Which Databricks Salesforce Connector Should You Use?

Databricks integrates with Salesforce through Salesforce Data 360 and Databricks Lakehouse Federation. Databricks ships two zero-copy connectors that read Data 360 data without moving it, plus an ingestion connector that copies records for batch work. Which one fits depends on whether you need live analysis in Databricks or governed customer data in Salesforce.
TL;DR
- Databricks ships three Salesforce connectors: two zero-copy Data 360 connectors and one ingestion connector.
- Zero Copy reads data where it lives — no ETL, no duplicate copy to govern.
- Data 360 is the connection hub; unify CRM data there before connecting.
- Batch copying suits snapshots and model training; zero copy suits in-place analysis.
Why this matters
Salesforce holds the customer record. Databricks holds the analytics horsepower. Until the two talk, teams export spreadsheets by hand or run pipelines that duplicate the same rows in both systems — and every duplicate copy is a second set of access rules, a second source of truth, and a second thing to break. Data integration work removes that seam, and the Salesforce–Databricks partnership does it at the data layer instead of the application layer, which is why it matters more in 2026 than any point-to-point workaround.
How does Databricks connect to Salesforce in 2026?
Databricks' own Lakehouse Federation documentation lists three Salesforce connectors:
- Data 360 query federation: Databricks reads Data 360. How data moves: Zero copy — SQL runs against Data 360 without moving data. Best for: Live analysis of Salesforce data in Databricks.
- Data 360 file sharing: Databricks reads Data 360. How data moves: Zero copy — Databricks reads Data 360's underlying cloud storage directly, with no JDBC hop. Best for: Federating large volumes quickly.
- Salesforce ingestion connector: Copies into Databricks. How data moves: Batch copy from Salesforce products, including Data 360 and Sales Cloud. Best for: Historical snapshots, reporting extracts, ML training sets.
The two zero-copy connectors serve one question: how do I analyze Salesforce data in Databricks without building a pipeline. The ingestion connector serves a different one: when is an actual copy the point.
What does Zero Copy actually do?
Zero Copy removes the extract, transform and load step between the platforms. Salesforce describes it as bidirectional in 2026:
- Into Data 360. Databricks tables become readable inside Data 360 without a copy — Salesforce calls this Bring Your Own Lake, and it is how an existing data lake stays exactly where it is.
- Back out to Databricks. Unified and enriched customer data in Data 360 is readable in Databricks without replication, so analysis sees what is in Data 360 now rather than last night's export.
What you stop doing: maintaining a second copy of customer records, reconciling two versions of the same field, and paying to store and move data that never needed to move. What you start doing: agreeing which side governs access to which tables, because with no copy in between, access rules matter more, not less.
When is a straight copy the right call?
Zero copy is not automatically the answer, even in 2026. A scheduled ingestion copy fits better when:
- You need a durable historical snapshot for model training or audit.
- The consuming workload runs offline — nightly dashboards, batch scoring.
- You are migrating or consolidating records, where the move itself is the task.
Copying is also what standard bulk tooling does: Salesforce's Data Loader guide describes it as a client application for bulk insert, update, delete and export of records, moving CSV files either way. Fine for one-off moves; not a substitute for a governed connection.
What are the limits of the integration?
Honest cons, because every option has them:
- Coverage depends on Data 360. The zero-copy connectors read Data 360 data — if a CRM object never gets into Data 360, Databricks will not see it without the ingestion connector.
- Federation has query latency. Reading in place is convenient, not instant; heavy analytic workloads may still justify a materialized copy.
- Permissions become a design task. With no copy in between, access rules on both sides have to be deliberate and testable.
- Two platforms to operate. The integration removes data movement, not the operational reality of running a CRM and a data platform.
How do you set the integration up?
- Confirm Salesforce Data 360 is enabled and your key objects are available in Data 360. What you check: The objects you care about are queryable in Data 360.
- In Databricks, create a Lakehouse Federation connection, pick the Data 360 connector type, and create a foreign catalog. What you check: The connection authenticates with your credentials.
- Run a federated query against one Data 360 object. What you check: Rows return in Databricks SQL with no ingestion job involved.
- Decide the second direction: expose Databricks tables to Data 360 via Bring Your Own Lake. What you check: Data 360 reads your lake tables without copying them.
- Document who can query what on each side. What you check: Role-based access is written down and testable, not tribal.
Step 5 is the one teams skip and then regret. With zero copy, the query reaches into the other platform's data — the access model has to be deliberate.
What should you decide before you connect anything?
The connector choice is quick. These decisions take longer, and they determine whether the integration pays off in 2026:
- Which customer data goes into Data 360 first. Start with the objects a Databricks workload actually needs, not everything the CRM holds.
- Who owns access rules. Zero copy means a query on one platform reaches data governed on the other. Decide which side is the source of truth for permissions.
- Which workloads need fresh data and which need a snapshot. Live dashboards suit federation; model training and audit extracts suit a copy.
- What happens to an existing lake. If your Databricks tables already hold the data you want in Salesforce, Bring Your Own Lake lets Data 360 read them where they sit.
A small first scope is the safer route. Pick one use case, connect it end to end, and confirm the access rules hold before widening the integration. The same approach works when a Snowflake setup is already in place: existing warehouses can stay where they are while Data 360 reads from them.
Where does this fit in a wider data plan?
A connector moves nothing useful if the underlying records are inconsistent. Duplicate accounts, mismatched identifiers and free-text fields all follow the data into Databricks. Cleaning the CRM side first, and agreeing on identifiers, means the unified customer view in Data 360 is worth analyzing. SOLVD.cloud's data practice covers cleanup, migration and the Databricks work as one engagement rather than three separate projects.
FAQ
Does Databricks integrate with Salesforce without middleware?
Yes. In 2026 Databricks connects to Salesforce Data 360 directly through built-in connectors, so no integration platform sits between the two for zero-copy reads. Complex routing across many systems can still justify an integration layer.
What is Zero Copy in the Salesforce and Databricks partnership?
Zero Copy lets each platform read the other's data without extracting and loading a duplicate. Databricks reads Data 360 data in place, and Data 360 reads Databricks lake tables in place through Bring Your Own Lake.
Does Salesforce Data 360 replace Databricks?
No. Data 360 unifies and activates customer data for CRM use; Databricks runs large-scale analytics and machine learning. The partnership exists precisely because teams keep both.
Can Databricks read standard Sales Cloud objects?
The zero-copy connectors read Data 360 data, so standard CRM objects should be brought into Data 360 first. The ingestion connector copies data from Salesforce products including Sales Cloud directly.
Do I need Data 360 to connect Databricks and Salesforce?
For zero copy, yes — Data 360 is the connection point on the Salesforce side. Without it, the realistic options are the ingestion connector or a custom pipeline.
One last thing
The connectors are the easy part. The questions that decide whether this pays off in 2026 are which customer data lands in Data 360 first and who governs access on the Databricks side. SOLVD.cloud is an official Databricks partner and a Salesforce consulting firm, so if you want a scoped opinion on the sequencing, book a conversation — bring the project, and SOLVD.cloud will tell you what the data work actually involves.