Product documentation

Knowledge and answer accuracy

What is Business Context in DataFact?

Business Context is an explanatory umbrella term for the reviewed terminology, definitions, relationships, instructions, and examples that help an Agent interpret a question in the intended business setting. In the product, business entities, entity relationships, general knowledge, business knowledge, optimization examples, Dataset knowledge, and Analysis Model versions remain distinct objects with different behavior. They are not one merged backend object.

These forms of knowledge can give an Agent more relevant terminology and decision context. Analysis Model versions make some semantic guidance reusable and reviewable over time. Dataset knowledge can stay specific to one usable SQL boundary. An optimization example can demonstrate an intended pattern without becoming a test assertion or a guarantee that a later answer will match it.

Business Context can improve answer quality only when it is accurate, scoped, current, and available to the workflow. It does not replace source data. It also does not prove that the Agent retrieved the right record, generated the right query, or interpreted an answer correctly.

Knowledge quality matters more than volume. Incorrect, conflicting, stale, or overly broad entries can reduce answer quality. Similarity scores indicate retrieval relevance; they are not accuracy scores.

Why it matters

The same term can mean different things across finance, operations, and product teams. Explicit context gives a workflow a reviewed interpretation, while versioning gives teams a place to inspect changes over time.

How DataFact uses it

Eligible knowledge can be attached to a Dataset or organized through compatible Analysis Models, then made available to configured Agents according to product state and access.

Example

A definition of “active customer” can specify status, time window, and exclusions. A generated query still needs review to confirm that the selected fields and filters implement that definition.

Review practice

State verifiable business definitions and scope.
Distinguish source facts from instructions or examples.
Build and inspect versions before activating them.
Re-test representative questions after changes.
Validate consequential answers against authoritative sources.

Limitations and edge cases

Conflicting knowledge can lower answer quality rather than improve it.
An active model can be correct for one Dataset and incompatible with another.
Retrieval similarity is not an accuracy or confidence score.
Context cannot supply facts that are absent from the authoritative source.

Datasets and Analysis Models
AI answer verification
Evidence and citations
Evaluation lifecycle