Master data management and data quality guide

Master data is the set of core records that a business reuses across its transactions and that change relatively rarely: item, bill of materials, supplier, customer, equipment and employee records. Master data management is the discipline that defines who creates and changes these records, under which rules, and how their quality is measured.

Production plans, cost calculations, quality records and maintenance work orders all rely on the same master data. A wrong unit or a missing lead time on an item record causes errors in every process from planning to purchasing. This guide shares the working framework the ORELSIS team uses on master data projects, with roles, rules, templates and checklists.

Master, transactional and reference data

Good governance starts with telling data types apart. Each type has a different change frequency, owner and control method. Rules built without this distinction are either too heavy or leave critical fields unprotected.

A practical rule: if a record is picked as a value in many transactions, it is master data; if it describes one event, it is transactional data. Reference data changes even less than master data and often comes from national or international standards (such as ISO 3166 for country codes and ISO 4217 for currencies).

Master data domains in manufacturing

In manufacturing companies, master data usually falls into the domains below. The critical attributes of each domain show which process suffers most when they are wrong, and they set the priority of quality rules.

Data ownership and stewardship roles

IT does not own master data; the business unit that uses the data and bears the consequences when it is wrong does. IT provides the platform and the technical controls. Writing roles down clearly removes the question “who will fix this record?”

• Data owner: The manager accountable for a domain's rules and quality. For item data, for example, the production planning or engineering manager.
• Data steward: The specialist who applies the rules day to day, reviews requests and closes quality issues.
• Data entry user: The person who opens a request or enters an approved record into the system.
• IT and data platform team: Builds validation rules, workflows, permissions and integrations.
• Data governance board: Approves cross-domain decisions, standard changes and priorities.

R: responsible, A: accountable, C: consulted, I: informed. Each row should have exactly one A; more than one A means accountability is diluted.

Naming and coding standards

The coding standard gives a record its identity; the naming standard helps people find it. Packing too much meaning into a code forces you to restructure codes when the product structure changes. So prefer short, stable codes with limited meaning, and keep detailed information in classification fields.

Item code structure template

This template produces codes like “RM-MET-000245”. Changeable information such as colour, size or supplier goes into attribute fields, not into the code. Adapt the template to your own classification; what matters is that the rule is written down and enforced by the system.

Data quality dimensions and sample rules

Data quality is not a single number; it is measured across several dimensions. Writing measurable rules for each dimension turns vague complaints like “the data is bad” into fixable issues. The table below shows starter rules for manufacturing master data.

Completeness, validity and uniqueness rules can usually be checked automatically and should be blocked at entry. Accuracy often requires field verification, counts or comparison with actual consumption, so it goes into a periodic control plan.

Profiling and cleansing existing data

Before writing rules, you need to measure the current state of the data. Profiling is the analysis that systematically reveals gaps, format errors, duplicates and outliers in the data. Its results determine the size and priority of the cleansing work.

Profiling steps

1. 01
   Choose the scope

   Pick one domain (for example items) and one system. Separate active and inactive records; flag records used in recent transactions.

2. 02
   Copy the data

   Move the data to a separate analysis environment so the live system is not affected. Verify the copy by comparing record counts with the source.

3. 03
   Compute field statistics

   For each field, compute fill rate, distinct value count, most frequent values, length and format distribution.

4. 04
   Apply the rules

   Run the quality rules on the data and list violations record by record.

5. 05
   Find duplicate candidates

   Group similar records by description, tax ID or manufacturer part number.

6. 06
   Review results with owners

   Present findings to the data owner and agree on fix priority and who is responsible.

Deduplication and cleansing

Duplicate records split inventory, distort planning and scatter purchasing power. Deduplication does not end with finding candidate matches; you also need to decide which record survives, how open transactions move and how the old record is closed.

1. 1.Define match keys: tax ID, manufacturer part number, normalised description.
2. 2.Normalise descriptions: make case, spacing, abbreviations and unit spellings uniform.
3. 3.Score candidate matches and separate certain matches from uncertain ones.
4. 4.Have the data steward review uncertain matches; don't merge them automatically.
5. 5.Choose the surviving record and move open orders, stock and BOM links from the others.
6. 6.Don't delete the old record; deactivate it and reference the new record.
7. 7.Log every merge with date, person and reason.

Golden record and create/change workflow

Golden record and source of truth

The golden record is the single, correct version of an object accepted across all systems. The source of truth is the system where that record, or a given field, is authoritatively created. Pick one source of truth for each domain and feed the other systems from it.

• Item records are usually created in the ERP; MES and quality systems read them.
• The design BOM can be managed in a PLM or engineering system and the manufacturing BOM in the ERP; the conversion rule between the two must be written down.
• Equipment records can be managed in the maintenance system; the finance system links to them by asset number.
• Different sources can be chosen per field; for example, a supplier's bank details in finance and its quality approval status in the quality system.

A data platform that combines several sources, such as OREL DM, can produce the golden record by running match and merge rules centrally. Whatever tool you use, the decision “which field comes from which system” must be written down and approved.

Create and change request workflow

When master data changes are left uncontrolled, quality drops quickly. A controlled request workflow makes sure every record is created according to the rules and every change is traceable.

1. 01
   Request

   The user opens a request on a standard form; it cannot be submitted without required fields and a reason.

2. 02
   Automatic checks

   The system runs format, required-field and possible-duplicate checks; if a similar record exists, it shows it to the requester.

3. 03
   Steward review

   The data steward checks classification, description and code structure; if anything is missing, the request goes back with a reason.

4. 04
   Domain approvals

   Each required department completes its own fields: purchasing the sourcing data, cost accounting the valuation data, quality the inspection plan.

5. 05
   Activation

   The record is activated and distributed to connected systems; the distribution result is checked.

6. 06
   Log and feedback

   The requester is notified; request lead time and rejection count are written to the KPIs.

Measurement and governance cadence

Master data quality degrades over time if it is not measured. A simple KPI set reviewed regularly works better than a long report. Break KPI values down by domain and data owner so errors are visible and owned.

• Weekly: data stewards review open issues and pending requests.
• Monthly: data owners assess the KPIs and assign priority fixes.
• Quarterly: the governance board approves standard changes, new rules and domain extensions.
• Yearly: the coding standard, roles and permission matrix are reviewed end to end.

Personal data in master data and KVKK

Employee, customer contact, sole-proprietor supplier and individual customer records contain personal data. In Türkiye these fields should be assessed under Law No. 6698 on the Protection of Personal Data (KVKK). Flagging personal data when you design the master data model is much easier than fixing it later.

• Flag fields containing personal data in the data inventory and write down the processing purpose.
• Collect only the fields needed for that purpose; don't add fields “in case we need them later”.
• Apply role-based access to sensitive fields such as national ID, contact details and health data.
• Mask personal fields in data copied to test and training environments.
• Define retention periods and the deletion or anonymisation method at the end of each period.
• Log access to and changes of personal data.

Common mistakes

• Treating master data as an IT project and not giving business units ownership.
• Packing too much meaning into codes and having to restructure them when the structure changes.
• Starting bulk cleansing before writing rules, so the same errors return a few months later.
• Leaving direct record editing rights open to many users.
• Deleting duplicate records and breaking open transactions and historical reports.
• Not choosing a single source of truth per domain and updating systems by hand in both directions.
• Measuring quality KPIs once and not tracking them regularly.
• Copying fields with personal data to test environments without masking.

Where to start

Pick one domain with high business impact, assign the data owner and steward, profile the existing data and write the first ten rules. Then launch the request workflow and track the KPIs weekly. A small model that runs continuously delivers more lasting results than a large one-off cleanup.