SAP prepares to add Joule generative AI copilot across its apps

By year-end, users of a couple of SAP applications should have the option to ask generative AI copilot, Joule, to help with their work — and the company plans to roll the feature out across all its applications one by one in the years to come.

Clicking on a dedicated button in compatible SAP applications will open up the Joule interface, which will then accept input in a variety of ways. Users can type their request in natural language, click on buttons for suggested actions, or enter dates from a calendar picker. Joule will respond accordingly with text, graphics, and tables drawn from SAP or external data sources, or with links to other applications and workflows.

Access to the Joule interface will be included in existing SAP software license agreements, but the generative AI use cases that Joule calls on will be priced based on their business value, SAP chief marketing and solutions officer Julia White said.

While SAP is opening up its AI assistant to the wider internet, it’s taking care to protect customers’ data. No customer data will be used to train external foundational AI models, said Bharat Sandhu, the company’s SVP for AI and application development platform. The company will host its own large language models (LLMs) in its own data centers where possible, and any data sent out to external LLMs for processing will be masked or anonymized, he said.

Other major enterprise software vendors have made similar generative AI announcements this month, with varying degrees of precision and precipitation in their timelines.

Salesforce went first, saying it will add a conversational AI assistant, Einstein Copilot, to the right-hand rail of all of its apps. It will conduct pilot trials with a limited number of customers before year-end, but has not said when the assistant will become generally available.

Oracle was next, saying it expected to add generative AI-powered capabilities to many of its cloud applications. Some of these are already in trials with a few customers, but users of its Fusion Cloud ERP, HCM, and customer experience tools may have to wait six months for Oracle’s generative AI features to become generally available, while the wait for users of its Clinical Digital Assistant could be up to a year.

ServiceNow, the last to announce, will likely be the first to deliver. It said it will make its generative AI chatbot Now Assist generally available from September 29 with the Vancouver release of its workflow platform. But where SAP and Salesforce plan to make their generative AI tools available to all users, only ServiceNow customers who purchase special add-on packs for Now Assist will get access to the generative AI functionality, limiting the number of users the company has to serve at launch.

White dismissed the threat from the competition, saying, “We’re aware of lots of claims and lots of hype in the broader ecosystem. We’re focused on delivering the customer value.”

Slow but steady

SAP’s slow-but-steady approach is more typical of the company, and perhaps understandable given the diversity of the company’s product portfolio and the complexity of its code base.

Joule is just one of three main layers in SAP’s AI strategy, said Sandhu. The foundation layer, embedded in its Business Technology Platform, handles orchestration, abstraction of third-party models, and such, while the second layer is composed of AI capabilities embedded in applications.

SAP is already using generative AI to help users create job descriptions or interview questions, process delivery notes, or forecast which clients will default on payments.

The new generative AI interface will eventually sit on top of all of SAP’s applications as a common engagement layer that can call on their existing functions, Sandhu said. But it won’t be available everywhere all at once.

“One of the unique aspects of SAP is it has a vast portfolio of applications,” Sandhu said. “We are rolling it out piece by piece. By next year you’ll see it fully integrated into most of our applications. It just takes time to integrate with all these different applications.”

For now, the company is only committing to making Joule generally available before the end of this year on SAP Start, the home page for its cloud applications, and in SuccessFactors, its SaaS HCM tool.

SAP plans to roll Joule out to the public cloud edition of its S/4HANA application suite early next year.

Customers using on-premises versions of SAP’s applications, even of S/4HANA, will have to wait for access to the generative AI features — perhaps forever. In July, SAP CEO Christian Klein told analysts that SAP’s newest innovations will only be delivered in the public cloud, or in private clouds via its Rise with SAP all-in-one offering.

White reiterated that position at the launch of Joule: “We will be delivering our AI capabilities like Joule as part of our cloud value. Generative AI really only works as part of a cloud model,” she said. 

Energy balance

SAP’s choice of name for its AI assistant — the joule is the SI unit of energy — puts the spotlight on one of the downsides of the increasing use of generative AI: the energy it consumes. SAP has been buying green energy for the data centers that power its own operations for years, and has committed to achieve net-zero carbon emissions along its entire value chain by 2030. Moving its customers to the cloud, where it can run applications more efficiently, is one way it hopes to achieve that, it has said.