Why Slow Data Is A Significant AI Risk In Life Sciences

Other sectors might tolerate slow data pipelines, but scientific research and patient care operate in real-time. In industries defined by urgency and precision, every delay is a potential catastrophe. However, most of the infrastructure supporting those operations was designed for after-the-fact analysis. Yet, even as organizations rely on systems never built for speed, many are waiting for the perfect solution to emerge. Meanwhile, the better option for most is real-time, event-driven orchestration.

Already, the industry has reached an inflection point, according to Saurav Ghosh, Vice President of Enterprise Digital Solutions for Data, Digital & AI at Genmab, an antibody science company. With over two decades of experience transforming life sciences with technology at companies like Infosys, Cognizant, IQVIA, and NNIT, Ghosh has a long history of driving growth and patient-centric outcomes. From his perspective, the most significant risk in life sciences today is inaction.

"Making precision decisions across different systems is difficult with static data sources. In biotech, data must be event-driven and point-in-time," Ghosh said. Because the current web of fragmented systems is obsolete, his philosophy is built on a simple principle: the timing of information is as critical as the information itself. To unlock AI's potential, leaders must create an architecture that acts on data instantly.

For Ghosh, the central challenge stems from what he called a "very messy architecture." Most biotech companies rely on a sprawling, fragile ecosystem of siloed SaaS systems, he explained.

• Universal donor: By creating a single, intelligent layer to interact with them, the Model-Controller-Prompter (MCP) route offers one viable path forward. But this power also comes with immense responsibility, Ghosh cautioned. "It comes with risk, because you are giving that layer elevated privilege to go into your systems, understand the data, and bring that context into AI to drive decisions."

According to Ghosh, the solution is a framework of strict guardrails, including prompt firewalls, immutable audit trails, and a human in the loop for oversight. Here, a phased approach to trust is a gradual journey, not a sudden leap.

• Assisted evolution: "I'm sure it will get to a point where we have a fully autonomous agent," Ghosh predicted. "But for now, I see a hybrid shadow mode with a human in the loop, trending towards a more fully autonomous state, almost like FSD in a Tesla."