Blog · 16 November 2026 · Paulo Cordeiro

Material attribute dictionary: how to build one by family

How to build a material attribute dictionary by family from your existing data and use it as the yardstick for cleansing, with a sheet and an example.

A material attribute dictionary is the set of definition sheets that states, for each material family, which characteristics an item has, how each one is written and which values are accepted. You build it from your own data: extract attributes from the descriptions and part numbers you already have, agree on them with engineering, maintenance and procurement, and test them on a sample. Only then does it make sense to start fixing records one by one.

A familiar scene in a cleansing project: the team gets the list of bearings and starts fixing descriptions by hand. One analyst writes "ID 25", another writes "BORE 25MM", a third writes "INNER DIA 25". The seal type shows up as "2RS", "SEALED", "W/ SEALS" and "RUBBER".

A few weeks later the data is "fixed", but in three different styles. Everyone did their job well; nobody had agreed in advance on what "right" looked like. That is the problem a dictionary solves.

There is no need to reinvent the wheel. ECLASS calls each product characteristic a property ("A property describes a characteristic of a product") and requires each one to have a preferred name, a definition that explains the property without repeating the name or including units, a data type and, for measures, a unit. The full set of accepted values for a property is called the value list (ECLASS, Property). Your company's dictionary can follow the same logic, whether or not you use ECLASS.

What is a material attribute dictionary?

It is the official reference for the technical characteristics of each material family. For the family "deep groove ball bearing", for example, it says the attributes are bore diameter, outside diameter, width, seal type and internal clearance, and for each attribute it holds a sheet with the definition, data type, unit, accepted values and the person accountable for it.

SAP classification follows the same order: you create the characteristics first, then the class that groups them. The SAP Learning lesson uses classes for nuts and screws, with characteristics such as length and thread type (SAP Learning). The dictionary is the well defined "characteristic" layer, before any system comes in.

In Brazilian public procurement, the Ministry of Health cataloging unit (CATMAT is the federal government's materials catalog) organizes items into Group, Class, descriptive standard and item code, and defines in descriptive standards, such as the one for medicines, which characteristics are mandatory and which are optional, with value choices supported by synonym tables (Brazilian Ministry of Health, CATMAT).

How is an attribute dictionary different from a description standard?

The dictionary defines each attribute on its own: what it means, how it is measured and which values exist. The description standard uses those attributes to build the item text: noun, modifier, attributes in an agreed order, unit.

One feeds the other. If the dictionary says seal type accepts "SEALED" and the controlled abbreviation is "SLD", the short description comes out the same on every item. The writing rules are covered in our guide on how to write material descriptions; this article is about what comes before.

Then there are classification codes. In Brazil, companies often keep a descriptive standard for materials (PDM) and for services (PDS), plus NCM (the Brazilian goods classification code, for materials and products), NBS (the Brazilian services classification code) and CEST (a Brazilian tax code tied to certain goods), along with the global UNSPSC. These belong on the record as classification attributes to be checked, separate from the technical attributes in the dictionary. They depend on the technical description, but they do not describe the item; they place it in a category. Keep them in a separate block and check that they match the description, as shown in our NCM review of the material master.

Which fields does each attribute need in the dictionary?

The table below shows the definition sheet for one attribute. The example is hypothetical: family "deep groove ball bearing", attribute "seal type" (and "bore diameter" when the field is numeric).

