Like most ERP vendors, SAP promotes its newest cloud versions as the best route to AI-driven digital transformation. Uniper, an energy provider based in Düsseldorf, Germany, has bought into that vision.
Early this year, it signed a contract to use the Rise with SAP migration service to replace legacy on-premises SAP ERP systems with S/4HANA Cloud Private edition in 2027. The goal is to modernize the company's ERP platform and take advantage of the latest AI technology, including embedding SAP Joule and Microsoft copilots across business processes.
AI requires cloud scale and processing power, but that doesn't mean organizations must first move to cloud ERP to use AI. Some organizations anxious to get started with AI opt for a hybrid cloud that retains on-premises ERP for certain business functions and uses cloud ERP for others, augmenting both with AI apps that run in the cloud.
Uniper is employing this hybrid approach as it prepares to move to S/4HANA Cloud. For example, it used SAP Joule Studio to develop a custom Joule agent to help employees find purchasing information and execute procurement tasks by communicating in natural language, according to Achim Löbke, Uniper's head of SAP transformation. The app runs in the cloud but draws data from on-premises ERP systems.
The company is in the middle of the journey to S/4HANA, Löbke said.
"Our technology backbone is still more than 20 years old, and that means until we go to the S/4 system, which will happen with a pilot next year, we will see how the story plays out in the future," he said. "My idea is to lead people step by step into the new system landscape and have the full potential of Joule on the new S/4 system."

Charging ahead with digital transformation
Uniper, which has around 7,000 employees, supplies power to several European countries but trades energy sourced from across the globe. Its comprehensive IT transformation is meant, in part, to support a strategy announced in 2023 to transition to green energy sources.
This isn't Uniper's first S/4HANA migration. It used Rise to replace a 30-year-old on-premises payroll system with the payroll module in SAP HCM for SAP S/4HANA, a version of the vendor’s previous on-premises HCM system adapted to run with S/4HANA. Uniper also replaced SAP Industry Solution for Utilities with the on-premises S/4HANA version, called S/4HANA Utilities, and will move that to the S/4HANA private cloud next year.
Microsoft technology is also central to Uniper's modernization strategy. Some enterprise applications and data centers already run on the Microsoft Azure public cloud, which is also the destination for future migrations. Employees use the Microsoft 365 office suite, including the Copilot AI assistant and its low-code development tools. SharePoint stores the important documents employees and AI must access to ensure compliance.The companies have also partnered to advance AI adoption in the energy industry.
The new SAP Joule procurement app follows an effort that began three years ago to use Business Technology Platform (BTP), SAP's software development and integration suite, to consolidate purchase requisitions from two SAP ERP systems and Ariba procurement software. It also aimed to make the information accessible from a "cockpit" built on Fiori, SAP's GUI platform. Procurement specialists have been using the app for the formerly time-consuming manual process for more than a year, Löbke said.
To build it, Uniper had to develop a custom search function using the generative AI hub in BTP.
"Sometimes the search function in Ariba itself is not the best, and it gives you many, many proposals, but not what you are looking for," he said.
A machine learning (ML) app that pulled purchase requisitions into purchase orders followed. Now, the focus is on making procurement information accessible to employees across different departments, including power plants, administration and trading, from Joule or Microsoft Teams by the end of this year.
"When they started, they had a machine learning model," said Vidit Kumar, an AI product marketing manager at SAP. "The problem they were trying to solve with that model was inconsistency in the data that is imported as part of users raising purchase-order requests. Their team had to do a lot of manual work, so they used this ML-based model that is built with the S/4HANA Cloud vector engine. It's basically like a vector search."
A key part of the ongoing work is to replace the ML functions with SAP RPT-1, a pretrained AI model designed for natural-language querying of the tabular, numerical data typical of ERP systems. The familiar large-language models of GenAI are built for unstructured text and therefore less reliable with such structured data.
Löbke said some users will access the RPT-1 tool from Microsoft Copilot while others will use the Fiori Launchpad.
"Ultimately, they want their users to access one interface that they can come to," Kumar said. "To users, all of the back-end details will be ambiguous or not very relevant. When a user submits a query to Copilot, it then integrates with Joule, which is our copilot for accessing the SAP ecosystem."

