Workday Data Cloud answers a question analytics teams have asked for years: how do we query Workday data in Snowflake without building an extract. It does not answer the question payroll, benefits, and identity teams ask every night: how do we send this system the file or the API call it will accept. Kiweely Roots, coming next, is specified for the second question. Zero-copy does not retire it. A headcount extract whose only consumer is a warehouse might be retired by it. An 834 file will not.
This article uses Workday’s Data Cloud admin guide and Snowflake’s zero-copy documentation, both read on 16 November 2026, plus the Roots page. A marketing URL on workday.com for a product named Data Cloud returned not found. The admin guide is the Workday source. Snowflake’s documentation and Snowflake’s blog do not use identical words for freshness, and both are quoted.
What Workday says Data Cloud is
Setup considerations for Workday Data Cloud, updated 18 September 2026, define it as a product suite for data flow between Workday and external platforms, using a zero-copy architecture so external systems can access Workday objects without a physical extract, transform, and load. The same page says the suite is bidirectional: query Workday data from outside, and view external data inside Workday.
Workday splits the outbound side into two tools, and the choice is a question on the page:
| Need | Workday’s tool | What the page says |
|---|---|---|
| NeedTransactional reads that need current data | Workday’s toolWorkday Live Data Query | What the page saysReal-time access via SQL. |
| NeedLarge-scale analytics where a periodic refresh is acceptable | Workday’s toolWorkday Data Lake | What the page saysIceberg tables. External platforms read where the data lives rather than exporting it. |
| NeedExternal data queried from inside Workday, without copying it in | Workday’s toolZero-Copy Data In | What the page saysPrism data-management tools. A live link to data that stays at the source. |
| NeedExternal data that needs heavy joins or a permanent copy inside Workday | Workday’s toolPrism ingestion | What the page saysMove, transform, and store. |
The requirements are not optional footnotes. Data Cloud is a separate SKU. The tenant must be public-cloud hosted. You need a contract with a supported partner. Workday’s examples are Salesforce, Snowflake, Databricks, and Google Cloud. You must subscribe to Flex Credits. Only certain business objects are in the catalog, and the page tells you to look at the Data Cloud Catalog rather than assuming every object is there. Zero-copy connectivity is limited to specific partners. It is not a general JDBC connection to an arbitrary tool.
Get started with Workday Data Lake lists who can connect: Snowflake, Google Cloud, Salesforce, Databricks (early access, via the account executive), and AWS. Setup on the Workday side includes the Data Cloud functional area, an integration system user, a registered API client, which objects and columns are published, and access roles. The page says data remains inside Workday’s security boundary until it is requested, and that Data Lake is the alternative to traditional ETL when a periodic refresh is enough.
Zero-Copy Data In is the other direction. Snowflake is one of four documented sources, with Databricks, Google Cloud, and Salesforce. Workday’s example is blending base pay and benefits, which stay in Workday, with stock compensation that stays in Snowflake, so the labor cost analysis does not require extracting the Workday side.
Security domains are spelled out: integration security for the integration system user, tenant setup for registering API clients, Data Lake access policies, and Prism domains for external catalogs and tables. An analytics connection is still an integration security problem. It is not “the warehouse is outside security.”
What Snowflake says
Snowflake’s documentation, About Workday Data Cloud and Snowflake, says the integration is generally available in commercial AWS, Azure, and Google Cloud regions, and not in Virtual Private Snowflake, government regions, or China. On the Workday side, Snowflake says it supports Workday Data Cloud on AWS only. The Snowflake account itself does not have to be on AWS or in the same region.
The mechanism on that page: you create a Zerocopy Connector, authenticate to the tenant with an API client a Workday administrator registers, and mount the catalog as a catalog-linked database. Namespaces become schemas. Tables become tables. Catalog-linked databases are read-only. Workday remains the system of record. Snowflake says it contacts Workday on each query, so the result is fresh rather than a cached copy.
