Why a Secure Data Foundation Drives AI Innovation for Healthcare and the Public Sector

In the healthcare and the public sector industries, AI only works when the data underneath it is trusted, governed and easy to access. During the recent Snowflake Accelerate 2026 virtual event, several organizations shared the key steps to getting real AI results, and they all started by solidifying their data foundation first. In healthcare, that’s resulted in reducing the time lost to fragmented patient and operational data. In the public sector, connecting siloed systems has helped departments better understand what’s happening across programs and services.

Jesse Cugliotta, Global Lead for Healthcare and Life Sciences at Snowflake, opened Accelerate Healthcare and Life Sciences 2026 with this stat, drawn from a conversation with his wife, an emergency department doctor: The median patient chart contains 46,000 words. The top end chart exceeds 1 million words — the length of the entire Harry Potter series. Even with a patient's full history available, no clinician with 24 people in the queue has time to read a chart that long. But an AI agent can, if it has a trusted and accessible data foundation it needs. That data foundation is where most AI initiatives actually stall. Advancing AI models makes any data shortcomings more visible: The sharper the tool, the more exposed the data gaps become. 

Why the data foundation comes first

In Snowflake's Data Trends 2026 reports, both healthcare and public sector organizations made the challenge clear: Teams must break down data silos and achieve interoperability before they can consider downstream capabilities like AI. 

AI is only as good as the data it runs on. In both healthcare and government, where a wrong answer can mean a denied claim, misrouted case or a failed audit, starting with the data foundation is operationally critical. That foundation must also be multimodal. Clinical notes, imaging reports, claims records, case files and unstructured documents all carry information that AI agents need to reason across. But a foundation that only handles structured tables leaves most of the picture out. Beyond data quality and breadth, AI also needs context. Without semantic and business logic attached to the data — for example defining what a “prior authorization” means in this system, or what “approved” means in that workflow — an AI agent is working with numbers that don’t carry meaning. And in highly regulated industries, there’s still more to consider. Governance and access controls must be embedded in the foundation itself so that AI operates only on data it is authorized to use, every query is auditable, and sensitive records such as patient data, law enforcement files or benefits information are protected by policy. The data foundation comes first because it is the layer that enables AI accuracy, context and the guardrails these industries require before any autonomous system can go into production.

Organizations presenting at Accelerate 2026 didn't build their data foundations to support some hypothetical future AI capability. Rather, they built them to solve a specific, often basic operational problem — and the capability to do more followed from there. When it comes to tackling operational challenges with AI, building the right data foundation to support that workflow is as paramount of a decision as which AI model to deploy. It's a key step in the process that cannot be overlooked.

How public sector teams turned a stronger data foundation into AI results

Colleen Herndon, Assistant Director of GIS and Data Insights for Metro Nashville, oversees data infrastructure for more than 55 city departments. The first dataset her team brought into Snowflake was Nashville's 311 system — an enterprise application that touched on everything from waste management to pothole repairs to social services requests. "A very manageable first step," she called it. What followed was more complicated.

Nashville's permits, land and licensing system, CityWorks, is used by more than a dozen departments. When CityWorks came in alongside Snowflake, a decade of stale reporting became something else entirely. For example, developers often complain that the city takes too long to process their plans, so the relevant department pulled the data to get to the bottom of the delays. The department discovered that the delays weren’t actually due to inaction from the city, but rather plans from developers were arriving incomplete. "This data is really enabling our departments to tell their story in a way they haven't been able to previously," Herndon said.

That story also changed what was possible at the program level. Nashville's health department ran a pilot co-dispatch program where when 911 calls had a behavioral health component, they dispatched police, fire and mental health specialists. Proving the program worked required joining two datasets — fire department call records and field notes from the mental health specialists — matched patient by patient. Prior to Snowflake, they were doing this manually, one record at a time. Herndon's team connected Snowflake to the fire department's records management system, automated the specialists' field data ingestion and built dashboards on top. The program was finally able to report on its own outcomes — and has since been deployed countywide.

The Township of King in Ontario had the same problem at a smaller scale. Marco Cheng-Perri, Manager of Digital Transformation, was tracking more than 100 public KPIs — but the data behind them was siloed and largely inaccessible. Years of unstructured council minutes sat untouched. When the foundation was finally in place, wins that had already happened started surfacing: SLA compliance in streetlight maintenance quietly improved from 50% to 100% in a single year, and no one in the organization knew. "It's a major win," Cheng-Perri said. "It got lost in the sea of data." New township administrators can now ask, “Has the council ever voted on this property?” and get the answer in seconds. The AI didn't create this institutional knowledge — it was always there. But the data foundation made it findable.

