What is MDM: the master data management that underpins the ERP

MDM (master data management) is the discipline that ensures a single, correct and shared record of customer, supplier, material, product and service across the entire company. It is not an ERP module: it is process, roles, quality rules and technology that validate data before it enters SAP, TOTVS or Oracle.

In summary

• MDM is the discipline that creates and maintains the single record of the entities used by the entire company, with a clear definition of who approves, who corrects and which rules apply before the record is written to the ERP.
• According to Gartner (2021), poor data quality costs organizations an average of 12.9 million dollars per year, and much of that loss originates in duplicate, incomplete or non-standard records.
• The ERP executes transactions and does not judge the quality of master data: purchase order, invoice and payment inherit the error that the record allowed through.
• With the tax reform established by Constitutional Amendment 132/2023 and detailed by Complementary Law 214/2025, poorly registered customers, suppliers and products begin to compromise the calculation of IBS and CBS.

What is MDM (master data management)

MDM (master data management) is the set of processes, roles and technology that maintains a single, accurate and shared version of the company's master data. DAMA-DMBOK (2017) defines master data management as the control over master data values and identifiers in order to maintain a consistent, accurate and shared version of the truth.

Gartner (2024) describes MDM as a discipline in which business and IT work together to ensure uniformity, accuracy, stewardship (the ongoing curation of data by a named owner), semantics and accountability over shared master data assets. In other words: MDM is not an isolated piece of software, it is a management agreement supported by software.

Master data, transactional data and reference data

DAMA-DMBOK (2017) separates master data, which describes entities such as customer, supplier and material, from transactional data, generated by events such as purchase orders and invoices. Reference data is the standardized list that feeds both, such as unit of measure, NCM and country code.

In ERP practice the difference becomes visible. According to SAP (2024), the MARA table stores general material data, KNA1 general customer data and LFA1 general supplier data. In TOTVS, SB1 plays an equivalent role for material. These records are master data: they exist before the transaction and are reused thousands of times.

The purchase order and the invoice are transactional. They are born inheriting the master record. If the material's unit of measure is wrong in MARA, the requisition buys the wrong quantity. If the supplier is duplicated in LFA1, the payable goes to the wrong CNPJ and the payment stalls. The error does not appear in the record, it appears in the next process.

Master data domains

MDM organizes the work by domain. The most common ones in Brazilian companies are customer, supplier, material, product, service and, in industrial operations, asset and cost center. Each data domain has an owner, its own rules and a lifecycle. Supplier requires approval and due diligence; material requires descriptive standardization; customer requires tax and credit validation.

Why MDM is not an ERP module

The ERP is excellent at executing processes and terrible at preventing bad records. It validates format, not meaning. It accepts "hex bolt 1/2" and "HEXAGONAL BOLT 1/2 IN" as two different materials. SAP (2023) unified customer and supplier entry into the S/4HANA Business Partner, replacing transactions XD01 and XK01, which improves the structure but does not replace the request, approval and quality rule flow that runs before the record is written.

Why MDM matters in money and in risk

MDM matters because bad master data becomes a recurring cost, not an isolated incident. According to Gartner (2021), poor data quality costs organizations an average of 12.9 million dollars per year. The money disappears in emergency purchases of items that already existed in stock under another description, in duplicate payments, in reconciliation rework and in rejected invoices.

There is also regulatory risk. Law 13.709/2018 establishes, in article 6, item V, the quality principle, requiring data that is accurate, clear, relevant and up to date according to the need of the processing. It is not a recommendation: it is a legal obligation applied to customer records, contacts and individuals in supplier bases. The same Law 13.709/2018 provides for fines of up to 2% of revenue in Brazil, capped at 50 million reais per infraction.

Tax risk has increased

Tax classification depends on the master record. According to the Receita Federal (2022), the Mercosur Common Nomenclature has eight digits, is based on the Harmonized System and is mandatory in the tax classification of goods reported on the electronic invoice. ICMS Agreement 142/2018 determines that the Tax Substitution Specifier Code, with seven digits, must identify goods subject to tax substitution on the tax document.

Constitutional Amendment 132/2023 created the IBS, under state and municipal jurisdiction, and the CBS, federal, replacing ICMS, ISS, PIS and Cofins. Complementary Law 214/2025 institutes the IBS, the CBS and the Selective Tax, detailing rules for registration, classification of operations and assessment. In 2026 the test year takes effect, with CBS at 0.9% and IBS at 0.1%, offsettable against PIS and Cofins.

The practical effect is direct: a company with multiple CNPJs and inconsistent customer, supplier and product records cannot correctly assess the new taxes. Those who have already organized their master data enter the transition period with an advantage, as we showed in the material on tax reform.

AI depends on reliable master data

AI does not magically fix bad data. A demand forecasting model trained on a base with the same material registered five times will forecast five fragmented demands. A purchasing agent querying a base with duplicate suppliers will choose the wrong commercial terms. Master data governance is the foundation for AI to work, not the other way around.