That last sentence is not how Workday’s Data Lake page describes freshness. Workday says to use Live Data Query for real-time reads and Data Lake when a periodic refresh is acceptable. Snowflake’s release note of 7 October 2026 repeats the “contacts Workday on each query” wording for the generally available connector. Snowflake’s own blog, announcing the same integration, distinguishes a Data Lake path that materializes on a schedule you set from a Live Data Query path against the tenant. Use the pair, not a blended paraphrase:
- If you need the scheduled, analytics-shaped share, you are on the path both vendors call Data Lake, and Workday says the refresh is periodic.
- If you need a query that is current at execution, you are on Live Data Query, and Snowflake’s connector documentation describes a query-time read.
- Neither page, read today, says a query writes a worker back to Workday. Snowflake’s documentation says the catalog-linked database is read-only. The reverse direction Workday documents is Zero-Copy Data In and, in Snowflake’s blog, Prism as external tables. Those are reads of Snowflake from Workday, not operational updates to a worker.
Snowflake’s blog names Worker, Position, and Payroll among curated objects, with money data described as coming soon. Workday’s admin page does not give that list. It says to consult the catalog. Do not design a payroll interface on the blog’s object names. Confirm the objects in the catalog for your tenant.
What zero-copy explicitly does not do
Workday’s setup page says Data Cloud “provides an alternative to traditional integrations” and can replace custom RaaS for some extraction. The examples are cross-functional analytics, model training, and ad-hoc SQL. The examples are not:
- an 834 enrollment file a carrier’s companion guide will accept,
- a payroll file an in-country provider posts,
- a joiner created in Okta on day one and removed on the last day,
- a cost-center file NetSuite or a partner still wants on SFTP.
Those are the systems named on the Roots page. They are not in Workday’s partner list for zero-copy. Connectivity “is not available for general-purpose JDBC connections to unsupported tools.” A carrier endpoint is an unsupported tool in that sentence. So is an identity provider’s user API. You can query worker data in Snowflake and still have to tell Okta that a person started.
The Roots page also lists Snowflake under finance and data, next to NetSuite and SFTP, with the example “headcount to Snowflake every morning for finance.” Read that example against Data Cloud before you build the extract. If finance only needs to query headcount, Data Lake or Live Data Query may remove the morning file. If finance needs a file in a layout a model consumes, or a push into a system that is not a zero-copy partner, the feed remains. Roots is specified to build that feed, test it on Sandbox, and keep no rows after the run. Data Cloud’s purpose is the opposite shape: the data stays in Workday and is queried in place, under a SKU and Flex Credits. Both can be true in one estate. They are not substitutes.
A decision table
| Consumer | Fit for Data Cloud, on the pages above | Fit for an operational feed, including Roots when it is available |
|---|---|---|
| ConsumerBI tool or notebook in Snowflake, periodic freshness acceptable | Fit for Data Cloud, on the pages aboveData Lake. Confirm the object is in the catalog. | Fit for an operational feed, including Roots when it is availableA scheduled extract is the legacy version of this row. |
| ConsumerDashboard that must match Workday as of the query | Fit for Data Cloud, on the pages aboveLive Data Query. | Fit for an operational feed, including Roots when it is availableAn extract will be stale by design. |
| ConsumerAnalysis inside Workday that needs a Snowflake fact, without copying Workday data out | Fit for Data Cloud, on the pages aboveZero-Copy Data In. | Fit for an operational feed, including Roots when it is availableNot a feed. |
| ConsumerBenefits carrier, 834 or carrier API | Fit for Data Cloud, on the pages aboveNot a documented zero-copy partner. | Fit for an operational feed, including Roots when it is availableThe operational feed. Carrier rules, not SQL, define success. |
| ConsumerPayroll provider | Fit for Data Cloud, on the pages abovePayroll as a queryable object may exist for analytics. The blog names it and the catalog governs it. The provider’s file layout is a different interface. | Fit for an operational feed, including Roots when it is availableThe file or API the provider posts. |
| ConsumerOkta, Entra ID, ServiceNow | Fit for Data Cloud, on the pages aboveNot in the partner list fetched today. | Fit for an operational feed, including Roots when it is availableIdentity feeds. Roots classifies these with directory feeds: fix, test, resume, notify. |
| ConsumerPartner SFTP | Fit for Data Cloud, on the pages aboveNot a zero-copy partner. | Fit for an operational feed, including Roots when it is availableThe file. Roots names SFTP explicitly. |
| ConsumerTraining a model on compensation history | Fit for Data Cloud, on the pages aboveWorkday’s own example for Data Cloud, governed by the access role you publish. | Fit for an operational feed, including Roots when it is availableDo not also drop a full compensation file on a server for the same project. |
Security and copies
Data Cloud’s argument against extracts is security, and it is a good argument. Workday says redundant copies are the risk, and that partner-side access still has to be configured by you. Publishing compensation history into a catalog that a wide Snowflake role can query is a copy in every way that matters, even if the bytes stay in Workday until the query. The access role and the column list are the control. The page gives administrators domains specifically for which objects and fields are available.
