Glossary Terms

A practical rule or shortcut used to guide a decision when an exact solution is difficult. Heuristics can reduce effort, but their assumptions and failure cases should be made explicit.

A configuration choice that governs a model or its training, rather than being directly learned from the training data. Evaluate choices using suitable validation and record them for reproducibility.

Performing an activity within the organisation rather than relying on an external provider. Assess the required capability, capacity and controls before assuming internal delivery will improve cost or quality.

A system that observes information and takes actions towards an objective. In business use, define its permissions, action boundaries and escalation requirements, with a named person accountable for its operation.

An assessment of an organisation’s resources, capabilities and current ways of working. It helps determine readiness for change and where strengths or constraints affect a proposed AI use case.

The process of examining and representing how a person moves through a service or task. Include actual user evidence, exceptions and handoffs so the map reflects operational reality.

A clustering method that assigns observations to a chosen number of groups based on similarity to cluster centres. Results depend on feature scaling, the chosen cluster count and the structure of the data.

A managed collection of information used to answer questions or support work. For AI-assisted retrieval, keep sources current, define access permissions and provide a way to check the underlying evidence.

A structured representation of entities and the relationships between them. It can support information retrieval and reasoning, provided relationships are sufficiently accurate, maintained and relevant to the business task.

A measure used to track progress against an important objective. Select indicators that reflect the intended business outcome, define how they are calculated and avoid confusing activity with improvement.