How to implement MDM step by step

1. Define domains and scope

Choose where to start, usually supplier or material, which are the domains with the most visible financial loss. Delimit which companies, plants and systems enter the first wave. An overly broad scope makes the project die halfway.

1. Appoint a data owner and a data steward

The data owner is the business manager accountable for the definition and the rules of the domain, such as the supply manager for supplier. The data steward is the one who performs daily curation, analyzes requests and corrects deviations. Without a name and a badge, governance never leaves the slide.

1. Write the standard before buying technology

Define taxonomy and data dictionary: the way to organize master data so that procurement, logistics, maintenance, tax and IT speak the same language. ISO 8000-110 (2021) defines requirements for master data exchange, including syntactic conformance, reference to a data dictionary and explicitly encoded semantics. Use this as a reference for descriptive standards.

1. Clean up the existing base

Before shielding the entry door, clean up what is already inside. Duplicate identification by similarity, attribute enrichment, descriptive standardization and blocking of inactive records. This work of data cleansing is what turns the current base into a reliable starting point.

1. Design the record lifecycle

Map request, automatic validation, approval by authority level, creation in the ERP, update, blocking and reactivation. Each stage needs a deadline, an owner and an audit trail. It is this flow that prevents records created "in a rush" to release an urgent order.

1. Automate validation with external sources and AI

Automatic queries to the Receita Federal database to check the CNPJ and its registration status before saving the record prevent unfit suppliers in the base. Normative Instruction RFB 2.229/2024 established the alphanumeric CNPJ, with the new numbers scheduled for issuance in July 2026 and the existing numeric CNPJs maintained, which requires prepared fields and validations in all systems.

1. Choose the architecture model and integrate

Decide between registry, consolidation, coexistence and centralized according to maturity and the number of systems involved. The decision about capabilities and architecture determines whether the MDM merely sees the data or whether it governs creation in the ERP.

1. Measure, publish and correct

Define quality indicators by domain, publish the results to the business areas and treat deviations as incidents. Governance without measurement becomes bureaucracy. Measurement without consequence becomes an ignored report.

The table below compares the four MDM architecture models, to help you choose the starting point according to the number of systems and the desired level of control.

Model	How it works	When it makes sense	Main limitation
Registry	The MDM keeps only the identifiers and points to the data that remains in the source systems	Many legacy systems and a quick need for a single view	Does not correct the data at the source, it only links records
Consolidation	Data is copied to a central repository that generates the golden record for analysis	Basis for BI, management reporting and consolidated tax assessment	The ERP continues receiving records without prior control
Coexistence	The golden record is built in the MDM and returned to the source systems, which keep operating	Companies with multiple ERPs or several units and CNPJs	Requires well-monitored bidirectional integration
Centralized	Every record originates in the MDM, with workflow and rules, and only then is written to the ERP	Operations that need strong compliance and authority-level control	Depends on process discipline and buy-in from the business areas

Practical example

A Brazilian consumer goods manufacturer with four CNPJs runs on SAP S/4HANA. Supplier records were created by three different teams, each one creating the Business Partner in a rush to release that day's order.

Before. The same maintenance provider appeared three times: one record with the CNPJ typed with punctuation, another without, and a third with one digit swapped. Finance paid into two different accounts, the tax area could not consolidate withholding, and the contract negotiated with a volume discount was never applied, because the volume was fragmented. In the material master, MRO items had free-form descriptions, and the analysis identified the same bearing under four codes, with idle stock at one plant and emergency purchases at another. In finished products, three SKUs had divergent NCMs for the same merchandise, a direct risk to the classification required by the Receita Federal (2022) and to the assessment in the 2026 test year, with CBS at 0.9% and IBS at 0.1% provided for in Complementary Law 214/2025.

What changed. The company rolled out a request workflow with tiered approval before writing to SAP. CNPJ validation now queries the public database automatically, rejecting any irregular registration status. The AI embedded in the software began detecting duplicates by similarity at the moment of the request and standardizing material descriptions through category templates with mandatory attributes. A procurement data owner and data stewards appointed by domain took over curation.

After. Duplicate suppliers stopped entering the database, payments stopped getting stuck over CNPJ discrepancies, buyers began finding the existing item before opening a requisition, and the tax team gained consistent NCM codes by product family. The ERP stayed the same. What changed was what happens before the record reaches it.

Common mistakes

• Treating MDM as an IT project. Without a data owner in the business area, no one decides the rule and everything turns into a ticket. Appoint owners by domain before the first line of configuration.
• Cleaning up the database without securing the point of entry. Data cleansing without governance has a short shelf life: within a few months, duplicates come back. Address both fronts within the same program.
• Assuming the ERP already does MDM. The ERP validates format, not meaning or semantics. Quality rules, approval workflows and similarity checks need to run before the record is written.
• Leaving material descriptions in a free-text field. Free text generates ten versions of the same item. Use templates by category, with mandatory attributes and an abbreviation dictionary, aligned with the coded semantics requirements of ISO 8000-110 (2021).
• Ignoring the personal data inside the master record. Customer contacts and supplier partners are personal data subject to the data quality principle set out in Article 6, item V, of Law 13.709/2018. Define retention, updating and accountability.
• Buying a tool before defining the standard. Software accelerates rules that already exist. If the rule doesn't exist, the tool merely automates the inconsistency.

