Q&A: How SAP Is Connecting Embodied AI and Robotics to Enterprise Business Processes

Following Hannover Messe, where SAP showcased its growing capabilities in embodied AI across warehouse, inspection, and industrial scenarios, SAPinsider spoke with Adelya Fatykhova, SAP’s architecture lead for embodied AI and robotics, to understand what comes next.

In this Q&A, Fatykhova explains how SAP is thinking about partner-led delivery, BTP-based architecture, production readiness, and the business-process challenges that make physical AI worth pursuing. For SAP leaders evaluating this space, the conversation offers a practical view into SAP’s embodied AI vision and the operational realities behind deployment.

SAPinsider: As the architecture lead for embodied AI, physical AI, and robotics at SAP, what does your role involve, and how does it shape SAP’s approach to bringing this into enterprise operations?

Adelya Fatykhova: From a technical perspective, my role is to ensure that embodied AI and robotics fit into SAP’s overall architecture strategy and that this is not just conceptual, but something that works end to end in real enterprise environments.

SAP is not building robots itself. Instead, it is working through a partner ecosystem of robot vendors, system integrators, and AI model providers to connect physical AI to business processes. This includes three main categories: robot vendors who provide the hardware, system integrators, such as accenture, who help deploy and connect these solutions into enterprise environments, and AI model providers, such as NVIDIA, who supply the physical AI models used by these systems. Together, these partners enable SAP to deliver end-to-end use cases that integrate physical execution into business processes.

We are focused on business processes end to end. Business processes have physical execution steps, and today there is limited visibility into those steps. With physical AI, this creates the possibility to automate business processes end to end.

Because this operates as a closed loop, with execution data flowing back into SAP, it also enables optimization. Customers can gain a full picture of how a business process is actually performed, including the physical execution, and use that insight to improve outcomes.

Q: Physical AI is emerging as the next step as AI moves from generative AI to operational AI and into execution. Where are you seeing early adoption?

AF: We are vendor-agnostic and robot-agnostic. Customers choose the type of robot based on their use case, whether that is humanoids, robot dogs, drones, or other forms.

What we are seeing is that use cases are closely tied to the type of robot. For example, in warehouse environments, pick-and-place operations are a strong early use case. These are well suited for wheeled humanoid robots and connect directly to SAP Extended Warehouse Management (SAP EWM).

Another area is inspection tasks. These are common in asset management and maintenance scenarios, using solutions such as SAP Asset Performance Management and SAP Field Service Management. In these cases, robot dogs, such as those from ANYbotics, are often used, especially in hazardous or hard-to-reach environments.

We also see applications in health and safety inspections, where automation can reduce risk in dangerous conditions.

Q: What types of decisions or actions are robots actually executing today?

AF: The focus right now is on tasks that are either difficult to automate or where the environment is unpredictable. These include pick-and-place operations, inspection tasks, and safety checks.

The key characteristic is that these tasks are either repetitive but hard to execute manually at scale, or they involve environments that are dynamic and changing. That is where physical AI brings value compared to traditional automation.

Q: What are the key challenges SAP encounters in delivering physical AI?

AF: Customers do not adopt technology for its own sake. They start with a problem to solve. Many of the challenges relate to hazardous environments, limited access to skilled labor, or operational constraints in areas like field service and asset inspection.

There is also a broader workforce trend behind this. Some industries are dealing with aging workforces and growing difficulty replacing specialized skills, so customers are exploring where physical AI and robotics can help automate tasks that are becoming harder to staff reliably.

On the technology side, key challenges include hardware readiness and certification, as well as integration. Customers expect robots to be fully integrated into their enterprise systems, not operate as standalone tools. Identifying the right use cases is also critical, since physical AI is most valuable in scenarios where tasks or environments are unpredictable.

Q: Can you outline the architecture behind a physical AI deployment in SAP?

AF: One of the most important aspects of our approach is that physical AI is treated as a central capability rather than something embedded into individual applications.

We build this on SAP Business Technology Platform. This allows multiple applications to use physical AI capabilities instead of having isolated integrations.

We also approach this from an agentic perspective. Embodied AI agents are essentially business AI agents that can interact with the physical world through robots. These agents handle business context, decision-making, and task execution from a process perspective.

That is why the architecture is BTP-first. The central services sit on SAP Business Technology Platform, and different applications can extend their capabilities through that shared foundation. SAP also sees this through the lens of Business AI and business AI agents. In that model, an embodied AI agent is essentially a business AI agent that has access to a physical tool such as a robot. Today, traditional robotics integrations tend to be point solutions, for example in SAP Extended Warehouse Management for AMRs or in digital manufacturing for shop-floor devices, and are not easily reusable across the portfolio.

