Evidence you can audit: citations in AI-assisted diagnostics

AI can read faster than any team of analysts. It can also state things with confidence that are not supported by the material it read. For a consultancy, that combination is a risk: a recommendation is only as good as the evidence behind it, and a board needs to be able to check that evidence for itself.

Citations as a design principle

UniversAIDS was built so that no finding exists without its sources. When it identifies a bottleneck, it records where it saw it: the workbook and row, the page and section of a procedure, the interview and session. When findings are linked into a graph, the links carry their evidence too.

• Spreadsheet findings cite the file, tab and row
• Document findings cite the file, page and section
• Interview findings cite the session and speaker role
• System findings cite the export and time window

If a finding cannot show where it came from, it does not go in the report.

Why it matters to leadership

Citations change the conversation in the boardroom. Instead of debating whether a consultant's conclusion is right, leadership can open the source and decide for themselves. Disagreements become specific and quick to resolve. Recommendations that survive that scrutiny are far easier to fund.

Why it matters to us

Citations also discipline our own work. A senior consultant validating a candidate root cause starts from its sources, not its summary. Weak findings are visible as weak because their evidence is thin. That is how we keep the AI in its proper role, proposing and correlating, while people decide.

Handling evidence responsibly

Raw evidence often contains sensitive operational and personal data. In UniversAIDS it sits in a partition for that engagement, processing runs only on a consultant's instruction, and only a partner can close an engagement and purge its raw evidence. Access follows the consultant's role throughout.