Metrics to track

• Duplicate rate by domain. Measure the percentage of records identified as duplicates against the total active base, using similarity comparison of name, CNPJ, description and attributes. Track it monthly by domain, not across the entire database.
• Completeness of mandatory attributes. Calculate the percentage of records with all mandatory fields filled in, by category. A material without a unit of measure, NCM code or manufacturer must count as incomplete.
• Average master data creation cycle time. Measure from request to creation in the ERP, separating queue time, validation time and approval time. This number shows whether governance is protecting or hindering the operation.
• Rejection rate at the point of entry. The percentage of requests blocked by automated rules, such as an irregular CNPJ or a detected duplicate. A high rejection rate at the start is a sign that the filter is working.
• Downstream incidents caused by master data. Count rejected invoices, blocked payments and emergency purchases whose root cause was master data. This is the indicator that translates quality into money.

Frequently asked questions

What does MDM mean in data management?

MDM stands for master data management. The DAMA-DMBOK (2017) defines the discipline as control over master data values and identifiers in order to maintain a consistent, accurate and shared version of the truth. In practice, it is the set of processes, roles, rules and technology that ensures customers, suppliers, materials and products exist only once and correctly across all systems.

What is the difference between master data and transactional data?

Master data describes entities the company reuses, such as customers, suppliers and materials. Transactional data records events, such as purchase orders and invoices. The DAMA-DMBOK (2017) makes this distinction explicitly. In SAP, according to SAP itself (2024), MARA holds general material data and LFA1 holds general supplier data, while the order and the invoice come later, inheriting whatever the master record defined.

Is MDM an ERP module?

No. The ERP executes transactions and stores the master record, but it does not guarantee that the record is correct. It validates format and mandatory fields; it does not detect that two records are the same thing written in different ways. MDM acts before that: request workflow, automated validation, duplicate checking by similarity and tiered approval, writing only the approved record into SAP, TOTVS or Oracle.

What are the most common master data domains?

Customer, supplier, material, product and service are the domains present in almost every mid-sized and large company. Industrial operations usually add asset, cost center and maintenance plan. Each domain has its own rules, approvers and life cycle: suppliers go through qualification and due diligence, materials require descriptive standardization, and customers require tax and registration status validation before sales release.

What is a golden record in MDM?

A golden record is the single, trustworthy record of an entity, assembled from the best information available across the various systems and adopted as the single source of truth. It consolidates one single supplier where there were three master records with the CNPJ entered in different ways. The golden record only holds up with defined survivorship rules, an accountable data steward and active integration with the source systems.

How does the tax reform affect customer and product master data?

Constitutional Amendment 132/2023 created IBS and CBS to replace ICMS, ISS, PIS and Cofins, and Complementary Law 214/2025 detailed the rules for registration, transaction classification and tax assessment. 2026 is the test year, with CBS at 0.9% and IBS at 0.1%. Without consistent NCM codes, customer records with the correct tax regime and organized CNPJs, the assessment of the new taxes is exposed to error and penalties.

How long does it take to implement MDM?

It depends on the number of domains, integrated systems and the current state of the database. Projects usually advance in waves: you start with one high-impact domain, such as supplier or material, with cleansing and governance running in parallel, and then expand. The frequent mistake is trying to tackle every domain at once. A scope bounded by wave delivers visible results early and sustains buy-in from the business areas.

Does AI solve the master data problem on its own?

No. AI accelerates specific tasks: detecting duplicates by similarity, standardizing material descriptions, suggesting attribute values and classifying items by category. But it learns from the existing database, and a bad database teaches the wrong lessons. A model trained on the same material registered five times fragments demand forecasting. Master data governance with defined rules and accountable owners is the precondition for AI to generate value.

How 4MDG helps with MDM (master data management)

We are MDM specialists and we work on both fronts the subject demands: MDM software with AI integrated into SAP, TOTVS and Oracle, and specialized services delivered by people who work with master data every day. The artificial intelligence embedded in our software detects duplicates by similarity, standardizes material descriptions and validates CNPJ and registration status before the record reaches the ERP. Our MDM Engineering method organizes domains, data owner and data steward roles, the record life cycle and the architecture, while MDM Academy trains the internal team to sustain the standard after go-live. From cleansing the current database to supplier qualification and customer onboarding, we take care of the point of entry and of the data inventory that already exists. Talk to 4MDG and bring the real problem in your operation to a conversation with people who do this every day.