Roles, not people – how to set up a permission model for product data in an orderly way

Permission models in PIM systems are often created in passing and grow uncontrolled after go-live. Why separating functional role and scope is the decisive step, and how to build a model in a structured way in seven steps.

Aaron Apelt PXM consultant
14 October 2026 · 14 min read

Summary

  • Permission models rarely fail because of technology, but because roles are built for people instead of tasks.
  • Separating functional role and scope prevents role explosion and keeps the model understandable even as the assortment grows.
  • A permission model is not finished at go-live. It needs a defined request process and a regular review of the rights granted.

In my last article, I looked at how a data model remains manageable after go-live. I deliberately left one question aside: who is actually allowed to change what in this data model and in the product data maintained within it?

The answer is provided by the permission model. However, hardly any part of a PIM project is treated as casually as this one.

Most projects follow the same pattern. During implementation, everyone involved works with extensive or even administrative rights, to prevent testing, data migration and acceptance from being blocked by missing permissions. This approach is essential for gaining speed and, in my view, also makes sense.

Shortly before go-live, roles are then created under time pressure – often guided by the question of which person needs to see which screens. Due to the time pressure and the fact that nobody has dealt with the topic in depth beforehand, the permission constructs that emerge here are often fairly simple and not well thought through.

After go-live, the first exceptions follow: a colleague needs additional access to another assortment, a service provider is to enter translations at short notice and an interface is given an administrator account for the sake of simplicity.

This careless handling of the topic results either in arbitrary assignments of roles to users or in a construct of dozens of roles whose differences and specific use cases nobody can keep track of any more.

Apart from being a security problem, this also has a direct impact on data quality. If it is not clearly defined who may maintain a piece of information, it is not clearly defined who is responsible for it either.

In this article, I would therefore like to show how to set up a permission model for product data in an orderly and structured way, and why separating two things that are almost always mixed up in practice is the decisive step.

The fundamentals are not new.

As with governance, it is worth looking beyond the PIM domain here too. The principles on which a good permission model is built have been established in information security for decades.

Principle of least privilege As early as 1975, Jerome Saltzer and Michael Schroeder formulated the principle of least privilege: every program and every user should operate only with the rights necessary for the respective task.1 Applied to product data, this means: whoever maintains marketing copy does not need write access to technical features. And whoever maintains product data does not need the rights to change the data model.

Role-based access control The NIST model for Role-Based Access Control (RBAC) describes that rights are not granted directly to people, but to roles. People receive rights exclusively through their assignment to these roles.2 The advantage is obvious: if a task changes, the role is adjusted – not every individual person.

Separation of duties Certain activities should not be in the same hands. A classic case is the separation of creation and approval. ISO/IEC 27001 also explicitly states that conflicting duties and areas of responsibility are to be segregated.3

Attribute-based access control In addition, NIST describes with the Attribute-Based Access Control (ABAC) approach that rights are decided based on properties, for example a user's department or a product's supplier.4 For product data, this idea is particularly relevant, as we will see shortly.

None of these principles is complicated in itself. But as always, the difficulty lies in transferring them to the structure of product data and reconciling them with real-world requirements and conditions.

The three dimensions of a permission in PIM

A permission in a product data system can be described in three dimensions.

DimensionQuestionTypical values
ObjectWhat may be worked on?products, attribute groups, classifications, data model, workflows
ScopeWhere does the permission apply?assortment area, language, channel, supplier, brand
ActionHow may it be handled?read, edit, create, approve, publish, delete

A complete permission answers all three questions. An example: a person may edit marketing copy for the "Garden" assortment in French, but not approve it.

Most PIM systems support these dimensions technically, but to varying depths. Some allow rights at the level of individual attributes, others only at the level of attribute groups. Some control the scope via categories, others via catalogues, product classes or dedicated user groups.

It is therefore worth checking very early in the project which of these dimensions the chosen system actually supports. A permission model that is conceptually sound but cannot be implemented technically inevitably leads to workarounds.

The real problem: role and scope get mixed up.

The most common problem when designing such permission models lies in mixing up role and scope. The "what" and "how" dimensions, by contrast, are cleanly translated into roles. The "where" dimension, however, is written into the role as well.

This results in roles such as:

  • Product management garden
  • Product management tools
  • Translation French garden
  • Translation French tools
  • Translation Italian garden

Each of these roles is understandable and unambiguous on its own. The problem, however, shows in the number of variants this creates.

Let us look at a deliberately simplified calculation. A company has 5 activities in product data maintenance, 12 assortment areas and maintains content in 6 languages. If the scope is written into the role, the worst case results in 5 × 12 × 6 = 360 roles. If activity and scope are separated, 5 roles plus 12 assortment areas and 6 languages as scopes are sufficient – that is, 23 building blocks that are combined with each other.

