According to the IDC 2024 Supply Chain Survey, 63% of respondents “have an AI strategy linked to business objectives” to improve operational efficiency, business resilience, and increase employee productivity.

Business Challenges in Integrating AI

Business leaders see significant challenges in integrating AI into business applications and processes, especially when trying to make sense of vast amounts of unstructured data. These challenges specifically stem from:

• Data quality: Data is inconsistent or not reliable nor current.

• Organizational Readiness: 

Organizations continue to rely on on-premise systems that pose challenges around stale data, integration, and scalability.

• Volatility: Global AI regulations are constantly evolving, for example the EU AI Act. Technologies such as large language models (LLMs) are also rapidly changing in performance and price with new models and providers popping up almost weekly.

These challenges underscore the need for a thoughtful approach to integrating AI into supply chain processes. SAP Business AI delivers a comprehensive set of relevant, reliable, and responsible AI solutions that effectively tackle many of these hurdles and ensure seamless AI adoption throughout your supply chain processes.

Our Vision of AI-Enabled Supply Chains

For decades, SAP has led the way in supply chains, empowering its customers to streamline their complex end-to-end supply chain processes for enhanced efficiency, agility, and resilience. The SAP Supply Chain portfolio introduces innovation through various technologies, including AI, to transition your supply chain from digital to adaptive, with the ultimate vision of instituting an autonomous supply chain.

There are three possible maturity levels where you could stand in your business transformation. It is also important to acknowledge that different lines of business are on different maturity levels in an organization – while supply chain planning may already be somewhat adaptive, manufacturing operations may still be at the start of a digital transformation.

• Digital: Digitalization is the starting point. The SAP Supply Chain portfolio automates manual processes, enabling a seamless, digital end-to-end business process and digitalizing paper-based systems. This results in better access to data, greater visibility, and control over the entire supply chain, setting the foundation. Most importantly, SAP provides the tools and systems to get data ready for more advanced stages.

• Adaptive: This step is marked by an integration of cutting-edge technologies into business processes, including the use of predictive analytics and simulation for decision-making and Big Data, as well as the introduction of Joule, SAP’s copilot for every supply chain cloud application. These technologies empower supply chain professionals with intelligent insights and recommendations, enhancing decision-making for greater agility and resilience.

• Autonomous: Next, we envision a path characterized by technological, procedural, and data enhancements that will propel the supply chain into an autonomous era. This transition will happen gradually but ultimately enable supply chains to operate autonomously with minimal human intervention, resulting in even greater efficiency, adaptability, and responsiveness. It empowers supply chain experts to focus on value generating activities and the most critical disruptions and business opportunities.

Technologies that Lead Toward Autonomous Supply Chains

SAP has been developing innovative AI capabilities that are fully integrated across digital supply chain applications, catering to diverse customer technology readiness levels. For example:

• Optimization models, rule-based predictive analytics, and heuristics improve, for instance, transportation plans, production scheduling, supply plans and spare part fulfillment. This allows a powerful way to automate decision-making in balancing conflicts of interests, such as customer service levels versus supply chain cost.

• SAP-owned machine learning is applied across the SAP Supply Chain portfolio; for example, in gradient boosting algorithms for demand forecasting, intelligent lead time predictions and failure curve analysis. SAP also offers customers a “bring-your-own-model” approach to enable extensibility for specific domains such as visual inspection and anomaly detection.

• Generative AI, when embedded in digital supply chain applications and combined with Joule, will allow users to conduct complex business transactions in simple conversational ways. We will augment tasks like conducting what-if scenarios for supply chain planning, assisting in new product ideation, analyzing manufacturing issues to accelerate the onboarding of new equipment, and assessing advanced failure modes of assets with generative AI.

With these technologies, our goal is to support customers wherever they stand in their journey and help them move from a focus on mere digital transformation to highly adaptive processes — and ultimately toward autonomous supply chain systems.

