Databricks has added a Genie plugin for ChatGPT that lets employees ask questions about SAP ERP data their organization has made available in Databricks. A buyer could identify late purchase orders that threaten next week’s production plan. A controller could trace a margin variance to the postings behind it. Genie uses company-specific business definitions, runs the analysis in Databricks, and returns results with links to the underlying sources.
Databricks listed the plugin in its November 11, 2026, release notes, two days after Genie Ontology became generally available. The ontology supplies the business definitions Genie uses to interpret questions. Databricks’ November 13 update describes the plugin as part of its multiyear partnership with OpenAI, announced in September 2025. For SAP customers, the route into ChatGPT builds on an existing integration between SAP Business Data Cloud and Databricks.
How SAP ERP Data Reaches ChatGPT
A company that already runs Databricks can use SAP Business Data Cloud Connect to share selected SAP data products into its existing workspace. SAP and Databricks administrators establish the connection, and shared products become available through Unity Catalog, the Databricks service that governs access to data assets. Databricks describes this as zero-copy sharing through OpenSharing. The workspace can query the published data product without maintaining a separate replica of it. SAP also offers an embedded deployment called SAP Databricks within SAP Business Data Cloud.
Zero-copy sharing describes how Databricks accesses the published data product. The freshness of an answer depends on when ERP changes reach that product and how recently any other datasets used in the query were updated. A planner looking at open purchase orders and warehouse stock may be combining datasets with different refresh schedules. The time of the latest update matters if a large receipt arrived after one of those datasets refreshed.
On the ChatGPT side, a workspace administrator publishes the Genie plugin for the relevant Databricks workspace. Employees then sign in with their own Databricks accounts. If the Databricks workspace restricts inbound traffic, its administrators may also need to allow the connection. ChatGPT sends the question to Genie One through the Model Context Protocol, or MCP. Genie identifies relevant data and definitions, generates and runs SQL when needed, and returns the result with links to Databricks sources. Unity Catalog enforces the permissions of the signed-in user.
What Genie Needs to Understand About SAP Transactions
A sales order, a shipment, an invoice, and a revenue posting can all represent stages of the same sale. They can also fall in different fiscal periods. When a controller asks why revenue fell last month, the answer depends on whether the query uses recognized revenue, billed sales, or booked orders. It also depends on the ledger, currency, and treatment of returns and intercompany transactions. A query can run successfully and still answer a different business question from the one the controller intended.
Genie Ontology combines definitions explicitly modeled in Unity Catalog with context Genie learns from sources such as dashboards and saved queries. An approved revenue metric can specify the ledger, fiscal calendar, and currency conversion rules that finance uses for reporting. SAP Business Data Cloud can also supply technical metadata, including field descriptions and key relationships, when its shares are mounted in Unity Catalog. Those relationships help Genie interpret and join data. The business team defines which transactions count toward the metric.
The same issue arises in procurement and inventory. A purchase order can have several schedule lines, each with its own delivery date and quantity. A stock figure may include material under quality inspection or already reserved for another order. An ERP team preparing data for Genie needs to decide which records answer each question and document the rule where Genie can use it.
Where the Integration Could Help ERP Teams
Finance teams investigating a margin variance
A controller could ask which company codes and product families accounted for the largest gross margin shortfall against plan, then review the postings behind each movement. The analysis would need actuals and plan data on a common currency basis, with an agreed treatment of freight, rebates, returns, and late adjustments. If a cost posting explains a material share of the variance, the controller could follow the source link and check the posting date and account against the close report. That gives the team a way to investigate an exception while keeping the ledger and approved reports as the basis for the final figure.
Buyers checking which late orders threaten production
A buyer could ask which purchase order schedule lines due in the next seven days have confirmed receipt dates that fall after production needs the material. To answer, Genie would need open quantities, supplier confirmations, goods receipts, and production requirements at the right level of detail. A partially received order can still leave a shortage, even if its header shows that a delivery has taken place. A useful result would identify the plant, material, remaining quantity, and relevant schedule line so the buyer can decide where to expedite.
Inventory planners looking for stock they can transfer
A planner could ask which plants have stock available to cover a shortage at another site within the current planning window. The query would need to account for reservations, blocked stock, lot restrictions, units of measure, and transfer lead times. A group can have enough inventory on paper while the only stock that can arrive on time is already committed. Combining ERP inventory with warehouse or transport data can help the planner identify feasible transfers, provided the result makes clear when each source was refreshed.
What It Takes to Act on an Answer
The Genie plugin gives ChatGPT an interface for ERP analysis. Changing a purchase order, releasing a blocked invoice, or posting an inventory adjustment requires a separately authorized ERP API or workflow. An agent could prepare a proposal to expedite three purchase orders and attach the production dates and shortage quantities that support it. The purchasing workflow would determine who can approve the change and record it in SAP.
Databricks’ November update also describes how developers can build custom agents using the OpenAI Agents SDK and deploy them through Databricks Apps. In an ERP setting, a custom application could combine an analytical result with an existing exception workflow. For example, a shortage monitor could open a review task for a planner, who would check the stock position before approving a transfer. Each action needs an assigned identity, defined permissions, and a record the process owner can review.
What SAF-Holland Shows About the SAP Data Layer
SAF-Holland, a manufacturer of chassis components for trucks and trailers, uses SAP Business Data Cloud and SAP Databricks to analyze more than 100 million SAP records, according to a case study published by Databricks. The data includes scheduling agreements, deliveries, customers, and materials. SAP S/4HANA remains its system of record. The analytics team compares expected demand with actual deliveries to see how customers revise orders as delivery dates approach.
David Gund, senior manager of global digital intelligence at SAF-Holland, estimated that implementing similar analytical workflows elsewhere would have taken at least five times as long. That estimate applies to the company’s SAP and Databricks analytics work. The case study provides an example of how ERP records can support production planning once the data is available for analysis. It reports no results for the ChatGPT plugin.
How an ERP Team Could Test the Integration
An existing purchase order exception report gives a pilot a clear baseline. The procurement team can ask Genie for the same exceptions using the same cut-off time and plant scope, then compare the results line by line. The test set should include partial receipts, canceled schedule lines, revised supplier confirmations, and users with different plant permissions. Reviewers can check whether the answer uses the approved definition, links to the relevant records, and makes the data timestamp visible. Tracking missed shortages, false alerts, and time to investigate gives the team a practical measure of whether the workflow is ready to expand.
The pilot should also account for what appears in ChatGPT. The source data can be queried through a zero-copy share, and query results can still appear in a conversation. An ERP team can define which fields are appropriate for that surface. A buyer investigating a late delivery may need the purchase order number, quantity, and confirmed date. Supplier bank details have no role in that analysis.
Sources
ERP News Editorial Team
We cover ERP and enterprise technology for the people who select, implement, and run these systems.
Pitch a guest article or work with our editorial team on a news story, customer case study, or executive interview.
Editorial Calendar · Pitch a Guest Article · Commission a Story
Advertising Options · ERP News Magazine · ERPNews.com
Editorial Inquiries · [email protected]
Copyright & Usage · Our original editorial content is produced by the ERP News Editorial Team, with expertise in ERP and enterprise technology.
© ERP News. All rights reserved. Except as permitted by law, reproduction, republication, and redistribution require prior written permission or a licence from ERP News, along with clear attribution to ERP News as the source.
ERP News Content Licensing · License full-text ERP News articles for your internal research tools, delivered through a private JSON feed. [email protected]
