How to Connect Salesforce Data Cloud and Databricks Without Moving Your Data

TL;DR

Zero-copy integration lets Salesforce Data Cloud and Databricks share data without copying or moving it. Salesforce can query data that lives in Databricks directly, and Databricks can read Salesforce CRM data via file federation, all governed through Unity Catalog. No ETL pipelines. No duplicated storage costs. This guide covers how it works, when to use it, and how to set it up.

What Is Zero-Copy Integration?

Zero-copy integration is a data access pattern where one system queries data directly from where it already lives in another system, without extracting, transforming, or loading it into a new location.

In the context of Salesforce and Databricks, it works in both directions:

• Inbound Zero Copy: Databricks (or Snowflake) data appears inside Salesforce Data Cloud as a live, queryable source. Salesforce pushes queries down to the external platform rather than pulling data in.
• Outbound Zero Copy: Salesforce CRM and Data Cloud data gets shared to Databricks via file federation. Databricks can then run analytics, train ML models, or build dashboards on top of Salesforce data without a separate ETL job.

The concept draws on two underlying technologies: Delta Sharing (an open protocol for secure data sharing across platforms) and Lakehouse Federation (a Databricks capability for querying external data sources without ingestion).

Why This Matters for Enterprise Data Teams

Traditional integration between a CRM like Salesforce and an analytics platform like Databricks involves a pipeline that extracts data from Salesforce, transforms it, loads it into the lakehouse, and then keeps it in sync. That pipeline has real costs:

• Storage duplication: data lives in two places simultaneously
• Latency: the lakehouse always reflects a snapshot, not live data
• Pipeline maintenance: every schema change in Salesforce can break the pipeline
• Governance complexity: data lineage tracking across two copies is harder

Zero copy removes all of this. Data stays in its authoritative location. Governance stays with the owning system. Queries run against live data.

For a CEE enterprise running SAP for financials, Salesforce for CRM, and Databricks for analytics, zero copy means a single unified data layer that all three systems can share without anyone having to build or maintain a replication job.

Two Integration Directions

Direction 1: Databricks Data Into Salesforce Data Cloud (Inbound Zero Copy)

This direction is useful when your primary analytics or ML data lives in Databricks, but your business users need to access it inside Salesforce, for example to power personalisation, customer scoring, or AI agents.

Salesforce Data 360 supports inbound zero copy via two mechanisms:

• Query Federation: Salesforce pushes SQL queries down to Databricks at query time. Results appear in Salesforce without any data being stored there.
• File Federation (Public Preview, 2025-2026): Salesforce Data 360 Objects are shared as Parquet files into Databricks Unity Catalog, where they become queryable as external tables. Databricks blog

Direction 2: Salesforce CRM Data Into Databricks (Outbound Zero Copy)

This direction matters when your data science or analytics team works in Databricks and needs access to Salesforce opportunities, accounts, contacts, or Service Cloud cases for model training or reporting.

Salesforce exposes CRM data through Zero Copy File Federation, sharing it directly into Databricks Unity Catalog. The Databricks team can then run Spark jobs, notebook analysis, or AutoML experiments on Salesforce data without any extract step.

Architecture: How It Actually Works

The integration relies on three components working together:

Salesforce Data 360 (Data Cloud)
         |
    Delta Sharing Protocol
         |
Databricks Unity Catalog
         |
  Databricks Compute (Spark / SQL Warehouse / ML)

Unity Catalog acts as the central governance layer. When Salesforce data is shared into Databricks, it appears as a catalog object governed by Unity Catalog. Access policies, lineage tracking, and audit logs all flow through Unity Catalog.

Delta Sharing is the wire protocol. It is open-source and platform-agnostic, which means the same pattern works across Snowflake, Google BigQuery, and other compliant platforms.

Setting Up Inbound Zero Copy: Databricks to Salesforce

Prerequisites:

• Salesforce Data Cloud license (or Data 360 edition)
• Databricks workspace with Unity Catalog enabled
• Admin access to both platforms

Steps:

1. In Databricks, create a share in Unity Catalog: CREATE SHARE salesforce_share;
2. Add the target tables or schemas: ALTER SHARE salesforce_share ADD TABLE catalog.schema.tablename;
3. Create a recipient for Salesforce: CREATE RECIPIENT salesforce_dc;
4. In Salesforce Data Cloud admin, navigate to External Data Sources and add a new Databricks connection using the share activation link
5. Map the shared tables to Data Cloud Data Model Objects (DMOs)
6. Test with a SOQL query or a Data Cloud segment that references the external table

