Data cleansing vs data scrubbing: what is in scope?
Data cleansing, cleaning and scrubbing are near synonyms; the scope behind them is not. A deliverables table to check what a cleansing project includes.
Data cleansing, data cleaning and data scrubbing are close to synonyms in everyday use. What changes from one proposal to the next is the scope behind the name: which master data domains are covered, which rules apply, what gets delivered and how you prove it was delivered. Before comparing price or timeline, turn the label into a list of deliverables you can check.
A familiar scene in any master data team: two proposals arrive in the same week. One is called a "customer data cleanup", the other a "master data cleansing project". Different prices, different timelines, and someone in leadership asks why the company should pay more for the same service.
It may not be the same service. One may cover only customers and check tax IDs. The other may include materials, description standards and a reload into the ERP. The name will not tell you. The list of deliverables will.
What does each word actually mean?
This article comes from Brazil, where the market uses three Portuguese words for this work: "higienização" (roughly, hygienization), "limpeza" (cleaning) and "saneamento" (sanitation or remediation). In English, the closest equivalents are data cleansing, data cleaning and data scrubbing. The Brazilian words carry slightly different shades of meaning, and the Michaelis dictionary, a Brazilian Portuguese dictionary, shows them clearly. Translated from its entries:
- To sanitize or remediate: among other senses, "to repair or reduce deviations or irregularities" and "to put an end to failures, irregularities or excesses". In the legal sense, "to rid a proceeding of nullities, flaws or defects".
- To hygienize: "to make hygienic or healthy" and "to use cleaning practices with the aim of preventing or avoiding the spread of disease".
- To clean: in the computing sense, "to remove unnecessary or outdated data from a file or disk".
That is the dictionary. Here is our reading: cleaning leans toward removing; hygienizing leans toward keeping the data healthy and stopping problems from spreading; sanitizing leans toward fixing irregularities. In the master data market, all three have become near synonyms.
That is also what the 4MDG master data cleansing page says: master data cleansing is also called data cleansing, data cleaning or data scrubbing. When people want to draw a distinction, scrubbing usually means correcting and standardizing, while cleansing covers the whole process. Data quality is the measured result; cleansing is the project.
Why does the same name hide different scopes?
Because each department, each vendor and each internal project uses the word for the part of the problem it needs to solve at that moment. Three public examples show how wide the range is.
In Brazil, CERC published a 2023 notice on payment arrangements whose title translates roughly as "Hygienization of the database". There, hygienizing meant leaving records inactive for more than 90 days out of a data delivery flow, with those records available only on request to support. No correction, no standardization: hygienization as taking inactive records out of a flow.
In a 2006 paper from the U.S. Census Bureau, researcher William E. Winkler describes record linkage, which combines information from different files to recognize the same entity, and notes that it is also referred to as data cleaning. Even in technical literature, "cleaning" can mean, first and foremost, finding records that are the same thing.
At the other end, the UK Government Data Quality Framework states that "Data quality is more than just data cleaning." It defines quality as fitness for purpose and recommends treating issues at source, with continuous monitoring.
Three legitimate uses, three scopes. None of them is wrong. The risk is signing for one while believing you bought another.
Read also: the difference between data cleansing and data governance.
What may (or may not) be in scope?
Use the table below to review a vendor proposal or to plan an internal project. In each row, the question closes the scope, the third column says how to check the delivery and the last column shows the sign that the item was left out.
