According to the study, only 35 percent of German companies have KPIs at board level that are directly linked to the introduction of AI. In your opinion, which metrics should supervisory bodies and CEOs definitely measure?
Birch: A central indicator for us is employee enablement. How many employees have already successfully completed training or upskilling programs related to AI? Without the appropriate skills, AI adoption will also fall short of its potential.
Transparency is equally important. Companies should know which AI agents are actually in use in their landscape. For this purpose, SAP offers the SAP AI Agent Hub, which automatically discovers and inventories agents from SAP and third-party environments. Customers have already identified thousands of agents, highlighting the need for centralized governance and transparency.
Companies should also have a complete overview of all AI use cases. There should be a reliable business case for every use case. We often experience two extremes: Either the board is under pressure to introduce AI as quickly as possible and allocates a flat budget for this. Or the management initially holds back. Then independent pilot projects are created throughout the company, individual departments procure their own tools and conclude their own contracts.
At SAP, we therefore follow a clearly structured selection process. Every idea first goes through an assessment of the expected business value. We then examine the technical feasibility, data availability as well as ethical and governance aspects. From a management perspective, it is crucial to have transparency about all ongoing AI projects at all times and to consistently prioritize them according to their business value.
“Agents also need a ‘hire-to-retire’ life cycle”
Even with the introduction of dozens or even hundreds of AI agents, governance becomes increasingly complex. What features do enterprise platforms need to manage AI agents at scale in a safe and controlled manner?
Birch: We consciously try not to humanize AI too much. Still, the analogy helps: Agents need a complete “hire-to-retire” life cycle. This starts with recognizing and registering an agent. He is then integrated into the company environment and receives the necessary authorizations and access to the data sources that he needs for his tasks.
Observability is equally important. Companies must be able to understand at all times what an agent is actually doing in the system. Additionally, they should capture performance metrics: Is the agent achieving desired results? How efficiently does it work? How many tokens does he use? How many processing steps does it need for a task?
Ultimately, it involves several building blocks: a complete inventory of all agents, appropriate governance, risk and compliance (GRC) mechanisms, transparency about agent behavior and continuous monitoring. This is the only way to ensure that AI agents permanently work within the defined guidelines and deliver the desired business value.
Which business processes do you think companies will actually fully delegate to AI agents in the next two to three years?
Birch: Such agents currently work particularly well in clearly defined use cases. SAP will release more than 50 (currently 34) specialized AI agents.
One example is periodic accounting. Bookings there must be made based on numerous rules that are stored in documents, emails or previous transactions. The agent analyzes these different sources of information, derives a recommendation from them and suggests the appropriate booking to the clerk.
According to what we hear from customer projects, employees of medium-sized companies now often spend around twelve hours per month on these tasks. With the help of an AI agent, this effort can be reduced to two to three hours.
Another field of application is production planning. If delivery dates change or new orders are received at short notice, the entire production plan must be adjusted. It is precisely such complex optimization tasks that are ideal for AI agents.
In principle, there are hardly any limits to the narrowly defined business processes in which agents can be used. However, they will not initially operate completely autonomously.
“Trust in AI starts with a stable foundation”
Many companies still find it difficult to trust AI agents. Finally, large language models work probabilistically and can hallucinate. This is particularly problematic in financial processes. How do you create trust?
Birch: Trust begins with a stable foundation. The ERP systems remain the reliable “system of record”. They work deterministically, contain the business logic and the relevant company data. AI agents build on this foundation – they do not replace it.
Equally important is the “Human in the Loop“-Prinzip. Employees must be able to understand what the agent is doing, check his results and intervene if necessary. That is why the qualifications of employees also play a crucial role. You need to understand how generative AI works and what its limitations are.
Of course, language models can hallucinate. At the same time, one must not forget that people do not work without errors either. The interaction between humans and AI is crucial. In this way, both the efficiency and the quality of many business processes can be improved.
Another important building block is transparency. For example, our global AI ethics policy requires users to always be able to recognize when AI is involved. In Joule you can see which data sources the agent used and which steps he went through when making his decision. This traceability is an essential prerequisite for trust.
What distinguishes an SAP agent from a general AI agent that only accesses an ERP system?
Birch: The crucial difference is that Joule and the SAP agents are anchored directly in the ERP system. There, for example, we built a “Knowledge Graph” that describes the semantic relationships between all tables, business objects and data fields.
To illustrate the scale: The SAP S/4HANA Knowledge Graph is based on around 452,000 ABAP tables, 7.3 million data fields and thousands of analytical views. The semantic relationships between these artifacts are modeled in the knowledge graph and made usable for AI applications.
For example, if a user wants to view all open orders, the agent does not have to laboriously search for the relevant information. He immediately knows which tables and objects are relevant and also knows the relationships between a purchase order, a purchase requisition, the responsible approvers and other business objects. This means that the agent not only works much more precisely, but also requires significantly fewer tokens because it can narrow down the search space significantly.
If you instead just try to add AI to an existing system or extract data from a relational ERP system, many of these relationships are lost. In a sense, you destroy the semantic context that is crucial for precise answers.
That’s why we see the ERP system as a huge strategic advantage. It has been the system of record for decades and contains around 50 years of coded business and process knowledge. This knowledge forms the basis for what we call “Autonomous Enterprise“ designate. The agents build on this knowledge and develop it further.
In the future, SAP agents will also communicate bidirectionally with agents from other providers using standards such as agent-to-agent (A2A).
According to your study, AI currently creates the greatest added value in decision-making, customer interaction and the acquisition of new insights – less so in classic productivity improvements. Will this change the way companies justify AI investments in the future?
Birch: In our study, productivity was only rated slightly lower than, for example, the acquisition of new knowledge. In the long term, however, productivity remains the real goal. Europe in particular has been suffering from comparatively weak productivity growth for years.
At SAP, we therefore first evaluate every new AI function based on its specific business benefit. We first create a value analysis for all agents and AI functions that we record in our “AI Feature Catalog”. We ask: What benefit does the function have for the user? Does it contribute to more sales? Does it increase productivity? Only then will it be developed further.
At the moment, the greatest added value often lies in bringing together information from structured and unstructured data sources and making it accessible using natural language. The next step, however, is to translate these findings directly into more efficient business processes. This is exactly where the greatest productivity gains will be realized in the future.
“The biggest mistake would be to transform the entire company straight away”
If you could give CIOs just one or two pieces of advice for transitioning from generative AI to AI agents, what would it be?
Birch: In my opinion, the biggest mistake would be to try to transform the entire company straight away or to first prepare all the data perfectly.
Instead, you should consider what kind of Agent can create high added value – and then implement it. Of course, this agent needs access to consistent and context-rich corporate data. This is exactly what we are working on at SAP with technologies such as the “Knowledge Graph”, which maps the semantic relationships of company data.
In addition, with “Data Products” and the “SAP Business Data Cloud” we provide tools that make data from different sources usable for AI agents. Thanks to zero-copy and data fabric approaches, information from legacy systems, Snowflake or ERP systems can be brought together without the data having to be extensively replicated first. In this way, a purchasing agent can be provided with exactly the relevant data for the respective use case.
The key point is: companies don’t have to wait until they have fully migrated to the cloud or consolidated their entire data landscape. With the technologies available today, data can already be made usable for specific AI agents, managed in a controlled manner and, on this basis, initial business added value can be quickly achieved. On the other hand, if you wait for the perfect starting position, you run the risk of losing touch.
