Start with a question the model should answer. “Which orders belong to CLI-001?” may need only two types and one relationship. Modeling the entire organization would make this first outcome harder to verify.
1. Choose concepts
Identify concepts with their own identity and lifecycle. Client and Commande are types. The amount is a property of Commande. Currency can remain a text code if you do not need a separate currency record.
Agree on definitions with users: does “active customer” mean an active contract or a recent order? Document the decision in the type or property description.
2. Define identifiers and properties
Choose stable business identifiers for each source. Display names cannot reliably match customers. Keep CRM and ERP references separate until their correspondence is established.
Record each property's type, unit, provenance and whether it is required. Before making a field required, check that historical records have values. See Naming conventions.
3. Describe relationships
Use a meaningful verb: Client places Commande. Choose direction and cardinality. Test a customer without orders and an order whose customer is not yet identified.
A source's client_id field alone does not establish a navigable relationship. Check the correspondence and the link actually available in the platform.
4. Test a small sample
Create or inspect one customer, two orders, a namesake and an unmatched order. In Instances, check values, origins and relationships. These cases expose ambiguity before importing larger volumes.
5. Evolve the model
List consumers: forms, sources, agents and APIs. Add an optional property, check persistence, then test an old and a new instance. Make it required only after planning how to handle historical records.
The outcome is a model where every type, property and relationship has an explained and verified purpose. See Evolve a used type and Model two sources.
