) and infers a clean taxonomy with three layers: a fixed Level 1 of 18 universal procurement towers anchored to the four global standards (UNSPSC, SIC,NAICS, NACE), plus Level 2 families and Level 3 sub-categories tailored to the specifics of your spend. Fixed Level 1 means your taxonomy is comparable with industry benchmarks, peer companies, and your own historical data. Every category is editable inline — rename, re-parent, or delete anything that doesn't fit.

Step 3 — Classify a sample (free)

A statistically representative sample is drawn from your dataset (95% confidence, 10% margin of error — around 100 rows for most datasets). Each row is classified into a Level 3 leaf category by an AI engine constrained to the global procurement standards (UNSPSC at the leaf level, NAICS / SIC / NACE cross-references at the top), with three layers of resilience: a shared dictionary of items that have been categorised before, deterministic keyword rules for high-signal patterns, and the LLM for everything else. If the LLM fails for a row after several retries, Structera falls back to a keyword match so every row gets a category.

Step 4 — Review the sample

Walk through the sampled rows. Approve the ones that look right; for the ones that don't, click the suggested category, search for a better fit, or create a new category on the fly. You see and fix the result on your own data before any payment is taken.

Step 5 — Apply to the full dataset (plus the insight pack)

When the sample looks right, unlock the full dataset with a single pay-as-you-go payment. The same classification logic runs across every row, in parallel, with progress visible in real time. When it finishes, you can export an Excel workbook with the three core sheets —Summary (one row per transaction with assigned Level 1/2/3), Transactions (full raw data plus classification metadata), and Taxonomy (the validated category tree) — alongside a six-sheetprocurement-consultant insight pack: an executive summary, a Level 1 category analysis with strategic recommendations, top-20 suppliers with action notes, supplier rationalisation candidates with estimated savings ranges, a tail-spend analysis, and a prioritised quick-wins playbook.

Why it works

Three design decisions matter for quality and speed:

  • Universal Level 1 towers — the top of the taxonomy is fixed to 18 widely-used categories so Level 1 is comparable across companies, industries, and time.
  • Sample-first review — you see the system's confidence on a real slice of your data before paying for the full run.
  • A dictionary that learns — items Structera has categorised before keep the same family on every run, which makes results consistent and runs faster.

Related