Semantic Interoperability

Definition

Semantic interoperability means that systems can exchange data and preserve the intended meaning of its concepts, measures and relationships. A shared model format helps exchange definitions, but teams must still agree on what a business term means in each context.

A semantic model describes business concepts, relationships and measures so data can be interpreted consistently. Interoperability has at least two checks: whether systems can exchange the model, and whether people agree on the meaning of what it describes. A shared file format addresses the first check but cannot settle a disputed business definition by itself.

In finance, small differences can matter. “Position” might be measured before or after settlement, at different times, or with different treatment of accrued interest. An AI workflow that combines the values needs these distinctions as explicit input; matching column names is insufficient.

What it covers

Structural exchange

Can tools read and transfer the same semantic model format?

Conceptual alignment

Do the participants mean the same thing by a term and calculation?

Provenance

Which definition, source and version produced a result?

Start with the decision the data supports. Identify the exact measures, sources, intended audience and impact of a wrong interpretation. Write down the population, units, time boundary, inclusion and exclusion rules for each term.

Separate the organization’s approved definition from each system’s local representation. A local field can map cleanly, map with a transformation, or remain incompatible. Preserve the source and version of every mapping so a later change can be reviewed.

In practice

Practical design patterns by Min.

• Choose a small set of high-impact terms and record their owners, definitions, calculation rules, time boundaries and exceptions.
• Map each source field to the shared concept, recording mismatches rather than silently forcing a match.
• Give AI workflows the approved definition and its version; check outputs against known examples before using them in consequential decisions.

Assign domain owners to review core terms. Publish a reference vocabulary incrementally and include concrete examples and counterexamples, not only short labels.

Use model-exchange formats such as Apache Ossie where they fit the toolchain, and compare relevant finance terms with FIBO. Neither replaces an explicit decision about local business rules.

Give AI applications the approved definitions at retrieval or query time. Validate generated queries and aggregated outputs against reviewed cases, and surface unresolved differences rather than guessing.

Research notes

Source-grounded details and editorial applications. Follow each link for the original document.

Working example

Why the controls matter

A risk assistant receives a “position” field from two portfolios. The first includes unsettled trades and the second uses a settled end-of-day snapshot. Before aggregation, the team records both definitions, chooses an appropriate normalization for the question, and labels any values that remain incomparable.

The assistant reports the definition and source versions used alongside the result. A reviewer compares a few known portfolios before the workflow is allowed to rely on the combined figure.

Common pitfalls

• Assuming a shared schema guarantees shared meaning.
• Treating a vendor model or industry ontology as an automatic substitute for local agreement.
• Letting an AI aggregate incompatible measures without showing the definition used.

A standardized exchange format can move an incorrect or ambiguous definition very efficiently. Validate meaning as well as syntax.

An ontology may define a general concept precisely while local accounting rules or regulatory purposes still require a narrower definition. Record that difference.

Do not present illustrative productivity gains or error reductions from a vendor essay as measured outcomes for this system.

Questions for operational review

Use these questions to decide what evidence your team needs. They are editorial prompts, not a compliance checklist.

• What does each source mean by the same term?
• Who approves a shared definition and its exceptions?
• Can a result be traced to its model and mapping versions?