Setting Up Outbound Zero Copy: Salesforce CRM Data to Databricks

Prerequisites:

• Salesforce Data Cloud or CRM Analytics Plus license
• Databricks workspace with Unity Catalog enabled
• Zero Copy File Federation feature enabled on your Salesforce org (GA as of 2025)

Steps:

1. In Salesforce Setup, open Data Cloud > External Connections > Databricks
2. Authenticate with your Databricks workspace using OAuth or personal access token
3. Select the CRM objects to expose (Accounts, Opportunities, Cases, Custom Objects)
4. Map to a target Unity Catalog schema in your Databricks workspace
5. In Databricks, run: SHOW TABLES IN salesforce_shared_catalog.crm
6. Run a test query: SELECT AccountName, AnnualRevenue FROM salesforce_shared_catalog.crm.account LIMIT 100;

The June 2026 Update: MuleSoft Agent Scanner for Databricks

In June 2026, Salesforce announced an expanded partnership with Databricks, adding the MuleSoft Agent Scanner for Databricks as part of the Data 360 Zero Copy ecosystem. This tool lets AI agents running in Salesforce discover and query data assets stored in Databricks Unity Catalog automatically, without a human configuring each connection.

Broader capabilities are rolling out through H2 2026. For teams planning a Salesforce + Databricks architecture, this makes the investment significantly more future-proof.

When Zero Copy Is the Right Choice

• Your data already has a governed home. If your Databricks environment is your system of record for analytics data, there is no point duplicating it into Salesforce.
• You need near-real-time access. ETL pipelines introduce lag. Zero copy queries live data.
• You are managing cross-functional access. Your marketing team can use Salesforce Data Cloud segments powered by Databricks data without needing a Databricks license.
• Governance matters. Regulated industries (finance, pharma, healthcare) need to keep data in one place with clear lineage.
• You want to avoid pipeline debt. Every ETL job is a liability.

When Zero Copy Is Not the Right Choice

Salesforce Ben outlines the key counter-cases:

• When your source data is not well-maintained. Zero copy exposes live data. If the source schema changes frequently without notice, downstream queries break.
• When you need heavy transformation before Salesforce can use it. Zero copy does not run complex transformations efficiently at query time.
• When network latency is a constraint. Cross-cloud queries add latency. If your Salesforce org and Databricks workspace are in different cloud regions, query performance will reflect that.
• When your Databricks workspace does not have Unity Catalog. The older Hive metastore does not support Delta Sharing.

Common Implementation Challenges

1. Unity Catalog not enabled
Many older Databricks workspaces still run on the Hive metastore. Migrating to Unity Catalog is the first step and can take 2-4 weeks for a complex workspace.

2. Salesforce Data Cloud edition requirements
Zero copy features require Data Cloud or Data 360 licenses. Check your contract before designing around them.

3. Network connectivity
If Databricks runs behind a private VNet or AWS PrivateLink, Salesforce needs a corresponding network connection configured.

4. Schema drift
When a Databricks table schema changes, the Salesforce DMO mapping breaks silently. A schema change notification process between teams is essential.

5. Token expiry and authentication
OAuth tokens used for the integration expire. Implement a rotation mechanism or use service principals rather than personal access tokens for production.

How Forest Digital Helps

At Forest Digital, we design and implement Salesforce and Databricks integrations for enterprise teams across Central and Eastern Europe. Our AI-driven development approach means we can set up a working zero-copy integration, including Unity Catalog migration, Data Cloud connection configuration, and governance policy setup, significantly faster and at a fraction of the cost of traditional system integrators.

If your team is planning a Salesforce + Databricks architecture or already has both platforms and wants to reduce pipeline complexity, contact us at forest-digital.com.

Conclusion

Zero-copy integration between Salesforce Data Cloud and Databricks gives enterprise teams a fundamentally better data architecture: no duplication, no pipeline debt, governed access, near-real-time data. The June 2026 MuleSoft Agent Scanner update makes it even more relevant for teams building AI agents on top of unified data.

The setup involves Unity Catalog, Delta Sharing, and either inbound or outbound federation depending on which direction your team needs. Both directions can coexist in the same architecture.

For CEE enterprises running SAP, Salesforce, and Databricks in parallel, zero copy is the foundation of a clean data ecosystem. The days of ETL pipelines connecting your CRM to your lakehouse are ending.