If a new assortment area is added, up to 30 new roles are created in the first case. In the second case, exactly one new scope is created.

In the literature, this phenomenon is known as role explosion.5 It is the main reason why permission models become incomprehensible over time.

Role and scope are separatedRole contains the scope
A role only describes the activity.A separate role is created for every combination of activity, assortment and language.
A new assortment area creates one new scope.A new assortment area creates many new roles.
Changes to an activity affect exactly one role.Changes to an activity have to be replicated across many roles.
Rights are requested via role and scope.Rights are granted as "the same rights as colleague X".
A role describes what someone may do. The scope describes where. Whoever mixes both in one role builds a model that grows with every new assortment and every new language.

Seven steps to a permission model

Below, I describe a process that can serve as a template for building a permission model. As with the governance model, the same applies here: this is a simplified approach that must be adapted to the circumstances of the respective company and the capabilities of the system in use.

Step 1: Capture activities, not people

The first mistake is often asking which person needs which rights. The result is a model that reflects the current organisation along with its coincidences.

The better starting question is therefore: Which activities exist in product data maintenance? The basis for this is the maintenance processes, which should be described in the project anyway. Typical activities are:

  • creating products and taking over master data
  • maintaining technical features
  • writing marketing copy
  • translating content
  • reviewing and approving product data
  • publishing products to channels
  • maintaining the data model (attributes, product classes, value lists)
  • managing users and rights

People do not appear in this step at all.

Step 2: Derive functional roles from the activities

In a second step, activities that are regularly performed together are combined into functional roles. The motto is: as few roles as possible, as many as necessary.

A sensible starting point for many companies are roles such as the following:

  • Product data maintenance: creates products and maintains technical features
  • Editorial: maintains marketing copy and sales-promoting content
  • Translation: maintains language-dependent content
  • Business approval: reviews and approves product data
  • Data steward: maintains the data model, but no product content
  • Administration: manages users, roles and system settings

Particular attention should be paid to the data steward role. It may change the data model, but may not maintain product content. This makes the separation between the structure of the data and the data itself visible in the permission model too. Changes to the model necessarily have to go through the responsible person in the Product Data Council.

Step 3: Define scopes separately

Only now does the "where" dimension come into play. For each role, it is defined by which criteria its scope can be restricted.

  • Product data maintenance, for example, is restricted by assortment area.
  • Translation is restricted by language.
  • Business approval is restricted by assortment area and, where applicable, by channel.
  • The data steward and administration usually have no restricted scope, but a very narrow area of responsibility instead.

This is where the value of the attribute-based approach becomes apparent. A scope does not necessarily have to be a category. It can also be controlled via an attribute on the product, such as the brand, the supplier or the responsible organisational unit. The prerequisite is that this attribute is reliably maintained, because whoever can change the attribute also changes the permissions.

The restrictions of the PIM systems must also be taken into account here, as many do not natively offer such granular control. It is nevertheless advisable to first think in the direction of the ideal solution and then adapt it to the restrictions. Decisions made because of limitations should be documented.

Step 4: Build the permission matrix

In the fourth step, roles, objects and actions are brought together in a matrix. It is the central document of the permission model. The following matrix is deliberately simplified and uses attribute groups as objects.

RoleMaster data from ERPTechnical featuresMarketing copyTranslationsApproval statusData model
Product data maintenancereadeditreadread–read
Editorialreadreadeditread–read
Translationreadreadreadedit––
Business approvalreadreadreadreadapproveread
Data stewardreadreadreadread–edit
ERP interfaceedit–––––

Two details of this matrix were chosen deliberately.

First, no human role can edit the master data from the ERP in the PIM. If the ERP is the leading system for these attributes, any manual change in the PIM would be overwritten with the next synchronisation. The permission thus reflects the system architecture.

Second, business approval has no write access to the content it approves. This means the four-eyes principle is not only agreed organisationally, but also enforced technically.

The matrix should not exist only as a concept. It should be stored under version control and serve as the reference for every later change.

Step 5: Define separation of duties

In the fifth step, it is defined which roles a person may not hold at the same time. For product data, three rules are typically relevant:

  1. Whoever edits content does not approve it themselves. This applies at least to the content a person has changed themselves.
  2. Whoever changes the data model does not maintain product content, to prevent attributes from being created "quickly" to solve a maintenance problem.
  3. Whoever grants rights does not grant them to themselves: administration needs a second person for changes to its own rights.

