Artificial intelligence (AI) has increasingly become an integral part of companies and therefore an investment that must pay off.
The new study “The SAP Value of AI Report 2026” from SAP and Oxford Economics shows: German companies are increasingly achieving a positive return on investment (ROI) with AI. At the same time, the majority are not yet fully exploiting the potential of AI and are hardly prepared for the use of AI agents. 2,600 managers from 13 countries worldwide were surveyed for the study, including 200 decision-makers from Germany.
Increasing return expectations through agentic AI
On average, the German companies surveyed in the study plan to invest around 35 million euros in AI this year. This puts them well above the average of around 24 million euros for all countries examined in the study. The expected return is also increasing: German companies expect an ROI of 24 percent for 2026. Last year it was 17 percent. One reason for this could be the expectations placed on AI agents. In the next two years, the ROI from agentic AI in Germany is expected to reach almost 18 million euros. That is more than four times the estimate from the previous year (around 4 million euros). Nine out of ten companies see the technology as having the potential to fundamentally change their organization. However, only a few feel prepared for this development: only four percent say they are fully prepared for the use of AI agents. The majority feel they are only partially prepared or not prepared at all.
“Germany has moved from the AI experimentation phase to implementation within a year and that is increasingly paying off,” says SAP Chief AI Strategy Officer Sean Kask. “At the same time, there is still a lot to be done. Because AI without context, without integration into processes, data and governance, at best creates activity without measurable benefits. At worst, it creates new risks.”
Governance is a weak point when introducing AI agents
One such risk could be the effective control of AI. Governance is becoming a key requirement for trustworthy and scalable AI applications. Nevertheless, only eleven percent of companies rate their capabilities in this area as completely sufficient and only 18 percent believe they have the necessary processes and framework conditions. These deficits become particularly clear when it comes to agentic AI. More than half of companies (57 percent) do not yet have a process that integrates human control into agentic workflows. A third have not set up authorization and access controls for AI agents and only 34 percent have a central directory of deployed agents. Given this, it is not surprising that 75 percent of companies either agree or are unsure about adopting AI agents faster than they can effectively control them.
“The next step to creating real value is to merge data and processes with AI. At the same time, companies need to ensure that AI has the necessary context and appropriate governance structures to deliver trustworthy results. This is exactly what we mean by the concept of the autonomous enterprise,” explains Sean Kask. “However, companies quickly realize that the benefits of AI are often harder to measure than expected and bring risks that are evolving faster than most governance structures can keep up with.”
Selective AI support and data quality slow down potential
An obstacle to combining data, processes and AI is apparently a lack of organizational requirements to anchor AI in the overall strategy. Although 35 percent of all tasks in an average German company are supported by AI, specific tasks dominate. The majority of companies surveyed (39 percent) predominantly pursue a fragmented approach to using AI. In addition, less than half of German companies (48 percent) have a clearly named manager who is responsible for the introduction of AI. Even rarer are concrete AI metrics for managers (35 percent) or training on AI skills and AI risks (47 percent). Accordingly, more than three quarters of companies are convinced that they have not yet fully exploited the potential of AI.
There are also problems with data quality. Although the proportion of companies that rate their data base for AI as sufficient is increasing, 81 percent report at least occasionally problems such as rework, delays or operational backlogs because AI results do not achieve the required quality. In addition, 77 percent doubt that their further training measures can keep pace with the rapid development of new AI applications. However, almost all companies expect AI to have an impact on personnel planning and qualification profiles. Another problem could be the use of so-called shadow AI – applications that are not officially approved or controlled: 71 percent of companies say that such applications are used occasionally to frequently.
The study shows that the economic benefits of AI are becoming increasingly measurable for companies, but at the same time there is a growing realization that successful AI strategies require more than the introduction of new technologies. The next development step is to connect AI more closely with company data, business processes and clear governance structures. Only then can agentic AI systems be reliably scaled and sustainably integrated into operational processes.
