empty filled in SAP found by enrichment

Engineering solutions for data governance

AI engineering that brings order to enterprise master data

We combine master data and SAP MDG expertise with data engineering and large language models. We enrich attributes, normalize values, find duplicates and analogs, and check new records before they are created — with a source for every value and a measurable result for every class.

  • 57% → 86%required attributes filled
  • 7.6×more complete records: 1,721 instead of 226
  • 37kvalues from manufacturer catalogs, each linked to its page
  • 3.2×more duplicate and analog pairs than SAP data alone reveals

Case study: electrical MRO catalog of an industrial holding — 8,008 items in 44 SAP MDG classes.

What we do

Data governance, end to end

From the class model to the data platform underneath it: we design the rules, build the engineering and measure the result on your data — not on a demo.

Master data & classification

Class and attribute models, value formats and mandatory fields — the structure that makes items comparable.

  • Class and attribute design
  • Naming rules
  • Change request governance

SAP MDG & ERP

How classification lives in your master system and the systems around it.

  • Classes and characteristics (CT04, CL02)
  • MDG change requests
  • Mass processing and import

AI enrichment you can audit

Models find and verify values instead of inventing them.

  • LLMs with web search
  • Part designations decoded, values calculated by standards
  • Source and confidence for every value

Duplicates & analogs

One catalog to buy, store and repair by — across classes where the same part was set up differently.

  • Duplicates, analogs, contradictions
  • Catalog unification
  • Checks before a new material is created

Data quality & monitoring

Rules, metrics and quality control at every step of the data lifecycle.

  • Canonical value keys
  • Validation at entry
  • Completeness per class, before and after

Data platforms & AI infrastructure

Storage, formats and search for large data — and language models that run on your own servers.

  • Cheaper analytics storage: Druid, Elasticsearch
  • Columnar formats and stream compression
  • Compact RAG stores and on-premise LLMs

Principles

Results you can verify

You see where every value came from — and you decide what goes into your master system.

  • Traceability

    Every value has a source and a confidence: SAP, item name, designation, standard or a catalog page.

  • The master record stays intact

    SAP values are never overwritten. A disagreement is logged as a conflict, with the alternative and its source alongside.

  • Experts decide

    AI prepares proposals and a review queue; your engineers and data stewards approve.

  • Measurable

    Attribute and record completeness before and after — for every class.

  • Inside your perimeter

    The model fits your security requirements: deployed on your infrastructure, or external models where they are allowed.

Case study · Industrial holding

Structuring an electrical MRO catalog

The holding needed a materials catalog people could search, compare and buy from. Two procurement groups — electric motors and wiring accessories — held 8,008 items in 44 SAP MDG classes, yet only 226 of 3,941 active records were complete.

One motor, three spellings

  • El.motor AIR100 S4 U2 3/1500
  • El.motor AIR100S4 3kW/1500 IM3001
  • Electric motor AIR-100S-4-U3 3kW comb.

Item names as found in the source catalog, transliterated.

  • 6%of active records complete
  • 57%of required attributes filled in SAP
  • 3,185placeholders instead of values, like “NOT SPECIFIED”

Required attributes filled, by class

In SAP After enrichment

Show as table
ClassIn SAP, %After enrichment, %
All 44 classes5786
AC electric motors7397
Wiring accessories4072
Cable lugs and sleeves5086
Plug connectors6189
Insulating tubing5986
Insulating tape6688
Batteries6494

01 · Enrichment

Attributes filled, every value sourced

  • Extended the class model: 375 new attributes on top of 368 existing ones
  • Filled values from item names, part designations, standards and manufacturer catalogs
  • Brought values to each class’s format and replaced 1,982 placeholders
  • Kept a source and a confidence score for every value
86%
required attributes filled; 57% in SAP
1,721
complete records instead of 226
37k
catalog values, each with a link
32
conflicts with SAP sent for decision

02 · Duplicates & analogs

Three times more matches

  • Compared all 3,941 active items on normalized attributes — within and across classes
  • Recovered frame size, IE class, cross-section and bolt size from item names
  • Handed engineers a report with a reference record per group and a decision field
Pairs of items found
Match typeAfter enrichmentFrom SAP data
Duplicates249
Likely duplicates15954
Analogs3,002928
Contradictions5513
Total3,2401,004

03 · New records

Repeats caught before creation

  • Parsed attributes of 49 materials created between 1 September and 19 October 2026
  • Compared each with every item in its class, including those marked for deletion
  • Suggested the class from the item name alone, without a language model
What the 49 new materials turned out to be
Already in the catalogMaterials
The same item2
An active analog25
An analog marked for deletion1
No match — creation justified21
28 of 49
already had the same item or an analog
98%
right class in the top three suggestions; 80% first

Client name withheld. Figures cover active items; SAP snapshot of 23 September 2026.

How we work

From catalog audit to everyday operation

  1. Audit

    • Snapshot of classes, attributes and materials from SAP; value profile
    • As-is report: completeness, placeholders and duplicates per class
    • Goals and metrics that define the pilot’s success
  2. Design

    • Missing attributes, each with a rationale
    • Normalization rules: formats, units, canonical keys
    • Governance: who approves values, duplicates and new attributes
  3. Engineering

    • Enrichment from names, designations, standards and the web
    • Duplicate and analog matching with reports for engineers
    • Steward assistant and integration with SAP MDG
  4. Operation

    • Entry control: class suggestion and checks for new records
    • Completeness monitoring per class
    • Re-enrichment after every new SAP snapshot

Start with a pilot on one procurement category

  1. Snapshot & auditAs-is report for every class
  2. Class modelMissing attributes, agreed with your data stewards
  3. Enrichment & duplicatesValues with sources; reports for engineers
  4. Assistant & entry controlFor the pilot’s users
  5. Back to SAP MDGTest system first, then production

What we need from you

  • A catalog export: classes, attributes and materials
  • A data steward and an engineer to review decisions
  • Access to a test SAP MDG system
  • Your security requirements for the choice of model

What you get

  • Before-and-after metrics for every class
  • A report on duplicates, analogs and contradictions
  • A steward assistant running on your data
  • A rollout plan for other procurement categories

Contact

Let’s look at your data

Tell us about your catalog or data platform. We’ll suggest where a pilot pays off first and what it takes to run it.