Sheet fieldWhat it is forExample (hypothetical family)Common error if missing
Preferred nameGive the attribute a single name across screens, reports and integrationsSeal typeThe same attribute shows up as "protection", "shield" and "cover" in related families
DefinitionExplain what the attribute is, without repeating the name or mentioning unitsKind of side protection that keeps contaminants out and lubricant inEach analyst reads it differently and fills the same field with different things
Data typeState whether the value is a number, a list value, yes/no or a dateList value (bore diameter: number)Numbers stored as text, which cannot be sorted or filtered by range
Unit of measureLock the unit for measured attributesBore diameter in mm"25" with no unit, which could be mm or inches
Allowed values listClose what can be enteredOPEN; SHIELDED; SEALEDFree values that grow with every new record
Mapped synonymsConvert legacy and supplier spellings to the official value2RS, 2RSH, RUBBER → SEALED; ZZ, 2Z → SHIELDEDThe conversion is redone by hand on every batch, with different results
Format and decimalsStandardize how numbers are writtenBore diameter as an integer, no decimals"25", "25.0" and "25.00" treated as different values
Included in short description, and positionState whether the attribute goes into the short text and whereYes, after the dimensionsDescriptions with the same data in different orders
Controlled abbreviationDefine the accepted short form when space runs outSEALED → SLDEveryone abbreviates differently: SL, SEAL, SLDD
Source that confirms the valuePoint to where the data can be checkedManufacturer catalog or designation suffixValues entered by guesswork, with no way to audit later
OwnerName the team that approves changes to the attributeMaintenance engineeringNew value requests stall because nobody feels authorized to decide

The arrow in the synonyms row means "converts to". SAP offers similar fields on a characteristic: NUM format, up to 15 characters, with unit and decimal places; CHAR format, up to 30 characters (up to 70 with extended fields in Customizing); optional allowed values, which become binding in classification unless an indicator allows additional values (SAP Learning). If your ERP has limits like these, the dictionary has to respect them from version 1.

How do you build the dictionary from the data you already have?

Do not start from a blank page. Your current material master already holds almost everything; it is just scattered.

1. Pick a family and pull everything it has

Export short description, long description, purchase text, part number, manufacturer and unit of measure for every item in the family. Include blocked items: they show old spellings that will come up during cleansing.

2. Mine the descriptions and part numbers

Split the text into terms and count how often each one appears. Numbers followed by "MM", an "X" between dimensions, designation suffixes and repeated words point to candidate attributes. In the bearing example, "25X52X15" reveals three dimensions; "2RS" and "C3" reveal seal type and clearance.

3. Group synonyms and variants

Bring together terms that mean the same thing: "SEALED", "2RS", "W/ SEALS", "RUBBER". Each group becomes one official value and one row of synonyms. Terms that fit no group go to a questions list.

4. Decide with the people who use the item

Bring a ready proposal, not an open question. Engineering validates definition and unit, maintenance says what it needs to find the item during a shutdown, procurement says what suppliers provide and how they quote. Each attribute leaves the meeting with a name, values and an owner.

5. Write the sheets

Fill in the fields from the table above for every attribute. If a definition has to repeat the name to make sense, it is not clear yet.

6. Test on a sample

Take a sample of items from the family, with good and bad descriptions, and fill in the attributes using only the dictionary. Whatever does not fit tells you what is missing: a value, an attribute, a synonym or a format rule.

7. Publish version 1

Freeze the family dictionary with a version number and date. From then on, changes go through a formal request only.

Hypothetical example: deep groove ball bearing

The example below is hypothetical, created to illustrate the method.

  • Bore diameter: integer, mm, included in the short description (1st position).
  • Outside diameter: integer, mm, included in the short description (2nd position).
  • Width: integer, mm, included in the short description (3rd position).
  • Seal type: list OPEN; SHIELDED (synonyms ZZ, 2Z, METAL); SEALED (synonyms 2RS, 2RSH, RUBBER). Abbreviation SLD for SEALED. Included in the short description (4th position).
  • Internal clearance: list NORMAL (synonym CN); C3. Included in the short description (5th position).
  • Cage material: list PRESSED STEEL; BRASS. Long description only.

Legacy description in the system: "BRG 6205 2RS C3 25X52X15 CAGE POLYAMIDE". Broken down by the dictionary:

  • Bore diameter: 25 (accepted)
  • Outside diameter: 52 (accepted)
  • Width: 15 (accepted)
  • Seal type: 2RS → SEALED (converted by synonym)
  • Internal clearance: C3 (accepted)
  • Cage material: POLYAMIDE (not on the list, goes to the decision queue)
  • "6205": manufacturer designation, goes to the part number field, not to an attribute

The attribute owner decides whether POLYAMIDE becomes a new value in version 2 or whether the record is wrong. Meanwhile, the item moves on with the other attributes fixed and an open issue logged.