| Deliverable | Question that closes the scope | How to check the delivery | Sign it was left out |
|---|---|---|---|
| Assessment with a measured baseline | What is the state of the data before work starts, measured by written rules? | Report with the rules applied and the result per rule and per domain | The proposal promises to "improve quality" without saying where it starts |
| Format standardization | Which fields get a standard format (tax ID, postal code, phone, unit of measure)? | List of fields with the format rule and a before and after sample | Only "address correction" is mentioned, with no list of fields |
| Description standards for materials (PDM) or services (PDS) | Will descriptions follow a standard per class, with short and long descriptions generated by rule? | Documented standards per class and descriptions generated from them | Descriptions "reviewed" by hand, with no written standard |
| Classification and review of NCM, NBS, CEST and UNSPSC | Which of these codes will be filled in or reviewed, for which items, and who decides the final code? | Field filled in with the evidence used and the recorded approval of the company's tax team | No one from tax appears in the approval flow |
| Deduplication | Which rule says two records are the same, and which one survives? | Written matching and survivorship rules, plus a list of groups with the chosen golden record | "Duplicate removal" with no rule for which record stays |
| Validation against official sources | Which official sources will be checked (Brazilian company tax ID, registration status, state tax registration)? | Source, check date and result stored per record | "Validation" with no date and no source |
| Enrichment | Which empty fields will be filled in, and from which source? | Field by field, the origin of each new value | Enrichment mentioned with no list of fields or sources |
| Handling inactive records | Will unused records be blocked, deactivated or deleted? Who approves? | List of affected records, action taken and history preserved | The word "cleanup" appears without saying whether anything is deleted |
| Domains included | Which master data is in: materials, services, products, customers, suppliers? Individuals and companies? | Domains listed in the scope, with the slice of each one | The proposal only says "the database" |
| Reload into the ERP | How does the data go back into the ERP, and who tests before and after the load? | Load plan, test in a separate environment and sign off by the people who use the data | The final deliverable is a spreadsheet |
| Log of every change | Will each change record the old value, new value, source, approver and date? | History you can look up per record | No one can tell why a field changed |
| Rules at the point of entry | What stops the data from getting dirty again after the project? | Validation rules and an approval workflow active for new records | The project ends at the load, with nothing for future records |
| Measurement afterwards | Will the data be measured again with the same rules used in the baseline? | Before and after report using the same yardstick | Results are shown with a different metric from the baseline |
A note on the codes: NCM, NBS and CEST are Brazilian codes. NCM is the Mercosur goods classification used in Brazil, NBS is the Brazilian services nomenclature and CEST identifies goods under the tax substitution regime. UNSPSC is an international classification of products and services. In this table they are master data attributes to classify or review. The decision on any code belongs to the company's tax team, and this article does not cover tax rules.
The description standards row is often the one that most separates one scope from another for materials and services. If it is in scope, it is worth understanding how a material description standard (PDM) or a service description standard (PDS) works before you approve.
For field by field detail on materials, the material master data cleansing checklist complements this table. For the baseline and the measurement afterwards, see how to measure master data quality.
What about the other names: deduplication, enrichment, data scrubbing?
These names usually point to one row, or a group of rows, in the table, not to the whole table.
- Deduplication: merging repeated records into a golden record, using a similarity score and a survivorship rule. For materials, this is where the problem shows up most; the warning signs are in how to identify duplicate materials.
- Enrichment: filling in empty fields from reliable sources.
- Bulk tax ID validation or registration updates: common labels when the work is about customers and suppliers, focused on checking data against the Brazilian Federal Revenue Service, state tax registration, address and postal code.
- Data scrubbing: usually correction and standardization. Data cleansing usually means the full process. As the Census Bureau example shows, the name alone does not settle the scope in English either.
How does the table work in a real decision?
Hypothetical example. A manufacturer receives two proposals. The first, called a "hygienization" of master data, checks tax IDs and registration status for customers and standardizes addresses and postal codes. The second, called "data cleansing", covers materials with PDM description standards, deduplication with a survivorship rule and a reload into the ERP, but does not touch customers.
The team lays both proposals side by side in the table and marks each row as in, out or not stated. Three things come up. The problem that triggered the project, repeated purchases of items that already exist under another name, would only be addressed by the second. Supplier validation, which is blocking payments, is in neither. And neither includes rules at the point of entry or a measurement afterwards with the same yardstick.
The final scope takes parts of each proposal and adds what was missing. The project name ends up being the least important detail.
Where to start
1. Write down the problem before the name
What triggered the project? Wrong purchases, rejected invoices, suppliers blocked from payment, reports that do not reconcile. Note which domain the problem shows up in.
