// DEEP DIVE | PRODUCT DATA ENRICHMENT
Automated Attribute Mapping and Schema Matching
Attribute mapping is the translation layer between systems. Every supplier ships their data in their own structure, every target system expects it differently. Where teams spend hours mapping and normalizing manually, AI-driven schema matching recognizes the structure and maps autonomously. "Color: Crimson" from the supplier becomes "color_code: #Red" in the shop, automatically and traceably.
THE PROBLEM
When every supplier speaks their own language
You are onboarding a new supplier. They ship an Excel file with 87 columns in their own logic: "art_nr," "bezeichnung_lang," "farbcode_intern," "vk_brutto," "gewicht_g." Your PIM calls the same fields "sku," "product_name," "color," "price_eur," "weight_kg." For the Excel to turn into something that fits the PIM, someone needs to map every single column. Across 50 suppliers, that is a full-time job.
Worse, the values themselves. The supplier writes "crimson," "bordeaux," and "wine red." Your shop only knows "red." Weight comes in grams, the shop wants kilograms. Prices include VAT, your system calculates net. Every one of these conversions has to happen somewhere, otherwise nonsense lands in the product database.
The solution is AI-driven mapping that solves schema (which column goes where) and values (what does this label mean) at the same time. What takes 2 hours of prep for the first supplier takes 5 minutes for the 50th, because the system learns from every file.
What is schema matching?
Schema matching is the AI-driven process that automatically recognizes which columns in a source file correspond to which fields in a target system. Unlike a hard-coded import, the system uses pattern recognition (column names, data types, sample values) to detect that "art_nr" is an SKU and "vk_brutto" is the gross price. This is complemented by value mapping ("crimson" becomes "red") and transformations as ETL steps (weight from grams to kilograms, prices from gross to net). All of this runs on a normalized pipeline so a new supplier format no longer requires a code change.
Four layers of translation logic
/ OUR APPROACH
01
Step 01
Schema matching with AI
The AI receives the source file and the target schema, proposes a column mapping with confidence scores. "art_nr" maps to "sku" with 98 percent confidence, "vk_brutto" to "price_gross" with 95 percent. High-confidence matches get accepted automatically, uncertain ones go to manual review. Per supplier format the mapping is set up once and stored as a template.
02
Step 02
Value mapping through lookup tables
Schema mapped is only half the work. Values need normalization. We build lookup tables: "crimson," "bordeaux," and "wine red" all map to the target value "red." These tables get initially proposed by AI and confirmed by the data steward. In operation, new variants come in as suggestions, the steward clicks them through.
03
Step 03
Transformations and calculations
Some fields need calculation, not just mapping. Weight of 1500 grams becomes 1.5 kilograms. Price of 119 euros gross at 19 percent VAT becomes 100 euros net. Date "15.03.2026" becomes "2026-03-15." These transformations run as ETL steps (Extract Transform Load), via regular expressions (regex) for pattern transformations or script functions for more complex cases.
04
Step 04
Supplier onboarding as a repeatable process
The first supplier costs a few hours of setup. The second costs half because the system has seen the fields before. By the tenth supplier, a new onboarding is a matter of minutes: upload file, confirm system suggestions, done. What used to be a time-intensive integration project becomes a routine operation in daily work.
The Babel fish for databases
Imagine every source system speaks a different language and your target system only understands one. Attribute mapping is the Babel fish that sits in between: reads the foreign language, knows the idiomatic quirks, translates by your house rules. What needs a dictionary the first time (lookup tables, manual confirmation) becomes a learned reflex (templates, automatic recognition). Supplier onboarding turns from bottleneck into routine.
Minutes
instead of hours for a new supplier format
Classic supplier onboarding with manual mapping of an 80-column Excel costs an integration manager 2 to 4 hours of prep before the first file lands cleanly in the PIM. With AI-driven schema matching and existing lookup tables from earlier suppliers, that drops to 5 to 30 minutes. Scaled to 50 suppliers per year, that is weeks of saved integration work.
Where the work disappears
Step
Manual Mapping
Xanevo AI Mapping
Column matching
1 to 2 hours per supplier
Suggestion with confidence score, manual confirm
Value normalization
Manual upkeep per value
Lookup tables with AI suggestions
Unit conversions
Excel formulas per file
ETL steps defined once, reused
New supplier format
Full new mapping pass
Template if format known; minutes instead of hours
Ongoing upkeep
High, no learning effects
System learns from every confirmation
Three questions integration managers and IT architects ask
If the AI finds no clear mapping with sufficient confidence, the column gets flagged as "unclear" and the integration manager assigns it manually. The manual decision is stored as a new rule, so the next supplier with a similar column name gets the mapping detected automatically. The system learns from every decision.
Yes. Existing rules from Talend, Informatica, AWS Glue, or custom ETL scripts can be imported as a starting point. We build the AI layer on top, which checks new suppliers against existing rules and only flags the differences. Existing investments are not lost.
Every mapping lives as a versioned configuration file (YAML or JSON), traceable and integrable into your Git repository. Changes are diffable, rollbacks possible. Lookup tables are maintained through a UI so non-technical data stewards can add or correct values without changing code.
Related Deep Dives
Golden Records
Attribute mapping is the prep work for golden records. Only when fields are comparable can survivorship rules kick in.
Taxonomy and Classification
Mapping between standards (ETIM to ECLASS, or supplier-specific categories to industry taxonomies) is a special case of attribute mapping.
// ATTRIBUTE MAPPING
We clean up your CSV in seconds.
Give us a typical supplier CSV or Excel and the schema of your PIM or ERP. We run our mapping pipeline over it and show you concretely which fields get detected automatically, where lookup tables would kick in, and where manual decisions remain necessary. Pseudonymized, in one week.