Examples of AI Innovations

Let’s explore specific innovations delivered in the SAP Supply Chain portfolio: These cases are classified according to their primary focus on digital, adaptive, and autonomous as a continuous journey. 

Digital

Intelligent Lead-Time Prediction for More Accurate Integrated Business Planning

This machine learning extension enables the extraction of historical lead times from goods movement data in SAP S/4HANA and analyses outliers and key influential factors for lead time changes to predict future lead times. The results are then uploaded as input for planning runs within SAP Integrated Business Planning, thus facilitating better decisions, and enhances plan adherence by considering the dynamic nature of actual lead times and future trends.

• Learn more here.

Logistics

SAP S/4HANA Transportation Management: Intelligent Goods Receipt Analysis

In high-volume logistics industries, the manual tasks involved in inbound goods receipt can be time-consuming and error-prone. By leveraging AI-based document extraction, this use case automates data extraction from freight documents, seamlessly integrating them into SAP S/4HANA, enhancing efficiency, accuracy, and time savings.

Asset Operations

SAP S/4HANA Enterprise Asset Management: Intelligent Maintenance Order Recommendation

LLMs optimize maintenance planning, promote agility, and improve asset performance and reliability in SAP S/4HANA enterprise asset management. In this innovation, generative AI supports users in the assessment of maintenance requests by recommending appropriate tasks and spare parts, thus enhancing operational efficiency. 

Autonomous

Supply Chain Planning



SAP Integrated Business Planning: Interactive Planning Assistant

Demand, supply, and inventory planners often use complex machine learning algorithms. By using generative AI and conversing with Joule in natural language, planners can gain insights into the models’ variables, constraints, and decision processes, leading to much more informed decisions and proactive planning. SAP helps customers boost explainability but also enable what-if scenario simulations for complex planning decisions.

Manufacturing / Industry 4.0

SAP Digital Manufacturing: AI-Driven Visual Inspection

In industrial manufacturing, human-led visual inspection can lead to potential quality issues and is synonymous with a higher degree of manual labor. This capability applies computer vision AI into the production process to automatically identify defects and allocate non-conformance for production decisions, thus supporting a consistent and objective quality analysis that contributes to cost savings associated with reworks and recalls. 

Ready to revolutionize your supply chain using artificial intelligence? Here’s how to get started:

• Dive into the IDC report “The Importance of AI in Supply Chain and Operations” for insights into AI’s pivotal role and its use cases in supply chain and operations

When a bear chases a group of people in the woods, you don’t need to be faster than the bear. Your need to be faster than the slow people.

That is one approach to network security. Your budget is not infinite and eliminating every possible vulnerability may be impossible. But being the most vulnerable target is a big problem.

Last week I attended a meeting of the Project Management Institute (PMI) which featured the founder of a cyber security company. He shared some war stories about some very damaging incidents in cyber security. He said there is a huge global shortage of network security personnel, and the number in the United States is about a half million positions that are not filled. 

His presentation addressed the broad information technology (IT) domain. The industrial domain is a subset of the broad IT domain and is specialized because it contains operational technology (OT). OT is unique because it requires determinism, and therefore utilizes some proprietary operating systems and technologies. 

Serendipitously, after the PMI event last week I listened to the January 23 episode of “Control Amplified”.  Since 2018 this podcast has provided monthly or biweekly episodes covering topics in automation. Most episodes are brief, ranging from 10 to 15 minutes. The Jan 23, 2025, episode is 34 minutes long. It is an interview of Joe Weiss, who is an industry veteran focused on cyber security in automation. Joe has been an evangelist for prioritizing security for quite some time. He is concerned that there has not been sufficient attention paid to security, and suggests there are significant omissions when security is addressed. Specifically, Level 0 and 1 device were singled out as lacking in security.