2. Pick a slice and measure it
Start where the problem weighs most. ABC prioritization helps decide where cleansing should start. Measure the baseline with written rules.
3. Go through the table row by row
For each deliverable, mark in, out or next phase. Anything left out should be written down as out, not forgotten.
4. Ask for the scope in writing
Whether the work goes to a vendor or to your own team, ask for the answer in the same rows. Compare proposals by the table, not by the title.
5. Agree on acceptance
Decide who tests the ERP load, how each change will be logged and when the data will be measured again. For materials, the material master data cleansing guide walks through the work item by item; for the full service overview, see the data cleansing page.
Frequently asked questions
Are data cleansing and data scrubbing the same thing?
In everyday use, they are close to synonyms. When people draw a line, scrubbing means correcting and standardizing, and cleansing covers the full process. Since each proposal puts different work behind the name, the list of deliverables is what defines the job.
Does a data cleanup delete records?
It depends on the agreed scope. In many cases, unused records are blocked or deactivated, keeping their history, instead of being deleted. If a proposal uses the word "cleanup", ask in writing what happens to inactive records, who approves and where the history is kept.
Does a customer data cleanup work for materials?
Usually not. Customers and suppliers need validation against official sources, such as tax ID, registration status and state tax registration in Brazil. Materials need description standards, units of measure, classification and deduplication. Check in the table which domains are included.
What drives timeline and cost?
The domains included, the size of the slice, how many rows of the table are in scope, the state of the data at baseline and who approves changes on the company side. Two proposals with the same name and different scopes cannot be compared on price alone.
How do I ask for the scope in writing?
Send the table in this article and ask for each row to be marked as included, excluded or optional, together with how the delivery will be checked. The same applies to an internal project: the team answers in the same rows before work starts.
How 4MDG can help
4MDG provides master data cleansing for materials, services, customers and suppliers: PDM and PDS description standards, NCM and UNSPSC classification, deduplication down to a golden record, validation against public sources, native SAP integration and rules at the point of entry so the data does not get dirty again. For materials, the starting point is material master data cleansing. If you want to see your own scope in the table with real data, request an assessment on a slice of your data, with nothing to install, or reach us through the contact page.
Informational content. Definitions come from a Brazilian dictionary and term usage reflects the market; this article does not replace a contract, a formal proposal or tax advice on NCM, NBS or CEST codes, which apply in Brazil.
Sources
- Michaelis, Dicionário Brasileiro da Língua Portuguesa (Editora Melhoramentos), entry "sanear" (in Portuguese): https://michaelis.uol.com.br/moderno-portugues/busca/portugues-brasileiro/sanear/. Accessed on November 12, 2026.
- Michaelis, Dicionário Brasileiro da Língua Portuguesa (Editora Melhoramentos), entry "higienizar" (in Portuguese): https://michaelis.uol.com.br/moderno-portugues/busca/portugues-brasileiro/higienizar/. Accessed on November 12, 2026.
- Michaelis, Dicionário Brasileiro da Língua Portuguesa (Editora Melhoramentos), entry "limpar" (in Portuguese): https://michaelis.uol.com.br/moderno-portugues/busca/portugues-brasileiro/limpar/. Accessed on November 12, 2026.
- CERC, "Arranjos de Pagamento: Higienização na base de dados" (in Portuguese, 2023): https://cerc.com/wp-content/uploads/2023/08/Arranjos-de-Pagamento_-Higienizacao-na-base-de-dados.pdf. Accessed on November 12, 2026.
- UK Government, "The Government Data Quality Framework": https://www.gov.uk/government/publications/the-government-data-quality-framework/the-government-data-quality-framework. Accessed on November 12, 2026.
- U.S. Census Bureau, William E. Winkler, "Overview of Record Linkage and Current Research Directions" (2006): https://www.census.gov/library/working-papers/2006/adrm/rrs2006-02.html. Accessed on November 12, 2026.
- 4MDG, "Master data cleansing": https://4mdg.com.br/en/data-cleansing. Accessed on November 12, 2026.