In smaller teams in particular, these rules cannot always be fully observed. In this case, it is better to document the exception consciously and compensate for it with a subsequent check than to silently ignore the rule.

Step 6: Handle special cases deliberately

Besides the regular roles, there are user groups that are neglected in many concepts.

Technical users for interfaces Each interface receives its own technical user. This user may only write those attributes for which the connected system is leading. A shared administrator account for all interfaces may be convenient, but in the event of an error it neither makes it traceable which system caused a change nor limits the potential damage.

External users Agencies, translation service providers and suppliers need access, but should only see their own section. For suppliers, it makes sense to control access via a supplier attribute on the product, instead of creating a separate role for each supplier. Access for agencies and service providers should always be granted for a limited time.

Emergency access Incident resolution and urgent corrections require access with far-reaching rights. It should be restricted to very few people, every use should be logged, and it should not be used for day-to-day work. In ISO/IEC 27001, such privileged access rights are treated as a topic of their own.3

Step 7: Test with scenarios

Before the model is used in production, it should be tested with concrete scenarios. Two kinds of tests are important here.

Positive tests check whether a role can fulfil its task, for example: "Can French translation translate a marketing text in the garden assortment?"

Negative tests check whether a role cannot do something it is not supposed to do: "Can French translation change the original German text?" "Can business approval edit a text and then approve it themselves?"

Negative tests are carried out far less frequently in projects. Yet they are the only way to determine whether the principle of least privilege has actually been implemented.

Time for testing should be explicitly planned. Such permission models are often very complex and require extensive business or domain knowledge. It is therefore difficult for a single person to design and set up everything without errors. With the help of AI, this review process can now be optimised.

After go-live: permissions need governance.

A permission model behaves similarly to a data model. With go-live, it begins to age: people change departments, new assortments are added and service providers are engaged and replaced again.

Without a defined process, the same gradual drift occurs that I described in the article on data model governance. Rights are often supplemented and added, but rarely revoked. ISO/IEC 27001 therefore prescribes a defined process for granting, changing, reviewing and removing access rights.3

Three building blocks have proven themselves in operation.

A defined request process Rights are not granted on request in passing, but via a request that contains the role, the scope and the justification. Approval is given by the person responsible for the respective scope on the business side and not by administration. Administration implements; it does not decide.

A regular review At fixed intervals, it is checked whether the rights granted are still needed. The 90-day rhythm from the governance model is well suited for external access and privileged rights. For all other rights, an annual review is usually sufficient, which can be combined with the strategic review of the data model.

Binding revocation If a person changes department or leaves the company, their rights are revoked. This sounds obvious, but in practice it is one of the most common gaps – especially when people change departments, where new rights are added but the old ones remain.

Conclusion

A permission model is not a technical configuration that should be done shortly before go-live. It is the translation of responsibility into the system. Whoever may edit a piece of information is also in fact responsible for it – regardless of what a role concept says on paper.

That is why the permission model belongs in the conception phase and not in go-live preparation. Its starting point is the activities in product data maintenance, not the people who perform them today or the organisational chart. The most important decision when building it is the separation of functional role and scope. It determines whether the model grows with every new assortment and every new language or remains understandable even years later.

However, the work is not finished at go-live. A permission model needs a defined framework similar to that of the data model itself: a request process, a regular review and binding revocation of rights.

A good permission model is barely noticeable in day-to-day work. A bad model, on the other hand, becomes apparent at the latest when someone asks who changed a certain piece of information – and nobody can answer that question.

  1. 1
    Jerome H. Saltzer, Michael D. Schroeder: "The Protection of Information in Computer Systems", Proceedings of the IEEE, Vol. 63, No. 9, 1975, pp. 1278–1308. ↩︎
  2. 2
    Ravi Sandhu, David F. Ferraiolo, D. Richard Kuhn: "The NIST Model for Role-Based Access Control: Towards a Unified Standard", 5th ACM Workshop on Role-Based Access Control, 2000. The model was adopted as the standard ANSI INCITS 359-2004 in 2004. ↩︎
  3. 3
    ISO/IEC 27001:2022, Annex A, in particular controls 5.3 (Segregation of duties), 5.15 (Access control), 5.18 (Access rights) and 8.2 (Privileged access rights). ↩︎
  4. 4
    Vincent C. Hu et al.: "Guide to Attribute Based Access Control (ABAC) Definition and Considerations", NIST Special Publication 800-162, 2014. ↩︎
  5. 5
    The term is used, among others, by D. Richard Kuhn, Edward J. Coyne, Timothy R. Weil: "Adding Attributes to Role-Based Access Control", IEEE Computer, Vol. 43, No. 6, 2010, as the motivation for combining RBAC and ABAC. ↩︎