SAP joins the AI agent era — but not all customers may benefit

Following the lead of its enterprise software rivals, SAP too has announced it will now be entering the agent era of AI.

Agentic AI is being billed as the next wave of AI in the enterprise, with a focus on operational decision-making over content generation. As opposed to generative AI chatbots, AI agents aim to solve more complex, decision-based tasks, which will  be key in a business context, as many companies are looking for concrete AI use cases that, above all, bring added value to business operations, productivity, and the bottom line.

Last month saw SAP rivals ServiceNow and Salesforce announcing agentic AI for their respective offerings. For ServiceNow, agentic AI is being fused into its Xanadu release of the Now platform. Salesforce, meanwhile, unveiled Agentforce, a suite for creating AI agents, just ahead of its Dreamforce annual conference.

SAP, which claims to cover business processes end-to-end with its ERP offering, could integrate its agents deeply into the business logic of its enterprise customers’ operations and support them with relevant data — a plus for the Walldorf, Germany-based company. In concrete terms, the success of its agentic AI strategy will depend on how quickly and how many use cases SAP gets up and running. The AI ​​strategy does not focus on the technology, but on the applications, said Philipp Herzig, SAP’s chief artificial intelligence officer.

“AI is an application feature,” he said at TechEd, and he may be right. 

More productivity through AI agents 

Specifically, SAP is expanding its generative AI copilot Joule, introduced last year, to establish “cooperative multi-agent systems,” according to the company. These agents will open up a new era of productivity for SAP customers, said Herzig, highlighting the fact that SAP’s agents will be deeply connected to the company’s systems. This makes it possible to handle more complex tasks in processes and workflows than with conventional stand-alone agents, which are usually trained only for a specific task, Herzig said.

Instead, Joule agents could work together collaboratively to answer more complex user queries. In these multi-agent systems, AI agents with specialized expertise would take on certain tasks in complex business processes, SAP said. Working together with other specialized AI agents, the cooperative system could dynamically adapt their strategies to solve the tasks assigned to them, the company claimed.  

SAP plans to release the first agents in the fourth quarter of 2024 and presented two application scenarios at TechEd: helping employees in finance departments with invoice workflow by pointing out errors in invoices; and automating financial processes such as the processing of payments or postings in the general ledger, thus making them more efficient. 

Joule is to be integrated ever more deeply into SAP’s software landscape. By the end of the year, the AI ​​tool will support around 80% of the most common business tasks within SAP, the company said. 

The view from here

The number of specific examples SAP cited for agentic AI isn’t overly impressive. The use case of finding errors in invoices or processing bookings automatically does not sound particularly spectacular on the face of it. In addition, the competence of its agents to solve highly complex tasks, which SAP emphasized, is not yet evident. A few weeks ago, Salesforce provided a much more impressive demonstration at Dreamforce, in which its AI agent fielded a service call about the return of goods with various options in terms of exchange, location, and timing. 

SAP is at least promising more speed. The foundations were laid in 2023, and now development is accelerating, Herzig announced. With each use case, development becomes easier and faster, he said. In the end, the only question that remains is who can actually use SAP’s new AI features. That remains a downside for large portions of SAP’s clientele: The company is sticking to its strategy of reserving the new technology solely for cloud customers with corresponding Rise and Grow contracts. The many on-premises customers are left out when it comes to Joule and multi-agent systems. As are, presumably, SAP cloud customers not on Rise or Grow.

Industry analysts have also questioned the cost of using Joule AI agents — something that was not revealed at TechEd.

At least SAP is offering its customers the prospect that the path to the cloud could be simpler and easier with AI assistance. For example, the new AI agents should be able to identify custom code and make suggestions as to how user companies should deal with it.

It remains to be seen whether this argument will work with customers. Perhaps we will find out more about this at the annual conference of the German-speaking SAP user group (DSAG) in Leipzig in mid-October.