Subbiah Sethuraman is a partner at ZS who leads AI engineering for healthcare and life-sciences applications. His work encompasses pharmaceutical analytics, enterprise machine-learning infrastructure and agentic systems grounded in business context, with an emphasis on whether automated recommendations address their actual causes.
In 2020, he outlined an approach to measuring machine learning’s business impact: choose metrics around commercial objectives, balance accuracy against interpretability and operational complexity, and validate deployed models through controlled experimentation. By 2023, he was contributing to a life-sciences machine-learning infrastructure initiative spanning deployment, orchestration, monitoring, data drift and governance.
Working with fellow ZS engineer Abhilash Asokan, Sethuraman subsequently confronted a failure common to pharmaceutical analytics: separate agents could correctly identify falling prescriptions and reduced insurance coverage, yet recommend deploying additional sales representatives—an action that did nothing to improve affordability. Their revised architecture rests on three principles:
- Deterministic signal detection: Conventional statistical methods identify, filter and prioritize prescription anomalies before an agent begins investigating.
- End-to-end ownership of reasoning: One agent retains responsibility for causal analysis and recommendations; specialized sub-agents can gather evidence without inheriting final judgment.
- Knowledge graphs as a control plane: Relationships among brands, geographic markets, insurers and performance indicators constrain which hypotheses the system investigates and connect potential explanations to underlying data.
For Sethuraman, effective enterprise AI depends on keeping statistical evidence, domain knowledge and recommended action coherent throughout the same decision process.
