Start with the question and lawful purpose
Define the decision the analysis should support and the minimum data needed. Confirm that each source is authorised for the intended use, that access terms permit processing and that personal information is handled appropriately.
Keep data-broker, public-record, customer-provided and internal sources distinguishable. Access to a dataset does not automatically grant permission to reuse or resell it.
Preserve provenance and confidence
Record source, retrieval date, permitted purpose, transformations and analyst actions. Distinguish verified facts, inferred signals and unconfirmed claims in the output.
Use corroboration for material decisions and provide a route to challenge or correct inaccurate information. Do not present an AI-generated summary as an authoritative source.
Put human review around consequential decisions
Use AI to organise information, identify patterns and draft analysis, but define human approval for decisions affecting eligibility, finance, employment, public services or other consequential outcomes.
Monitor data quality, bias, drift, access permissions, retention and auditability. Make it possible to explain which evidence informed a recommendation.
Trustworthy intelligence depends on authorised sources, traceable evidence, transparent uncertainty and accountable human review—not simply more data or more AI.