Subbiah Sethuraman and Abhilash Asokan describe a pharmaceutical commercial-analytics pipeline in which individually plausible findings led to poor recommendations because handoffs lost the investigation's business context. A sales signal might require an insurance explanation rather than a change in representative activity.
Subbiah Sethuraman and Abhilash Asokan propose deterministic statistical detection before agent investigation and a single owner for reasoning and judgment. Delegated work can still gather evidence, but it should not fragment responsibility for the complete causal account.
Subbiah Sethuraman and Abhilash Asokan use a domain knowledge graph as a control plane for entities, metrics and candidate hypotheses. Relationships guide tests against underlying data rather than automatically becoming proven causes. The case study concerns commercial analytics, not clinical advice.
Watch on YouTube
