IIIOrganization & Project ManagementGuide
Maintaining Master Data: Customers, Items, Prices in One Place
The same customer with three addresses, the same item with two prices: unclean master data cause errors in every workflow that uses them. A short charter and clear responsibilities keep them clean.
Key takeaways
- Master data are the basic data that many workflows share: customers, suppliers, items, and prices. Errors in them repeat in every quote and every invoice.
- According to Wang and Strong (1996), data quality is measured by whether the data are usable for the user, not only by whether they are correct.
- In 20 case studies analyzed, good results depended mainly on people's behavior and clear roles, less on technology.
- The master data charter defines for each data type a leading system, an owner, few mandatory fields, a change procedure, and a review rhythm.
Why unclean master data cause errors in every workflow
A customer has moved. The accounting program has the new address, the CRM the old one, and the field sales spreadsheet a third. The invoice goes to the wrong address, so does the reminder, and in the end the customer calls, annoyed. The same happens with prices: the price list was raised in January, the quote template was not.
Such errors seem small, but they are expensive because they repeat. A wrong entry in a quote affects one quote. A wrong master record affects every quote, every delivery, and every invoice that uses it, until someone notices. With e-invoicing, which passes data on without human checks, this becomes even more visible.
This article answers one question: how does a small company ensure that customers, items, and prices are maintained correctly in one place and that all workflows use the same data? To answer, we first clarify the terms, then present a thinking model of data quality, examine findings from case studies, weigh central against distributed maintenance, and finally describe a charter for master data.
An error in a quote affects one customer. An error in a master record affects every transaction that follows.
What are master data, and what is a leading system?
Master data are basic data that change rarely and are shared by many workflows: customers with address and contacts, suppliers, items and services, prices and terms. Transaction data, by contrast, are created anew with each process, such as a quote, an order, or an invoice. Transaction data draw on master data. That is why errors in master data act like a copying error that reappears in every new transaction.
The leading system is the one place where a data type is maintained and which prevails in case of doubt. All other programs look it up there or receive the data from there. This rule corresponds to the principle 'one piece of information, one place' from the article on how to reduce tool sprawl.
An old saying goes: a man with one watch knows what time it is; a man with two is never sure. Master data behave the same way. Two customer lists are not twice as good as one, but worse than one, because nobody knows which one applies.
The thinking model: data quality from the user's perspective
In 1996, the researchers Richard Wang and Diane Strong examined what data quality means to the people who work with data. Their starting point was the observation that improvement efforts in companies usually focus narrowly on accuracy. Using a two-stage survey and a sorting study, they developed a framework with four categories. Their core finding: data consumers have a much broader understanding of quality than the professionals who run the systems (Wang and Strong 1996).
| Category | Meaning | Example for master data |
|---|---|---|
| Intrinsic quality | The data are correct and credible in their own right. | The address is right, the price is the valid one. |
| Contextual quality | The data fit the task: complete, current, relevant. | The invoicing email is entered for e-invoicing. |
| Representation | The data are understandable and presented consistently. | Item names follow the same spelling. |
| Accessibility | The data are reachable for authorized users and protected. | Field sales sees the current price on the phone. |
The model explains why master data are often considered poor even though each individual entry is correct. A correct address without an invoicing contact is useless for accounting. A correct price that field sales cannot access gets replaced by a private list, and there are two watches again. For sales data in the CRM, the article on CRM data quality names four properties: complete, current, unique, and honest. They fit into this framework. For master data, representation and accessibility are added, because many different workflows read the same data.
What case studies show about data maintenance
Walsh, McAvoy, and Sammon (2025) analyzed 20 published case studies of programs for steering data in companies, known as data governance, meaning the rules for who may maintain and use which data and how. They condensed the experiences into 17 insights, ordered by preparation, launch, and sustaining. The clearest result: success was determined mainly by people and their handling of data, less by theory or technology. The models from research hardly played a role for practitioners.
Several insights transfer directly to small companies. Successful programs defined clear roles and responsibilities from the start and wrote them down. They treated data maintenance not as a task for IT alone, but as a joint task of the areas that create and use the data. And they trained users in their duties, because even a good system delivers bad data if people do not know how to enter them. The authors also cite an estimate according to which a lack of trust in data can divert up to half of knowledge workers' time into searching for reliable data. It is a rough estimate, not a measurement, but it shows the magnitude of the problem.
Haug, Staskiewicz, and Hvam (2023) followed a large international company over an extended period as it introduced central master data management. They show that neither central nor distributed maintenance solves all problems. Each has its own advantages and drawbacks, and the authors derive four distinct strategies from this. The study makes clear that organizing maintenance is a deliberate decision.
Central or distributed maintenance? Two perspectives
The first perspective says: one person maintains master data. Only then do spellings stay uniform, duplicates rare, and prices consistent. Anyone who needs a new customer requests it, and the responsible person creates it.
The second perspective says: data are best maintained by whoever learns about them first. The salesperson knows the new address, the buyer the new supplier price. Every detour via a central office delays the change, and shadow lists emerge in the meantime.
