// DEEP DIVE | PRODUCT DATA ENRICHMENT
Golden Records: The Single Truth for Your Products
Master data management starts with one question: which value is right when ERP, supplier, marketing, and shop all say something different? A golden record is the clean essence drawn from all sources through clear survivorship rules and fuzzy matching. Instead of one truth per system, a single source of truth emerges that gets distributed to all channels. Changes happen once, consistently, everywhere.
THE PROBLEM
When every system says something different
The same Adidas sneaker sits in the ERP as "Adidas Sneaker model X, color red, price 89 euros." In the supplier feed as "Adi-das running shoe, dark red, 84 euros." In the marketing tool as "Adidas Performance Runner, red, 89.99 euros." Three systems, three truths. Whoever runs reports gets three different answers.
For the customer that means: a different price in the newsletter than in the shop, a different image in the marketplace listing than in your own shop, a different product label on the invoice than on the website. Trust drops, return rates rise, service inquiries explode. What starts as a data detail becomes a conversion brake.
The solution is not "align all systems" (the expensive variant that rarely succeeds). The solution is an additional master layer that builds the best dataset from all sources: the golden record. Original sources stay untouched, on top sits a consolidated truth distributed to all channels.
What is a golden record?
A golden record is the best possible consolidated dataset for an entity (product, customer, supplier), merged from multiple source systems through defined survivorship rules. Unlike a manually maintained "master dataset," a golden record emerges through rules that determine which source wins per field. Example: for prices, ERP wins. For marketing copy, CMS wins. For technical specifications, the supplier wins. The result is a single source of truth (SSOT) that gets distributed to all downstream systems (shop, marketplace, print).
Four steps from chaos to golden record
/ OUR APPROACH
01
Step 01
Matching with fuzzy logic
Before consolidation can happen, duplicates must be detected. "Adidas Sneaker" in system A and "Adi-das running shoe" in system B look different to classic match rules. Fuzzy matching uses confidence scores: based on attributes like EAN, manufacturer code, description patterns, and image similarity, the AI recognizes with 98 percent probability that it is the same product. What is above the threshold gets matched. What is below goes to manual review.
02
Step 02
Merging by best of breed
Once duplicates are identified, data gets merged. Best of breed means: per field, take the best source. Price from ERP (that is the accounting truth), image from supplier (that is where the hi-res originals live), marketing copy from CMS (that is where brand voice is maintained). The result is better than any single source.
03
Step 03
Survivorship rules
The rules deciding which source wins are configurable and traceable. Examples: "For prices: ERP beats supplier beats shop." "For copy: marketing beats supplier." "For technical data: supplier beats ERP." When no rule covers a conflict, the system creates a task for the data steward. The steward decides once, the decision is learned as a rule.
04
Step 04
Distribution as single source of truth
The finished golden record gets distributed to all consumer systems: shop, marketplace listings, print catalog, customer service. When an attribute changes in the source system, the pipeline runs again, the golden record updates, the change arrives consistently in all channels. Manual catch-up disappears.
Data does not disappear, it gets consolidated
A common concern: "If we merge records, do we delete information?" No. The golden record is an additional master layer, not a replacement. ERP, supplier feeds, and marketing tools stay in operation unchanged, sources remain traceable. Every value in the golden record is traceable back to its source (lineage). Whoever asks "where does this price come from?" gets a clear answer. We do not delete anything, we add an intelligent layer.
5, 20 %
duplicates in typical product master data
Before consolidation, product data in grown assortments typically contains 5 to 20 percent duplicates or quasi-duplicates. These are products maintained in two or three variants because different suppliers deliver them under different names or because deduplication was skipped during data migration. Systematic consolidation with fuzzy matching cleans 70 to 90 percent of these cases automatically, the rest gets clarified in manual review.
Example: the same sneaker in three systems plus golden record
Field
ERP
Supplier
Golden Record
Name
Adidas Sneaker X
Adi-das running shoe
Adidas Performance Runner (Marketing)
Price
89.00 € (wins)
84.00 €
89.00 € (ERP)
Color
Red
Dark red (wins)
Dark red (supplier)
Material
Textile
Mesh + synthetic (wins)
Mesh + synthetic (supplier)
Marketing copy
Empty
Datasheet style
From CMS (wins)
Three questions data stewards and CIOs ask
No. Original sources stay in operation unchanged. We create a new master layer on top that contains the consolidated truth. Every value in the golden record is traceable back to its source (data lineage), and you can adjust or roll back consolidation at any time. No data deletion, just a consolidated view.
The system creates a task for the data steward, showing all source values and a suggestion based on historical decisions. The steward decides once, the decision is stored as a new rule and applied automatically the next time a similar conflict appears. Manual upkeep shrinks over time because the system learns.
Depends on the need. For e-commerce assortments, consolidation typically runs as a nightly batch job so shops see the latest state in the morning. For time-critical cases (price changes during pricing campaigns), we can trigger the pipeline in real time so a change in the ERP arrives in the shop within seconds. Both are configurable.
Related deep dives
Data Governance
Survivorship rules are part of data governance. The governance layer defines which source has maximum authority for which field.
Attribute Mapping
Before consolidation, fields from different systems need to be comparable in the first place. Attribute mapping delivers that prerequisite.
// GOLDEN RECORDS
How many duplicates do you have?
Give us an excerpt of your product inventory from two or three source systems. We run our matching pipeline over it and show you concretely how many duplicates get detected, what the confidence scores look like, and how the consolidated golden record would shape up. Pseudonymized, in one week.