RPT-1 will continue to play an important role in Uniper's plans. Löbke said that while the 2026 version of the procurement app focuses on answering users, the vision for next year's version is for agentic AI to complete the transactions.
Business-IT collaboration
Several business and IT entities are involved in the Uniper projects and have developed a good working knowledge of how to integrate processes and AI, Löbke said.
A procurement team has expertise in business processes and technologies, including AI. For its part, IT provides two people for front- and back-end development who work with the AI experts in procurement. IT service providers Adesso, DXC Technology and SAP Services and Support also help.
They follow a scaled Agile development framework. Planning is done once a quarter. Developers meet daily, and every two weeks the business side checks on their progress.
"This works quite well, because in the scaled Agile framework, there are milestones," Löbke said. "The business always has a two-week touchpoint to see what's coming out of the two weeks. For three or four years now, it has brought us big value from developing in a structured way."
Meanwhile, business users employ the low-code tools in Microsoft Copilot to program agents for their departments. Both beginners and advanced users are doing it successfully. One example is an agent for the consolidation process for Uniper's quarterly financial reports. Users in different departments, such as accounting and legal, can use the app to request changes and trust that perfect accuracy will be maintained, with daily changes reflected in the reports, Löbke said.
"We support it a little bit from the platform perspective, but the business built it on their own," he said. "It's just a small use case, but it has a big impact because 300 end users are affected."
Löbke said when the company moves to S/4HANA next year, it will get more use out of low-code tools because a standard, greenfield implementation offers more opportunities to connect dedicated APIs that make it even easier for business users to build small, low-code apps.
Connecting on-premises ERP to AI in the cloud
Löbke explained how it's possible to add AI to SAP ERP Central Component (ECC), the on-premises predecessor to S/4HANA, the first versions of which also ran on premises.
"It's just that you have to develop it on your own," he said.
BTP serves as the anchor, connecting to the ECC system using SAP's OData service, a standardized web service for connecting to back-end data and logic.
Uniper's invoice-checking GenAI app is a good example.
"Normally, before we pay out our invoices, colleagues in accounts receivable have to check each paper or PDF invoice against the value in SAP. Now, this can be done automatically via AI," Löbke said.
The AI also reads notes that employees attach to invoices — such as requests to delay payment. A human in the loop then double-checks a proposal list of payments before approving them.
Kumar said the issue with running AI with an on-premises ERP is ensuring the AI has the right data to be accurate.
"Depending on where that data resides, it can be either on the cloud, S/4HANA Cloud or your ECC, which is on-prem," he said. "Your on-prem systems are also exposed via APIs that these agents can connect to. Ultimately, these agents are executing on the cloud."
Sid Misra, chief marketing officer for the SAP Business AI platform, said SAP is working to expand the connectivity tools. Last month, it augmented the platform by announcing the Enterprise AI gateway, which helps manage AI traffic by centralizing connectivity standards, such as Model Context Protocol, agent-to-agent protocol and APIs, along with authentication, policy enforcement, routing, monitoring and auditing.

"That component plays a big role," Misra said. "For these agents to really mobilize and work for the customer, they need to connect to your on-prem systems as well as cloud systems."
Despite the success of Uniper's hybrid architecture, Löbke described the 20-year-old, on-premises ECC system as a drag on progress, and it will take next year's move to S/4HANA Cloud to show Joule's full potential.
It helps that the procurement app was written on BTP, so it won't need to be rewritten to run on S/4HANA.
"That was already in mind when we started it, and we can simply connect it to the S/4HANA system," Löbke said. "But, especially for the other areas like finance and plant maintenance, my hope is that we get a lot of standard functionality [from S/4HANA]."
AI and S/4HANA bring different challenges
Every company that moves on-premises ERP to the cloud struggles to get users to drop their old ways and adopt new processes, and Uniper is no different.
"Even with private cloud, for us to be on SAP standard means a lot of effort to convince our business colleagues to follow, because in some dedicated cases, they will lose some functionality," Löbke said. "In other cases, they will gain some. But it's hard, debating every day."
Perhaps by 2030 or 2031, he said, Uniper can move to the public cloud version of S/4HANA, which, unlike the private cloud, allows little to no customization. One holdup: the lack of a public cloud version of the maintenance software used by Uniper power stations.
"Maybe for finance and procurement it would work, but for plant maintenance, it's not ready yet," he said.
The ethical, security and governance issues of AI are important for a company like Uniper that deals in critical infrastructure, according to Löbke. But the required approvals can take a long time because every AI use case must go through a governance board, where lawyers ask lots of questions.
"They are coming from their legal perspective, and we are talking two completely different languages," he said. "Sometimes, it's not easy to get to the same understanding" about levels of risk, for example.
Another challenge is addressing worker fears that AI will take their jobs. Löbke said Uniper plans to hold AI experience days to encourage a mindset shift from fear to fun. They can explore what AI can do and recognize that while AI might disrupt current work, it can free people for new challenges.
"We also try to bring this message to our top leadership and board, so they all speak the same language," he said.
Löbke is looking forward to the S/4HANA Cloud pilot, which will give Uniper access to more than 400 embedded AI features, he said.
Unlike S/4HANA Cloud Public Edition, which is a shared, highly standardized multitenant SaaS version of S/4HANA, Private Edition is a single-tenant version of the software that is customizable. Nevertheless, it is standardized enough that Löbke expects Uniper to save significantly on maintenance costs — money to put toward the AI-driven digital transformation after accounting for its added costs.
"For three years, we developed our own tool, and it's very costly," he said. "Procurement is nice, but in the end, it's a supporting function for our core business."
Löbke advised others considering similar transformations to avoid the common pitfall of siloed thinking and align activities across the organization. Besides securing support from executive leaders and boards, he said promoting the project at events can help motivate the project team.
David Essex is a senior editor who creates in-depth content on enterprise applications, emerging technology and market trends for TechTarget's CIO group.