Virginia State Police CDO Steve McLaughlin described data operations before Snowflake as "Spreadsheets, macros, copying and pasting, and VLOOKUPs." His team migrated records out of legacy on-premesis systems into Snowflake, then built Cortex-powered conversational interfaces, so analysts and executives could query data in plain language. A purchase order line-item lookup that once involved navigating a clunky legacy interface for minutes or hours now takes 25 seconds — the time it takes Cortex to return an answer.

New Jersey Department of Education CIO Shashiya Lembatla was explicit about sequencing: "We wanted to make sure that we built a strong foundation layer before we worked on figuring out what kind of analytics to run," she said during the Accelerate Public Sector 2026 event. With security, data quality checks and governance in place first, the department moved from annual classroom snapshots to near-real-time insights. For example, teachers are seeing what's happening in their classrooms today, not last year, and the department also began automating certification workflows to address chronic teacher shortages.

Download the “Data Trends 2026: Public Sector” ebook to see how these trends map the path forward across every function of government.

Watch the Public Sector Accelerate session on demand to hear how Virginia State Police and New Jersey's Department of Education are making that transition. 

What a trusted data foundation makes possible for regulated AI

Dr. Paige Killian, Chief Medical Officer at Innovalon, put the patient cost of administrative overhead in terms that have nothing to do with technology: "I think of the entire healthcare industry as an enormous, inverted pyramid balanced precariously on what happens in a medical encounter. Every time we add administrative burden, we reduce the time the patient gets with that healthcare professional."

Most of that burden is a data access problem, namely, that data exists but can't be reached when it's needed. Innovalon is closing that gap with agentic prior authorization workflows on Snowflake. A 200-page chart review that once required a nurse to manually locate the right codes, diagnoses and supporting evidence across a submission — a process measured in days or weeks — now resolves in minutes with access to a unified data foundation that connects siloed health data across systems. The time saved doesn’t get used for more documentation and admin, it goes back to patient care. 

The Francis Crick Institute also faced a substantial data access challenge. A researcher studying the intersection of rare cancers and long COVID needed to combine patient datasets from 38 countries. Traditional sharing methods were ruled out by re-identification risk. Trellis — a Trusted Research Environment built on Snowflake — made the seamless and secure exchange of health information possible without requiring any one institution to relinquish control of its highly sensitive data. As Infinite Lambda's Jim Ternyila described it: "Not a vault around one piece of data, but a trusted environment that multiple parties can access from multiple locations simultaneously." Without this secure data infrastructure that enabled collaboration at scale across regions, it would have been arduous for the science to move forward.

The thread connecting the stories of Nashville's 311 integration to the Crick's Trellis also runs through Township of King and Innovalon. In each case, the data was already there. The value was already inside it. What was missing was the infrastructure to harness it. The foundation didn't generate the outcome — it gave organizations their first real way to reach data that they’d been accumulating but was siloed and untouched for years. It made the outcome visible, shareable and actionable for the first time.

To dive deeper into creating a trusted data foundation for AI and the role of an open semantic layer, we encourage you to check out the ebook “Data Trends 2026: Healthcare and Life Sciences.” It explores how AI fails to scale in healthcare not because the models aren't good enough, but because data doesn't carry consistent meaning across systems — and that until clinical terms, diagnoses and metrics mean the same thing everywhere, no AI agent can be trusted to act on them. 

Watch the Healthcare and Life Sciences Accelerate session on demand. 

Download the Data Trends 2026: Healthcare and Life Sciences ebook to see how these trends map the path from fragmented data to trusted AI across every part of the organization.

What the leaders at Accelerate understand

Both healthcare and public sector organizations operate under compliance requirements that many organizations treat as a constraint on speed, such as HIPAA, FedRAMP High or Department of War (DoW) IL5. The organizations at Accelerate had a different interpretation. Building a trusted data foundation — governed and traceable with access controls enforced from the start — supports compliance with those standards. And it is exactly that foundation, not the certification itself, that removes the adoption hurdles for agentic AI in highly regulated industries. The investment in trust doesn't trade off against speed; it makes AI possible.