A role model should describe real jobs in the product. Adding admin, manager, power user, and super user without clear differences creates ambiguity for customers and a large set of permission combinations for engineers.
Start with the smallest roles that support the pilot's actual responsibilities. Grant the actions each job needs and document which important actions remain restricted.
List actions before naming roles
Write the product operations: read a record, update it, export data, invite members, manage billing, delete resources, and transfer ownership. Include operational actions such as retrying jobs or inspecting private content if customers can perform them.
Then group actions into jobs. A reviewer might read and annotate, an operator might manage workflow state, and an owner might manage access. Resource relationships can still matter: an operator may edit assigned requests but not every request in the organisation.
Design a fictional request-management pilot
Suppose a service company needs staff to review enquiries and track follow-up. The pilot has two customer roles: operator and owner. Both can read and update service requests. Only the owner can invite staff, export all customers, or change account ownership.
A separate support role belongs to the product operator, not the customer organisation's ordinary role set. It has a scoped diagnostic workflow rather than unrestricted access to every customer record.
Test the two customer roles across read, update, export, invitation, and ownership actions. Add a user who belongs to a second organisation with a different role. A title in one organisation must not grant the same powers elsewhere.
Make exceptions visible rather than adding broad roles
If one staff member needs export access, determine whether that is a distinct recurring job or a narrow permission. A new broad administrator role may grant far more than the request requires. Conversely, dozens of individual exceptions can become impossible to explain.
Keep a permission table that customers and engineers can read. Use the same action vocabulary in acceptance cases and support tools. When a role changes, review the affected operations rather than assuming an interface label update changes server enforcement.
The tradeoff is granularity versus comprehensibility. Fine-grained permissions can express precise needs but increase configuration and test burden. A small pilot can begin with a simple model while documenting the boundary that would justify a new role.
For general principles, see OWASP's authorisation guidance. The StackFlight portfolio account supplies related SaaS context; this service-company model is proposed guidance.
Questions about MVP role design
Should every product have a viewer role?
Only if a real job requires read-only access. Roles should follow responsibilities, not a standard-looking dropdown.
Can owners bypass every product rule?
Define their powers explicitly. Ownership need not permit invalid state transitions, access to another tenant, or unsafe support actions.
How should a new role be approved?
Write its job, actions, excluded powers, and affected tests. Review that table in a SaaS permission discussion before implementing it.