Northwestern Medicine’s Sama Says AI Optimization Requires a New Data Foundation

Why It Matters

Without a robust data foundation, AI projects risk failure and inflate costs, threatening the promised clinical and financial gains for health systems.

Key Takeaways

• Health systems spend 80% time on AI, underinvest in data
• Legacy data silos hinder AI deployment across hospitals
• LLMs enable AI on messy, incomplete clinical data
• Build custom tools for unique workflow gaps; buy generic solutions

Pulse Analysis

Healthcare executives are caught in an AI frenzy, yet surveys show that less than half of health systems have modern data platforms capable of supporting real‑time analytics. This mismatch creates a hidden cost: fragmented EHR extracts, legacy interfaces, and siloed clouds generate technical debt that slows innovation. By prioritizing enterprise‑wide data lakes, FHIR‑based APIs, and event‑driven pipelines, organizations can lay the groundwork for scalable AI, reducing integration overhead and improving data quality at the source.

Large language models are reshaping how clinicians interact with imperfect data. Instead of waiting for exhaustive data curation, LLMs can parse unstructured notes, imaging reports, and streaming sensor feeds, delivering actionable insights while embedded safeguards ensure clinical reliability. Coupled with standardized HL7 and FHIR exchanges, these models accelerate use‑case deployment—from predictive readmission alerts to automated triage—by leveraging existing data flows rather than building new ones from scratch. Event‑driven architectures further enable instantaneous triggers, turning data events into automated workflow actions.

Strategically, health systems must differentiate between capabilities that merit internal development and those best sourced from tech giants. Custom solutions should address institution‑specific gaps such as nuanced patient‑safety pathways, while generic AI services—like cloud‑based vision or speech APIs—can be purchased to avoid reinventing the wheel. Embedding ROI requirements into annual operating plans forces pilots to demonstrate measurable clinical or financial outcomes, ensuring that AI investments translate into sustainable value rather than speculative projects.