AI with ERP: who needs it?

Resilience, observability and governance

Many AI experts in the SAP community are convinced that the concept of resilience will prevail as a key indicator of digital AI performance. The performance of large language models (LLMs) such as ChatGPT from OpenAI or Gemini from Google is based on highly sophisticated statistical methods and machine learning. Deterministic algorithms, such as those used in SAP's ERP software for over 50 years, are not intended for LLMs. This means that there is always a risk of hallucinating when using traditional AI.

AI systems can quickly escalate across applications, cloud regions, payment systems and external services due to minor disruptions. ERP users, however, demand reliability, availability and security - in other words, resilience and observability. These requirements can no longer be viewed as separate disciplines, but must be integrated as a common AI goal: SAP has not yet provided proof that an S/4 BTP system has the ability to absorb disruptions, react quickly and ensure a consistent user experience under load.

The AI community problem

A study commissioned by Dynatrace with FreedomPay shows how fragile digital ecosystems have become and how quickly technical failures lead to customer frustration and financial losses. In the UK, an estimated £1.6 billion in revenue is at risk from payment failures each year, and in France around €1.9 billion. A single disruption can spread across networked ERP and AI systems, highlighting how tightly coupled modern end-to-end processes are.

Customers feel these failures immediately. Patience wears thin within a few minutes, and many cancel ERP transactions if the problem persists for more than fifteen minutes. However, the average duration of an outage is more than an hour - by which time the damage has long been done. Almost a third of all customers lose trust after just one incident, with younger, digitally-savvy target groups reacting particularly sensitively.

This situation requires a common approach to resilience, says Dynatrace. Companies need a common understanding of how ERP services behave, how errors propagate and how recovery impacts the customer journey. Resilience is measured by how AI and ERP systems respond under pressure, not just how they perform in normal operation.

2026: CES (Las Vegas, USA) and Viva Technology (Paris, France)

Events, congresses and trade fairs are an important part of the AI discourse. In the second week of the new year, the leading managers of the IT industry and their customers met in Las Vegas at CES 2026 - The Most Powerful Tech Event in the World, Consumer Electronics Show. Last year, the European AI hotspot Viva Technology was in Paris in June. Naturally, Nvidia was very well represented there. However, ex-SAP CEO and current Siemens Supervisory Board Chairman Jim Hagemann Snabe was also on the keynote stage. SAP was absent from Paris 2025, although almost all of SAP's major consulting partners were represented at Viva Technology.

Germany is the preferred host country for Viva Technology 2026. Whether SAP will find its way to Paris this year is uncertain. However, if SAP CEO Christian Klein is really serious about AI, he should follow his predecessor Jim Hagemann Snabe to Paris and give a strong sign of life on the Viva keynote stage. True to the motto: Do good and talk about it!

AI challenge: Agentic AI and observability

No IT company - whether legacy or start-up - will probably be able to overcome the AI challenges alone. SAP will also need the public AI discourse and IT partners. The solo effort with partial partnerships that SAP has undertaken so far has not led to any overwhelming AI successes. The subject matter is difficult to master!

Agentic AI is an example of the power of AI and at the same time much more difficult to master than traditional LLM and machine learning systems. When AI agents on the SAP BTP (Business Technology Platform) coordinate tasks, exchange context and trigger further ERP actions, even a well-structured digital environment can tip over into unpredictable behavior. However, the SAP community can experience what is already possible on the SAP BTP and with SAP Business AI in two AI experience workshops in the April in Heidelberg and June in Salzburg experience. Together with SAP partner Snap, E3 Magazine is organizing these AI workshops with a strong practical focus: BTP, Clean Core and SAP Business AI.

The IT company Dynatrace has also recognized that most companies are not (yet) prepared for this AI change. Without strong observability and clear governance, AI/ERP systems will become increasingly difficult to understand and control. This realization applies in particular to the topic of agent-based AI. Each AI agent acts independently based on instructions and inputs - not only from humans, but also from numerous first and third-party agents. A single customer contact can trigger hundreds of background processes in which AI agents independently make decisions, switch roles and instruct other agents.

But every agent should be accountable to a human or a higher-level agent, and supervision always remains with the human - at least that is the plan: the explosive growth in agent-based communication can therefore no longer be controlled without observability. The challenge no longer lies in optimizing individual models. The decisive factor is to control the network of autonomous interactions in real time. Observability thus becomes the basis for secure, scalable and controllable agent ecosystems, writes Dynatrace in a recent statement.