Most health systems replicate Clarity (and Caboodle where it is implemented) from the reporting database into the warehouse on a nightly schedule, keep the raw copy versioned, and build shared models on top. They choose between four routes: an in-house ETL build, an ingestion connector such as Fivetran's Epic Clarity connector, a consulting partner such as phData or Hakkoda that builds and runs the pipelines, or a platform such as Databasin that extracts Epic and maintains the models for you. Snowflake and Databricks can both hold Epic data well; the choice usually follows your analysts' tools and the platform you already license.
A 900-bed academic medical center licenses Snowflake on Azure. The contract was the easy part. A year later the Epic data is still arriving as CSV exports, and every Epic upgrade breaks something downstream. The platform is fine; the work between Epic and the platform was never scoped.
What has to move, and how often?
Clarity is the main relational reporting database, refreshed nightly from Chronicles, Epic's operational store. Querying Chronicles directly for analytics is generally off limits. Caboodle, Epic's star-schema warehouse, is more query friendly but its coverage depends on how much each institution has implemented.
Plan for thousands of Clarity tables, of which a few hundred carry most reporting. Decide which subject areas come first (encounters, orders and results, professional and hospital billing), whether you need history beyond what Clarity keeps, and how fresh the data must be. Clarity's nightly refresh already sets the floor: most workloads run on yesterday's data.
FHIR APIs cover standard resources and help with near-real-time needs, but they rarely expose an institution's full custom build. For cohort, revenue cycle and finance work, teams end up back at Clarity.
Who can build and maintain the pipelines?
| Route | What you get | What stays with you |
|---|---|---|
| In-house ETL | Full control over tables, timing and modeling | Engineers who know Clarity, schema changes on every upgrade, on-call |
| Ingestion connector | Managed replication of Clarity tables into the warehouse | Modeling, definitions, reports and the Epic knowledge to use them |
| Consulting partner | A team that builds pipelines and models, often with published Epic accelerators | Ongoing cost and a plan for who runs it after the engagement |
| Databasin | Epic extraction, versioned raw layer, shared models and an answer layer, implemented by our team | Your warehouse contract (or none, if you use our lakehouse) and the decisions about definitions |
Partners such as phData and Hakkoda publish their approaches to Epic data on Snowflake. Fivetran documents a dedicated Epic Clarity connector. Each fits a different staffing model; ask every option who maintains the pipeline after an Epic upgrade.
What breaks after go-live?
Epic upgrades and local build changes alter Clarity columns and tables. A pipeline that loads straight into reporting models finds out when a dashboard is wrong. Keep the raw load untouched and record its schema each night, so a change is detected at the raw layer and resolved before it reaches anyone's report.
Definitions drift too. Readmission, length of stay and denials end up calculated three ways in three reports. Write them once in a shared layer and point every report and AI question at it.
Snowflake or Databricks for Epic data?
Both platforms run on Azure, both can hold Clarity's volume, and both support governed access to PHI under a BAA. The practical differences are about people and existing contracts.
| If your team mostly | Snowflake tends to fit | Databricks tends to fit |
|---|---|---|
| Writes SQL and builds BI dashboards | Yes: warehouse-first, simple SQL scaling | Works, with SQL warehouses |
| Works in Python notebooks and machine learning | Works, with Snowpark | Yes: Spark and notebooks first |
| Needs open table formats across engines | Iceberg tables supported | Delta native, Iceberg readable |
| Already licenses one of them | Use it; the Epic work is the same | Use it; the Epic work is the same |
The Epic extraction, schema tracking and shared definitions are the same work on either platform. Choosing the platform first and leaving that work unscoped is the common reason projects stall.
Where does Databasin fit?
Databasin was built at Washington University School of Medicine because its own Epic pipelines kept failing. Its Epic connector extracts from Clarity directly, supports Caboodle where an institution has it, and treats FHIR as a supplemental stream. Every load lands in a versioned raw layer, shared definitions live in one layer above it, and the result lands in your Snowflake or your Databricks Unity Catalog, or in an open Iceberg lakehouse inside your own Azure tenant. People can then ask questions in plain English and see the SQL behind each answer.
Our team implements the connection, the first subject areas and the definitions with your Epic and analytics staff. The connector uses your organization's own Clarity access, the same as any in-house ETL job, so your Epic team's data governance still applies.
Example: a first Epic workload on either platform
An illustrative first month, the same on Snowflake or Databricks:
| Week | Work | Result |
|---|---|---|
| 1 | Agree the first question and the Clarity subject areas it needs | A scoped list of tables and definitions |
| 2 | Land nightly loads with schema tracking | A raw copy that shows upgrades as changes |
| 3 | Build shared definitions and reconcile against an accepted report | Numbers the reporting team recognizes |
| 4 | Hand the governed tables to analysts and BI | The first report running on the warehouse |
Checks before you start
- Pick the first questionStart with one report your leaders already use, so you can reconcile against it.
- Confirm access and approvalsAgree with the Epic team how Clarity is read and who approves new subject areas.
- Plan for upgradesAsk how each route detects a Clarity schema change before reports break.
- Name the ownerDecide who maintains pipelines and definitions a year after go-live.
Sources
Fivetran, Epic Clarity connector documentation, accessed 2026 · fivetran.com/docs/connectors/databases/epic_clarity
phData, How to activate EHR data from Epic and Caboodle into Snowflake, accessed 2026 · phdata.io/blog
Hakkoda, Epic healthcare data, accessed 2026 · hakkoda.io/resources/epic-healthcare-data
Databasin, Epic data architecture, 2026 · databasin.ai/tech/epic
Discuss your Epic and warehouse plans
We can walk through your Clarity setup, the platform you already license and the first workload worth moving.