Central
- Uniform spelling and few duplicates
- Clear responsibility
- Bottleneck at the responsible person
- Changes arrive late
Distributed
- Changes where they become known
- Fast and without detours
- Inconsistent entries and duplicates
- Nobody feels responsible for the whole
The findings of Haug and colleagues suggest that neither form works alone. For small companies I consider a hybrid right: responsibility for each data type lies with exactly one person, while entry happens where the information arises, according to fixed rules. The responsible person reviews and cleans up regularly but does not type everything personally. This matches the principle of the RACI matrix: many may perform, but only one person is accountable.
The master data charter: five rules on one page
The master data charter is the tool of this article. It records five rules for each data type. Like the filing charter from the article on digital filing, it fits on one page and is kept where everyone can find it.
| Data type | Leading system | Owner | Mandatory fields | Review |
|---|---|---|---|---|
| Customers | Order software | Head of inside sales | Company, billing address, invoicing email, contact | monthly sample, duplicates yearly |
| Suppliers | Accounting | Purchasing | Company, address, bank details, payment terms | callback for every change of bank details |
| Items and services | Order software | Product owner | Number, name, unit, tax rate | quarterly |
| Prices and terms | Order software | Owner | List price, valid from, discount limit | at every price change, fully once a year |
Worked example
Three points deserve special attention. First: few mandatory fields. Every additional mandatory field increases the likelihood that someone fills it with a placeholder. Second: a change procedure for sensitive data. A supplier's changed bank details are accepted only after a callback to the known phone number, because forged change notices are a common fraud scheme. Third: archive instead of delete. A customer who has not ordered for years is marked inactive so that old invoices remain traceable.
How to clean up an overgrown data set in four weeks
Week 1: define the leading system
Decide for each data type which system applies. All other lists are no longer maintained from now on, only read.
Week 1: draw a sample
Check 50 customer records and 50 items against the four categories. Count which errors occur most often.
Week 2: merge duplicates
Sort by name, postal code, and phone number. Merge duplicate entries and archive the surplus ones.
Week 3: close gaps
Fill in the mandatory fields for active customers and items. Inactive ones are marked, not updated.
Week 4: introduce the charter
Present the charter to the team, briefly train who enters what, and put the first monthly sample in the calendar.
Worked example with freely chosen numbers. A record can have several errors. The most frequent errors concern usability for the task, not correctness.
The example shows a typical pattern that the Wang and Strong model would lead you to expect: most errors are not wrong entries but missing or outdated ones. The missing invoicing email becomes important when e-invoicing becomes mandatory for sending as well from 2027 or 2028, as the article on the e-invoicing mandate describes. Clean master data are also the first layer of any digitalization, as the article on digitalization in the right order shows.
Conclusion: how do master data stay correct in one place?
We set out to ask a small company maintains customers, items, and prices correctly in one place. The answer has three parts. First, there is exactly one leading system for each data type, and all other lists are abandoned. Second, according to Wang and Strong, quality is measured by usability for users, meaning completeness, consistent representation, and access as well, not just correctness. Third, according to the case studies analyzed, people and clear roles matter most. For small companies this means: responsibility with one person per data type, entry where the information arises, and a short review every month. Master data do not become good through a one-time cleanup, but through a rule that applies after the cleanup.
The next step
- Define one leading system each for customers, suppliers, items, and prices.
- Name a responsible person for each data type.
- Limit mandatory fields to what quotes, invoices, and deliveries really need.
- Draw a sample of 50 customer records and count the errors.
- Set a change procedure with callback for bank details.
The sample takes barely an hour and almost always shows where most errors in quotes and invoices originate.
Frequently asked questions
What are master data?
Master data are basic data that change rarely and are shared by many workflows, such as customers, suppliers, items, and prices. Transaction data like quotes or invoices draw on them. Errors in master data therefore repeat in every transaction.
How do you maintain master data properly?
With one leading system per data type, a responsible person, few mandatory fields, a fixed change procedure, and a regular sample check. Data are entered where the information arises, with one person accountable.
What is a leading system?
The leading system is the one place where a data type is maintained and which prevails in case of doubt. Other programs look it up there or receive the data from there. Two equal lists almost always lead to contradictions.
How do you recognize good data quality?
According to Wang and Strong (1996), by the data being usable for users: correct, complete, and current, consistently presented, and accessible to authorized users. Correctness alone is not enough.
How often should master data be reviewed?
A monthly sample of 20 to 50 records is usually enough in small companies. A search for duplicates is worthwhile once a year. Prices are checked at every change and fully once a year.
Sources
- Wang, R. Y.; Strong, D. M. (1996): Beyond Accuracy: What Data Quality Means to Data Consumers. Journal of Management Information Systems, 12(4), 5-33.
- Walsh, M. J.; McAvoy, J.; Sammon, D. (2025): The Data Governance Journey in Practice: Insights from Case Study Research. Information Systems Management, 42(3), 471-488.
- Haug, A.; Staskiewicz, A. M.; Hvam, L. (2023): Strategies for Master Data Management: A Case Study of an International Hearing Healthcare Company. Information Systems Frontiers, 25(5), 1903-1923.