A reminder about the Purdue model describing the levels in an industrial network:

Level 0 – Instrumentation: the physical connection to the process for obtaining data and taking action

Level 1 – Digitization: the devices that convert signals to digital form and process logic

Level 2 – Supervisory: the control network and applications that rely on deterministic processing, such as HMI, historical data, and advanced controls

Level 3 – Manufacturing Execution System: the interface between the lower deterministic layers (operational technology) and the non-deterministic enterprise layer (Enterprise Resource Planning)

Level 4 – Enterprise Resource Planning: the non-deterministic financial and management applications for the enterprise

A reasonable question is whether every level requires security in its design, and if so to what extent?

• Imagine a “dumb” Level 0 sensor measuring pressure and providing a 4-20 milliamp signal. What security would it require? There is nothing digital in it, so security does not apply.
• Now imagine a smart “digital” sensor that is remotely accessible. It has processing and memory, and could even have logic in it. If that sensor is used in an interlock or control loop, a remote connection might allow someone to hack it and give it a false reading, which would be dangerous.
• However, if that same smart sensor is physically wired to a Level 1 controller, and that controller is on a Level 2 control network, is it sufficient to secure Level 2? Some control networks are proprietary and have security in the design, which can isolate any Level 0 or 1 devices from security concerns.

Therefore, considering different implementations of Level 0 devices the security requirements can differ.

Nearly every project I have worked on in my career involves network communication. I have performed security audits for critical infrastructure (power plants) complying with North American Electric Reliability (NERC) standards. I managed projects to audit security patches and network vulnerability for federal government agencies. During one of my projects, I witnessed a malware attack in progress that we killed by pulling the internet connection out of the switch and recovered from a backup. I have seen facilities such as municipal and federal utilities that air gapped their system (no physical connection outside the building) to isolate them from security threats. I consider myself sensitive to these threats.

I have also seen way too many industrial control networks with Windows 95 machines with no passwords and no security patches. I have seen thumb drives plugged into Level 2 devices without considering the possibility of a virus on that drive, which happened with Stuxnet. I have seen server rooms with no lock on the door. I still have a key fob from a facility I last visited 8 years ago that doesn’t seem interested in having it returned.

While many facilities have significant vulnerabilities, some may have gone too far. There is a major producer in industry that had a severe ransomware attack. They responded by tightening up their security. Sometime afterwards I had to visit a facility because they said the instrumentation system no longer worked. When I arrived, I found there was nothing wrong with the instrumentation system. The new cyber security approached killed OPC communication, preventing anything from the instrumentation to get to the network historian where they looked at the data.

In our industry we have a lot of facilities that need better attention to security. As stated, there is an under supply of qualified network security resources. It becomes more difficult when you need people that understand both the broad IT technology and the OT technology (such as OPC) that make the system work. It needs to be understood by decision makers that network security is the cost of doing business. If the right investments are made in the right resources, you can shed some anxiety and focus on selling production.

Don’t be the most vulnerable facility in industry. Make the bear chase someone else.

Metzger's and Dixon's columns above are thought provoking to read.

The holy grail of being able to access your data anywhere, securely and quickly continues to be a challenge. Most suppliers will tell you they can give you two out of three.

Every instrument, every sensor is an entry point. Then there is the old example of the USB found lying in the parking lot that someone picks up and sticks in their computer to "see what is on it."

AI brings its own set of challenges--we think about speed and access, not so much about security. It just may be that while we are securing all entry points, we also need to be further developing systems that can recognize the bad guys once they get in.

Think about the ubiquitous human employee security badge. It doesn't exist so much to identify employees as it does to identify non-employees that breach security and are spotted on the inside. Maybe this is the next step in solving the security, access and speed conundrum.

Automation and robotics have become essential for businesses striving to remain competitive in today's fast-evolving packaging landscape.

38% of procurement leaders identified enhancing data analytics and spending visibility tools as their top initiatives for the next 12 months.

The first of a two-part series discusses best practices to help engineers significantly reduce model hallucinations

How to overcome the "OEM blockade" that introduces cybersecurity risks for nearly every manufacturer.