Healthcare AI trust depends on foundations built on data, not just algorithms, according to executives at Snowflake and Komodo Health. As artificial intelligence moves from pilot projects into production, the industry is realizing that a compelling demo doesn’t guarantee clinical trust. The most stubborn problems are structural, and solving them requires a data foundation built before any model is trained.
The 46,000-Word Chart Problem
Jesse Cugliotta, vice president and global head of healthcare and life sciences at Snowflake, pointed to a stark example: the average patient chart is 46,000 words — roughly the length of “Fahrenheit 451.” If a physician walking into an emergency room has 27 other patients, they cannot read through an entire chart history, even if they could access it. “The ability to leverage AI is to provide a more complete picture,” Cugliotta said, which has a critical impact on caregiver experience and patient outcomes.
Related: Cyera Doubles Valuation to $12B in Five Months
Healthcare data is notoriously fragmented, with claims, labs, and prescriptions scattered across systems. No single organization holds a complete picture of a patient’s health journey, making AI training incomplete without a unified data layer. Komodo Health has spent a decade assembling that foundation, stitching together more than 350 million patient journeys across its Healthcare Map, according to Amit Sangani, the company’s chief technology officer. The industry still relies on fax-era workflows, Sangani noted, and AI is only as good as the data it can access.
“When a patient goes to a doctor, gets diagnosed, then gets a prescription, the data is all over the place,” Sangani said. “How do you combine all of that data for a single patient and create a full journey? You aggregate that across different segments or therapeutic areas. Once you do that, it becomes really powerful.”
Related: Disaggregated infrastructure transforms private cloud landscape
In the world of AI, deterministic answers are difficult to get, Sangani noted, but by building guardrails, Komodo can provide them to customers. That transparency is the competitive differentiator for the company’s Marmot platform, which layers multi-agent orchestration on top of the Healthcare Map using Snowflake for storage, compute, and governed caching.
Trust Through Auditable Steps
A Snowflake research study found that 85% of healthcare leaders view interoperability as foundational to scaling AI. That finding maps directly to Komodo’s architecture, where every analytic step is logged, auditable, and reproducible. For life sciences analysts running deep research on GLP-1 therapy migration patterns or clinical trial cohort design, each step in the workflow can be inspected, modified, or replicated, Sangani explained.
Related: The Shifting Landscape of News Consumption
Healthcare providers, he said, don’t want just an answer. They want to understand the underlying data and how the answer was derived. “The first step is the plan,” Sangani said. “How is the AI going to create a plan and help me understand what cohorts they are creating, how they are doing the filtering, what SQL is being written, what Python basically generates the report. That gives them the trust that when the final answer comes, it’s like a kid doing a math problem. If they just give you an answer, you say: ‘What are the steps you followed?’ That’s exactly what we are doing.”
The industry’s shift toward production AI is forcing organizations to confront legacy technologies like fax machines and disjointed electronic health records.