An operational feed has the mirror problem. The file is a copy the moment it is delivered. Roots’ rule is to not keep a second copy on the platform after the run: read, sent, deleted, with times, counts, and outcomes retained. Workday’s own 834 documentation includes a document-retention service that defaults to 30 days inside Workday. Those policies answer different auditors. Data Cloud tries not to make the analytics copy. Roots tries not to make a platform copy of an operational file. A full extract into a bucket makes both copies and gets neither policy.
Flex Credits are part of the Data Cloud price, on Workday’s page. Roots does not publish a price and does not say it uses Flex Credits. Do not infer a discount, a conflict, or a requirement. They are different products. A tenant can need the SKU for Snowflake and still need a feed for a carrier.
What to do with the feeds you already run into Snowflake
- List the Snowflake consumers. If they are SQL, a dashboard, or a model, open the Data Cloud Catalog and see whether the objects and the freshness (Live Data Query versus Data Lake) match. If they do, the extract is a candidate to retire after the zero-copy query has been compared to it, not before.
- List the non-SQL consumers of the same file. A headcount file that also lands on SFTP for a partner is two interfaces. Retiring the Snowflake half does not retire the SFTP half.
- Check region and hosting. Snowflake’s page excludes Virtual Private Snowflake, government regions, and China, and it requires Workday Data Cloud on AWS. A public-cloud requirement is also on Workday’s page. A tenant or a Snowflake account outside that list does not get this integration, whatever the announcement said.
- Keep payroll files, 834 files, and identity writes on the operational path. Use Sandbox for the change, and do not treat a successful Snowflake query as evidence the carrier accepted a row.
- If the operational path is what you want an agent to build and maintain, that is the Roots list. It is coming next. Register interest with the feeds Data Cloud will not take, not with the extracts Data Cloud was built to replace.
Frequently asked questions
Does zero-copy mean Snowflake has a full copy of Workday?
Workday says the data stays in Workday and is read in place, and that this avoids redundant copies. Snowflake says catalog-linked databases are read-only and that a query contacts Workday. A query result in a worksheet is still a result someone can download. The architecture removes the pipeline copy. It does not remove every export a person can run.
Can Roots send headcount to Snowflake?
The Roots page includes that example. Prefer Data Cloud when the consumer can query. Use a feed when the consumer needs a file or a push on a schedule into something that is not a zero-copy partner. “Headcount to Snowflake every morning” is the example to re-examine, not the example to rebuild by default.
Is Databricks the same as Snowflake here?
Workday documents Databricks as a Data Lake target in early access and as a Zero-Copy Data In source. This article does not claim Databricks is generally available on the same terms as Snowflake. Workday says to contact the account executive for the outbound connector.
Kiweely Roots is coming next, for the feeds in the right-hand column. Data Cloud is the product to evaluate first for anything whose destination is a query. Building a new nightly extract into a warehouse, in Studio or in Roots, is the integration Workday’s own page is trying to make unnecessary.