Enterprise Master Data Cleaning: The Business Case for SAP Supplier Data

Part 1 of a series on cleaning enterprise master data, for business and leadership. Part 2 shows, hands-on, how AI finds duplicate suppliers, and Part 3 (coming soon) is the technical blueprint.

Your supplier master is the phone book your company pays from. Every purchase order, invoice and payment run looks someone up in it. If the book has three entries for one supplier, a wrong bank account, or a company that closed in 2021, the mistake follows every transaction that touches it.

Nordic companies feel this in an interesting way. The Nordics run some of the most digitized business registries in the world and adopted e-invoicing early, so the bar for clean, machine-readable data is high. Yet the vendor records behind it were often built up over decades of mergers and regional ERP rollouts.

Seven ways supplier data goes bad

What it costs

Analysts such as Gartner put the average cost of poor data quality at roughly $12–15 million a year per organization. Figures vary, but the mechanisms are easy to recognize:

• Money leaks: duplicate payments, missed early-payment discounts, and spend split across duplicate vendors so you never hit the volume tier.
• Fraud exposure: duplicated or unverified vendors and changed bank details are a classic payment-fraud route.
• Compliance risk: sanctions screening, tax reporting and GDPR all lean on accurate partner data.
• Wasted time: AP and procurement teams spend their week hunting and fixing records.
• Failed transformations: migrating dirty data into S/4HANA just moves the mess to a pricier system, and AI projects inherit it too.

Why cleanups keep failing

Most enterprises have run a cleanup project. Quality jumps, then decays as vendors move, merge and close, and you’re back where you started within a year or two.

The other problem is tooling. Exact-match rules miss “Müller GmbH” against “Mueller Gesellschaft mbH”, and stewards drown in false positives.

A Nordic lens

• Legal suffixes move around. AB, Oy, AS, A/S and ApS appear before or after the name, so “Nordic Steel AB” and “AB Nordic Steel” look like two companies.
• Diacritics get flattened. å, ä, ö, æ and ø become a, o or “aa”, which breaks search, matching and sanctions screening.
• Local payment rails. Bankgiro and Plusgiro sit next to IBAN, and bank details are the highest-fraud-risk field.
• Authoritative registries. Bolagsverket, Brønnøysundregistrene, CVR and PRH/YTJ can confirm that a supplier exists and is active.
• Peppol e-invoicing. It depends on correct organisation numbers and endpoint IDs, so bad master data breaks automated invoice flow.
• GDPR. Sole traders (enskild firma) and contact persons make some vendor data personal data.

The Nordic preference for consensus and transparency helps. A cleanup with a named owner, visible rules and an audit trail is a very lagom answer: steady and sensible, not a heroic one-off.

Where AI genuinely helps

AI doesn’t replace governance. It makes governance affordable at scale.

• Deduplication: fuzzy and semantic matching finds near-duplicates across names, addresses, tax IDs and languages, then proposes a golden record.
• Onboarding: it extracts vendor details from invoices, registration forms and bank letters straight into SAP fields.
• Enrichment: it fills gaps and standardizes names and addresses from trusted sources.
• Monitoring: it scores quality continuously and flags suspicious bank-detail changes.
• Assistants: SAP’s Joule in MDG or custom agents draft change requests and summarize them for approvers.

Here is the most common case, three records becoming one:

Behind the scenes, the data moves through three layers. Bronze is the raw supplier data exactly as it came out of SAP, kept untouched as evidence. Silver is the cleaned version: standardized, validated, with duplicates flagged and fixes proposed. Gold is the trusted result: the golden records that people approved and that SAP now holds. The layers matter to leadership because they make the process auditable. You can always show what the data looked like before, what changed, who approved it and what it looks like now.

The rule that keeps it safe

Vendor master data decides where money goes, so the pattern is never “let the AI fix it.”

Low-risk, high-confidence fixes such as a country code can run automatically. Anything that moves money goes to a person with the evidence in front of them.

Five questions for your next leadership meeting

1. Who owns supplier master data, by name?
2. What is our duplicate rate, and how do we measure it?
3. Which changes are automated, which need approval, and who decided?
4. Can we show an auditor who changed a vendor’s bank details, when and why?
5. Are we cleaning data, or stopping it from getting dirty again?

If any answer is “not sure,” you have your starting point. Part 2 shows the core matching idea in a notebook you can run, and Part 3 (coming soon) shows how to build the whole process with Airflow and human-in-the-loop approval.