How to clean supplier master data in a day

Published 14 August 2026 · 6 min read

Supplier master data is the single biggest reason spend analysis produces the wrong answer. If “AWS”, “Amazon Web Services” and “AMAZON WEB SERVICES EMEA SARL” are three different rows, your top-supplier report is wrong and every consolidation decision built on it is wrong too.

You do not need a master data management project to fix this. You need about a day.

Step 1: normalise the strings

Before anything clever, do the mechanical work. Uppercase everything, strip punctuation, collapse whitespace, and remove legal-entity suffixes — LTD, LIMITED, PLC, GMBH, SARL, BV, INC, LLC, SA, AG, PTY. In most datasets this alone collapses 10–20% of apparent duplicates.

Step 2: cluster what is left

Sort the normalised names alphabetically and read down the list. Near neighbours are usually the same supplier. For anything larger than a few thousand suppliers, use fuzzy matching — a similarity ratio above roughly 0.9 on the normalised string is a reliable candidate, below 0.8 is mostly noise.

Do not auto-merge. Produce a candidate list and have a human confirm. Two genuinely different suppliers with similar names is a rarer problem than a wrongly merged parent, but it is a much more embarrassing one.

Step 3: decide your parent policy before you merge

This is the step people skip and regret. Do you want spend rolled up to the ultimate parent, or kept at the trading entity?

Roll up to parent if the question is negotiating leverage. You want to know you spend GBP 4m with one group, not GBP 500k with eight subsidiaries.
Keep at entity if the question is contract compliance or payment terms, which are usually entity-specific.

You can hold both if you keep the raw name in a separate column. Never overwrite the original — you will need it to reconcile back to the ledger.

Step 4: handle the one-timers honestly

Most supplier files have a long tail of vendors used once. Cleaning those individually is not worth anyone's time. Set a threshold — spend below some floor, or a single transaction — and treat that group as a bucket. You are not trying to achieve perfect data, you are trying to make the top 80% of spend trustworthy.

Step 5: write down what you decided

The half-day you lose next time is remembering whether you rolled up subsidiaries and what your merge threshold was. A short README next to the file is enough.

What good looks like

You have not finished when every name is perfect. You have finished when the top 50 suppliers by spend are unambiguous, the parent policy is documented, and the raw column still exists. That is a working supplier master, and it is achievable in a day.