How does the dictionary become the yardstick for cleansing?

Once the dictionary is published, fixing data stops being a matter of opinion. Every legacy description goes through the same yardstick and lands in one of four outcomes:

  • Accepted value: it is on the list or in the right format. Save it.
  • Known synonym: convert it to the official value. Save it.
  • Value not on the list: send it to the attribute owner's decision queue.
  • Missing attribute: send it to enrichment, using the source named on the sheet.

The decision queue is what changes the pace of the project the most. Instead of each analyst settling a doubt alone, identical questions are grouped and the owner decides once for all items. Each decision goes back into the dictionary as a new value or a synonym, so the yardstick gets better with every batch.

With attributes split out, identical items written in different ways show up side by side, which makes it easier to follow the steps to identify duplicate materials. To check the other fields on the record, use the material master data cleansing checklist.

How do you keep the dictionary alive instead of a forgotten spreadsheet?

A dictionary that never changes goes stale within months: new suppliers arrive, new technology appears, engineering changes a spec. Three things keep it in use:

  • Version, date and history. Each change records what changed, who asked, who approved and when. A value that falls out of use is flagged as obsolete, not deleted, because older items still use it.
  • Request for a new attribute or value. A simple form: family, attribute, proposed value, supporting source, requester and reason. The data steward does not create values on their own; they request them.
  • Owner with a response deadline. Each attribute has an accountable team and an agreed deadline to decide. Without one, people fall back on free text so the purchase is not held up.

It helps a lot when the dictionary lives where records are created, in the MDM or the ERP, not in a side spreadsheet. When the value list is on the form, the wrong value never gets typed. Deciding which attributes are mandatory in each family is a separate step, covered in our post on product data standardization by category.

Where to start

  1. Pick a family with meaningful volume and frequent complaints from procurement or maintenance.
  2. Export descriptions, part numbers and manufacturers for the family and mine the recurring terms.
  3. Draft the proposed attributes, values and synonyms, and take them to a short meeting with engineering, maintenance and procurement.
  4. Write the sheets using the fields in this article's table and test them on a sample.
  5. Publish version 1, name an owner for each attribute and open the channel for new value requests.
  6. Only then run cleansing for the family, with the decision queue working from the first batch.

Frequently asked questions

How many attributes should a family have?

As many as you need to tell one item from another and to buy the right one. Start with those that appear in good descriptions and in purchase orders. An attribute nobody fills in or looks up can wait for a later version.

Should we use ECLASS or UNSPSC as is, or build our own dictionary?

Public standards are a good starting point for names, definitions and data types. ECLASS, for example, publishes properties with definitions and value lists. Even so, the final call on what goes in belongs to the people who use the catalog. Many companies use a standard as a reference and adapt it to their own needs.

Who approves a new value?

The attribute owner named on the sheet. Usually engineering for technical attributes, maintenance for application attributes and procurement for commercial ones. The master data team coordinates the request and updates the dictionary.

What do we do with a legacy value that is not on the list?

Send it to the decision queue with the item and its original description. The owner picks one of three paths: create the value, map it as a synonym of an existing value or fix the record.

Can the dictionary live in a spreadsheet?

For the first version, yes. For day to day work, the value lists should live in the MDM or the ERP, on the creation form, so that a value outside the list is blocked at entry.

Do we need the dictionary before deduplication?

It helps a lot. With standardized attributes, comparing items becomes objective. Without a dictionary, deduplication depends on comparing free text, which means more doubt and more manual review.

How 4MDG can help

4MDG standardizes and enriches material master data on an MDM platform, with automation and AI. Each family gets a characteristics structure defined with the people who use the catalog, with a template of mandatory and optional attributes and short and long descriptions built field by field. See how material standardization (PDM) works or request a diagnostic of your catalog. If you would rather talk first, get in touch.

Sources

Have a question or want to talk about this?

Talk to a 4MDG specialist — we reply through the channel you prefer.

The international code is added when you save.