The Cost of Moving Data: Delta Sharing with Adobe Experience Platform
Picture this: the data team finishes a purchase propensity model. It is governed, tested and ready. Marketing still cannot use it, because it needs the CSV file, the SFTP transfer and the process someone has to maintain so everything arrives on time. How much does it really cost you to move data from one team to another?
Databricks gave that cost a name: the "integration data tax". On April 20, 2026, together with Adobe, it announced a way to stop paying it. Before I explain it, a note on availability, because the pages do not agree with each other here.
According to the Databricks blog, Delta Sharing support for Adobe Experience Platform was announced that day, and anyone who wants to explore it joins a waitlist. Adobe's documentation, updated on October 21, 2026, says something else: that the limited beta has ended and general availability is planned for 2027, with the Adobe account team as the contact. The usage guide in that same documentation still mentions a limited beta "until August 1, 2026". My reading is that today this is an announcement with a feature on the way. To activate it, you will have to wait.
Delta Sharing with Adobe Experience Platform: from the file that travels to the table you query
According to Databricks, until now marketing teams waited on CSV exports, SFTP transfers and costly ETL processes just to activate their own data in Experience Platform. Databricks summarizes three problems:
Latency: data can take 24 to 48 hours to reach marketing systems.
Cost: moving large volumes between clouds generates egress charges and duplicated storage.
Governance: copying the data breaks the lineage and the access controls defined in the catalog.
What Databricks describes is a change of direction. Experience Platform becomes a recipient of Delta Sharing, an open protocol. The data team prepares governed tables and shares credentials. In Experience Platform those tables appear as virtual tables, and with Adobe Data Distiller they can be queried without moving the records.
Adobe's documentation confirms it with more precision. The data is represented as virtual, read-only datasets. Only metadata is stored, such as the schema, the lineage and the connection details, and the records stay where they are.
The same idea has another name at Salesforce
If you follow the Salesforce world, this sounds familiar. Salesforce explains the Zero Copy concept on its Latin America blog: accessing data in another warehouse without moving, copying or reformatting it. According to that page, in the traditional model data is only accurate as of the last sync point, and in the Zero Copy model the data stays in its original location.
My reading is that the market is converging on one idea: the data stays where it is born and the marketing system queries it. Two cautions. The Salesforce page shows no date and its example uses Snowflake. And what each vendor says about its own method belongs to that vendor. I read it as a signal of where the platforms are heading.
When yes and when no
Sharing without copying also brings upfront work. According to Adobe's documentation, a Databricks administrator has to create the share, the recipient and the credentials, and those credentials carry an optional expiration date. Adobe also asks for concrete things.
When yes:
When your tables are Delta tables and already governed, with an owner and clear definitions.
When your Experience Platform organization is in a supported Azure region and the feature is enabled, something Adobe lists in its requirements.
When the value lies in using a model or an audience that already exists, for example a churn score, and you can save yourself another pipeline.
When no, at least for now:
If you depend on tables with row filters or column masks. According to Adobe's documentation, Databricks does not allow sharing that kind of table.
If the data team has not tidied its tables. Sharing a dirty table only makes the problem faster.
If you need to solve something this quarter. With general availability planned for 2027, it is wise to wait before building a plan on this feature.
The uncomfortable question is about responsibility: who answers for the table that marketing is going to consume live? When data traveled in a file, the file was a pause that left time to review. With a direct query, that pause disappears. If your data team and your marketing team still disagree on what "active customer" means, Delta Sharing will not decide it for them.
Before you think about the pipeline, review your tables. And if you move files every week today, at least count how many hours it costs you. That number is your own cost of moving data.
Sources:
Comments