With embodied AI, SAP is aiming to enable physical task execution across multiple applications through a shared, BTP-based service. This is paired with an agentic approach, where business AI agents can use robots as a tool to interact with the physical world. The agents handle business context, exception handling, and decision-making, while partner technologies handle motion control, perception, and low-level robotics functions.

Integration is designed to be flexible, including agent-to-agent (A2A) patterns, so that systems can react to dynamic, real-world situations while staying aligned with business processes. In this model, robots are not standalone tools but become part of end-to-end, business-driven execution within SAP.

Q: Where is operational data collected in a physical AI setup, and how is it different from traditional SAP data?

AF: SAP already has a strong foundation for business data through solutions like Business Data Cloud. This is where structured business data is typically stored and analyzed.

With physical AI, we introduce a new category of operational data coming from the physical world. This can come from robots, sensors, cameras, or other devices.

In many cases, this data is handled through partner systems or specific SAP solutions such as digital manufacturing. The goal is to bring both business data and physical-world data together so that customers can optimize processes based on a complete view of operations.

 Q: You mentioned embodied AI agents handling unpredictable environments. How is this different from traditional automation?

AF: Traditional automation works well in structured and predictable environments. Physical AI is designed for situations where either the task or the environment is dynamic and constantly changing.

For example, in a warehouse, layouts may shift, items may be misplaced, or conditions may change in real time. In these cases, systems need to adapt rather than follow predefined rules.

That is where embodied AI comes in. It enables systems to react to real-world conditions while still operating within business processes.

Q: What are the biggest barriers to scaling physical AI deployments?

AF: Good question. I think to state that the other way around, one of the reasons companies are interested in physical AI, especially humanoid robots, is the assumption that if an environment is designed for humans, it should also work for humanoid robots without much adaptation.

So, from a physical and process readiness perspective, a lot of the hope is that customers don’t need to change so much. If it’s a space designed for humans, humanoid robots can also work there, so the gap is smaller. That’s something customers find very attractive.

But there are still real barriers. One is certification and regulation. For production environments, the hardware needs to meet requirements such as CE marking. This is changing now. Two years ago, and even last year, almost no one was certifying humanoid robots because the hardware was still evolving. This year, vendors are starting certification processes, which is a good sign. Without that, customers cannot scale.

Another barrier is integration. Customers don’t want robots as standalone resources. They expect them to be integrated into their systems, both SAP and non-SAP. That expectation for integration is actually one of the biggest drivers for why SAP is involved in this space.

So, if we talk about scaling, it comes down to the readiness of the technology, especially certifications, the readiness of the customer environment, which may require less change than expected, and the ability to integrate robots into enterprise systems at scale.

Q: What needs to be in place before a customer can move from pilot to scaled adoption?

AF: One common assumption behind humanoid robotics is that if an environment is already designed for people, it should require relatively little adaptation for humanoid robots. That is one reason customers find this area attractive: in some cases, the barrier to deployment may be lower than they expect from a physical-environment standpoint.

That said, scaling still depends on certifications, regulations, and technology maturity. If the hardware is not certified, customers will not scale it into production. Beyond that, customers need to think carefully about use-case selection. Physical AI is not a replacement for every kind of automation. It is most relevant when the task itself is variable, dynamic, or difficult to automate in traditional ways, or when the environment is unpredictable and changing.

If the work is already highly standardized and predictable, traditional automation or simpler technologies may be the better fit. The priority for customers is to identify where embodied AI is genuinely necessary rather than using it where other tools would do the job more efficiently.

Q: What should organizations prioritize if they want to adopt physical AI?

AF: The most important step is identifying the right use cases.

Organizations should focus on scenarios where tasks are dynamic, environments are unpredictable, or traditional automation is not sufficient. If those conditions are not present, simpler solutions may be more effective.

Physical AI should be applied where it provides clear value, not as a replacement for existing technologies.

This is something that is exactly the same when it comes to this technology. Organizations need to evaluate if this is really the right fit for their use case. If we look at the common thread across the use cases, it is that they are difficult to automate today because the task itself is not standard, it changes frequently, or it is unpredictable.

In some cases, the task may be routine, such as inspection, but the environment is unpredictable, for example outdoors. So from that perspective, customers need to prioritize by looking at the challenges they have and clearly identifying which ones actually require this technology versus those that are better addressed in other ways.

The scenarios where physical AI makes sense are those where either the task or the environment is dynamic, unpredictable, and changing. If that is not the case, then simpler solutions, such as traditional robotics or even cameras, may be sufficient.

What This Means for SAPinsiders