Transforming Healthcare Referrals with Fivetran, Agentic AI, and Databricks Genie

Every health system stands to benefit immensely from AI: from improving patient outcomes and reducing clinician burnout to optimizing the revenue cycle. Despite an influx of technology, hospitals and health systems continue to struggle to realize this value.

For healthcare delivery organizations, the core problem is a highly fragmented data and AI estate. Patient records in the electronic health record (EHR), medical imaging in PACS systems, streaming telemetry from bedside monitors, and unstructured clinical notes are all siloed. This fragmentation leads to disjointed security policies, the lock-in of critical operational and clinical data, and AI sprawl without proper enterprise semantics.

Furthermore, building and productionizing HIPAA-compliant AI agents is incredibly difficult, especially when trying to balance computational cost versus clinical quality and safety.

Maximizing AI Potential Starts with Your Data

Databricks unifies and democratizes all of your data and AI, designed to handle the complexity of healthcare environments safely and efficiently. But to build a single source of truth, you first need to unlock your data from legacy silos. Fivetran can help. 

Traditionally, extracting data from complex EHRs meant engineers spending weeks building brittle, custom ETL scripts. Today, Fivetran’s Epic Clarity connector accelerates this process by providing a fully managed, zero-code pipeline. It automatically and securely extracts rich clinical and operational data from Epic Clarity and syncs it directly into Databricks.  

By automating change data capture (CDC) and schema mapping, Fivetran eliminates manual extraction efforts, allowing health systems to flow critical data into Databricks in minutes rather than months.

With Fivetran feeding the platform, the Databricks strategy is built on two foundational steps:

• Develop a common data foundation: Unify your hospital's data architecture with the lakehouse architecture for massive-scale data warehousing, Lakeflow for ingest and streaming (perfect for real-time HL7/FHIR feeds and IoT bedside monitors), and Lakebase for serverless Postgres. This unified foundation delivers the lowest Total Cost of Ownership (TCO) and breaks the cycle of vendor lock-in typical of legacy healthcare IT.
• Unify all of your governance: Leverage Unity Catalog to provide unified, strict governance for all assets. In healthcare, this is critical for PHI and HIPAA compliance. Unity Catalog governs not just traditional tables, but also predictive models, AI agents, PDFs (like scanned referral documents), and raw files.

Optimizing Referral Management: Start with Your Enterprise Context

Once a governed data foundation is established, health systems can deliver their unique clinical and operational context to AI. By applying AI that actually understands your specific hospital protocols, patient populations, and payer contracts, Databricks enables transformative use cases for healthcare delivery.

For hospital leadership, patients receiving timely specialty care while keeping them within the health system's network is a top priority. Optimizing referral management to stop network leakage is a massive challenge. Up to 50% of referrals fall into a "black hole" because the data is fragmented across paper faxes, Epic Clarity databases, and external payer systems. 

They simply lack visibility into whether a patient actually scheduled or attended their follow-up.

By tapping into the full potential of Fivetran-ingested data within Databricks, leaders can unlock immense business value. AI models can:

• Predict referral no-shows
• Identify root causes of out-of-network leakage, and
• Automatically recommend the next best in-network specialist based on availability and proximity.  

Once data has been unified under a common platform, your data team will need to model that data and prepare it for reporting and activation. Fivetran offers a robust toolset for transforming the data, including in-platform Quickstart data models for common SaaS platforms, integrations with dbt and Coalesce, and net new capabilities like:

• AI Column Builder to enrich your data with additional fields, powered by your AI provider of choice like ChatGPT, Claude, and Gemini
• Audience Builder to build custom audience segments

Then, when that data is ready, you can push it into the applications where your team works, to help them use data to drive decisions in their daily platforms. 

Data activation ensures that data is not siloed in dashboards, static reports, or data warehouses. Healthcare administrators and medical professionals need to access sensitive, timely data with a minimum of friction or delay. It isn’t practical for doctors and nurses — or their staff — to learn SQL and spend hours trawling through business intelligence platforms when their attention is urgently needed elsewhere.

Fivetran Activations supports the end-to-end management of referral lifecycles by integrating data from orders, scheduling, and completed visits. With that data, you can model patient journeys across providers, flows, capacity and leakage across provider networks, and standardize referral metrics.

Critical insights from these models can then be sent to operational systems of all kinds, whether to inform end users or trigger automated actions.

Building Production AI Agents and Democratizing Insights for Care Teams

With Fivetran and Databricks, we’re making it simple to empower health systems to build and deploy secure, clinical-grade data and AI apps.

By interacting with your data in natural language with AI/BI Genie, hospital administrators, nursing managers, and medical directors can literally talk to their data. They can explore questions about patient flow, readmission rates, or ED wait times and receive instant insights via natural language, data tables, and visualizations.

Health systems can deploy production AI agents that achieve higher quality at a lower cost. For example, similar to how AstraZeneca utilized Agent Bricks to parse through over 400,000 complex clinical trial documents, a hospital can deploy agents to extract structured data points from decades of historical unstructured medical records—without requiring their IT teams to write complex code.

By accelerating data ingestion with Fivetran and unifying analytics and AI on Databricks, health systems can move from fragmented IT to a truly intelligent healthcare delivery network.

Join us at HIMSS!

If you’re headed to HIMSS26, check out our session at the Databricks booth (#5453), Wednesday, March 12:00 - 12:00 PM PT.  And if we don’t get a chance to see you in Vegas, join us at our upcoming webinar, Wednesday April 15 at 9am PDT (link).