Why We Killed Our Multi-Agent Pipeline — Subbiah Sethuraman and Abhilash Asokan, ZS Associates

Splitting analysis among agents can produce a recommendation that loses the reason for the problem. In this pharmaceutical analytics case study, the speakers illustrate a pipeline that identifies insurance affordability as the cause of declining prescriptions, then recommends more sales representatives.

Their redesign gives one agent responsibility for the final judgment while retaining focused subagents. Statistical code detects changes before an investigation begins. An expert-built graph maps possible relationships between business measures, guiding which explanations the agent checks against actual data. The graph supplies hypotheses, and one agent keeps the evidence connected to the recommendation.

“it has actually derived the right fact, but then there is no single agent which is owning which understands the end-to-end picture basically”

Subbiah Sethuraman

“So, the agent's job is to investigate, not to identify.”

Abhilash Asokan

“So, the knowledge graph is not just something the agent looks up for data. It is a control plane for the agent.”

Abhilash Asokan

“it's able to produce something an analyst was able to produce maybe in three or four weeks in like maybe 20 30 minutes.”

Abhilash Asokan

“any complex workflows will have deterministic parts and agentic parts. Don't let agents actually run the deterministic part, right?”

Subbiah Sethuraman