Must all source databases be redesigned?

Not necessarily. Mappings can connect existing structures to the shared representation.

Does conformance prove data quality?

It proves only the checks actually performed; semantic accuracy and factual correctness require additional validation.

Is Common Data Model always a Microsoft name?

Capitalized usage often refers to Microsoft's specific model, while the lower-case phrase can describe the general concept.

AI Glossary / Data engineering and architecture / Intermediate

Common data model

Definition

A common data model is an agreed way to describe shared entities, attributes, relationships and meanings across systems. It helps different applications exchange and interpret information consistently. The phrase can be generic; Microsoft Common Data Model is a specific published collection of extensible schemas and metadata for common business concepts.

What it means

Two systems may both store customers but disagree about identifiers, required fields or what counts as a customer. A shared model provides a target vocabulary and structure for mapping those differences. It does not require every source system to use identical physical tables. Microsoft Common Data Model supplies defined entities and metadata within its ecosystem, while other organizations or sectors can define their own common models. Name the specific model and version when interoperability depends on it.

How it works

Model designers define entities, field types, relationships and semantic descriptions. Source data are mapped to those definitions, with transformations for units, formats and identifiers where needed. Validation checks whether records conform to the agreed structure and constraints. Governance manages extensions and versions so one team's customization does not silently change another team's interpretation. A model describes how data should be understood; an integration pipeline or application must still implement the mapping and maintain the actual records.

Three practical examples

Sales and support applications map their account records to a shared customer representation for reporting.
An organization standardizes product identifiers, units and category meanings before combining inventory data from several systems.
An AI team prepares training features from multiple sources using documented entity definitions, making it easier to identify when similarly named fields actually mean different things.

Think of it this way

Think of agreeing on a common set of labels and measurement units before several workshops exchange parts. Matching labels can make assembly easier, but only if the parts really follow the agreed definitions. The analogy is limited because data models also describe relationships, optionality and semantics; they cannot verify the truth of every incoming record merely by checking its shape.

Worked example

Suppose system A stores an order amount as 12,500 cents and system B stores 125.00 in major currency units. A common model defines Amount as a decimal number of major units plus a separate Currency field. Mapping A requires division by 100 for the specified two-decimal currency, while B may map directly. Omitting Currency would still leave ambiguity between different monetary units. The example is a representation exercise, not a universal currency conversion rule: currencies and source conventions vary, and exchange-rate conversion is a separate transformation requiring its own date and policy.

Illustrative case study

Fictional case: Rowan Products adopts a shared product model. One source uses “active” to mean available for sale, while another uses it to mean the record has not been archived. The team creates distinct fields and documents their meanings instead of mapping both to one flag. It adds representative examples and versioned validation checks. This fictional case illustrates semantic modeling rather than a claim that adopting a schema automatically cleans all product data.

Common misconceptions and mistakes

Do not confuse a common data model with a central database, a data lake or a synchronization service. A model can be shared even when data stay in separate systems. Matching column names does not establish matching meaning. Overly broad models can become difficult to govern, while uncontrolled extensions can undermine interoperability. Also, the Microsoft Common Data Model is not automatically the same thing as Microsoft Dataverse or every model used by a Microsoft product; refer to the relevant documentation and service boundaries.

Business use

Businesses can use a shared model to reduce repeated mapping work and improve consistency across analytics and applications. Begin with the concepts needed for a defined use case rather than attempting to standardize every field at once. Assign owners to definitions and changes. For AI, record the model version used to construct a dataset and retain source lineage, since structural consistency does not remove bias, missing values or incorrect labels.

Check your understanding

Two systems map a field named Status into one shared field, but one means payment status and the other means shipping status. Is the mapping sound because the names match?

Answer: No. The meanings differ and should be represented separately or transformed according to an explicit business definition.

Related terms

Schema
Interoperability
Data governance
canonical data model
entity
semantic metadata
data mapping

Learn next

Select one shared entity and compare its fields in two source systems. Write definitions, units, allowed values and ownership for each candidate shared field. Test the mapping with edge cases and document any information that cannot be represented without an extension.

Sources

Primary product references: Microsoft Common Data Model documentation, https://learn.microsoft.com/en-us/common-data-model/ ; published schemas and description, https://github.com/microsoft/CDM . Official documentation and repository search excerpts consulted; full model-version and editorial review remain pending. Generic common-model principles are distinguished from the named Microsoft implementation.

Previous: Common cause Next: Common ground