Getting started with ISLM: how to embed machine learning in SAP S/4HANA

As more businesses begin to look toward integrating artificial intelligence into their daily operations, SAP has a solid ground on which to manage machine learning throughout its platform: Intelligent Scenario Lifecycle Management (ISLM). It is a tool that helps in simplifying and making the entire process of machine learning easier to directly integrate intelligence into ERP systems.

While ISLM is a powerful and often underutilized capability, it’s important to understand that it represents only one of several ways to bring Artificial Intelligence into the SAP landscape. Broadly speaking, SAP offers three complementary approaches:

• Embedded AI, which integrates intelligence directly into the ERP stack through standard “Intelligent Scenarios” and frameworks like ISLM, leveraging HANA’s built-in machine learning libraries (APL, PAL).
• Side-by-Side AI, where advanced models are built and hosted externally (e.g., SAP BTP AI Core, third-party ML platforms) and connected back to S/4HANA through secure APIs.
• Business AI Services, increasingly delivered by SAP as pre-packaged capabilities (e.g., Joule, AI-powered Fiori apps, or industry cloud solutions) that embed AI into end-to-end business processes without requiring custom development.

Each option serves different levels of complexity and maturity: standard embedded scenarios provide quick wins, side-by-side approaches enable scalability and data-science-driven innovation, and business AI services accelerate adoption through turnkey capabilities. In this article, we focus specifically on ISLM, SAP’s native framework for managing the machine learning lifecycle inside the ERP system.

What is ISLM ?

ISLM is a framework that allows businesses to manage the entire lifecycle of machine learning use cases from development and training to deployment and operations within SAP systems. What is unique about ISLM is the fact that it uses datasets directly in SAP ERP, meaning that there is no divergence of business logic and machine learning.

Fundamentally, ISLM seeks to implement machine learning in the SAP environment in a way that is both accessible and scalable.

ISLM relies on two SAP Fiori applications. The Intelligent Scenarios application enables users to create and manage machine learning scenarios. The Intelligent Scenario Management application manages training, deployment, and monitoring of machine learning models.

Together, these applications offer a user-friendly interface for enterprise customers to create, train, and administrate machine learning scenarios.

Deployment architectures: embedded vs. side-by-side

ISLM supports two deployment models:

• Embedded: runs directly within the SAP S/4HANA stack using HANA libraries (APL, PAL). It is best for simpler  use cases that need close integration and fast results.
• Side-by-side: runs outside of the ERP system using a remote ML provider with ISLM orchestrating communication between SAP S/4 HANA and the ML provider. It is ideal for advanced use cases.

The choice between the two depends on use case complexity, system landscape, and performance requirements.

Supported scenario types

Different types of Machine Learning scenarios can be handled by ISLM: Classification and Regression.

Classification is a machine learning method used to predict categories instead of numbers. Classification models help you make decisions by sorting data into groups.

Regression is a machine learning method that uses historical data to predict a number like the sales quantity of a product. It looks at how variables such as product price, order date or customer type have affected sales in the past in the example of a sales scenario. By finding the trend between these data points, it can forecast future sales.

Why use ISLM?

ISLM provides several key advantages to SAP users:

• Fills gap between data science and business applications
  
  
• Extremely simple to embed machine learning scenarios into SAP processes
  
  
• Enables adoption of Machine learning with known tools and environments
  
  
• Improves business outcomes through data-driven decisioning

ISLM enables embedding of machine learning into ERP processes instead of as a decorrelated sideline project.

Use cases

Standard use case

SAP offers pre-delivered smart scenarios, including the Sales Predicted Delivery Processing Delay app, which can forecast likely delays on current deliveries based on a number of variables and historical data. These scenarios are already built into SAP applications, they simply require to be switched on and trained on appropriate business datasets prior to presenting predictions. This makes them easy to deploy and very useful for organizations wanting instant payback from AI with minimal development effort.

Customized use cases

When business requirements go beyond standard offerings, ISLM enables the creation of custom machine learning scenarios:

Scenario Definition & Model Training: Define a unique business scenario, train the model on historical data, and activate it for inference. Below an example on the different steps to perform the scenario creation and Training:

1. In the intelligent scenario app, click on create and select embedded.

• Define your scenario parameters, here we are using a regression.
• Click on “Add Model”

• Define you model parameters, here we are using the standard sales document view but it would be recommended to create a specific custom view that only includes the relevant parameters for the model training.
• Click on Add

• Publish the Scenario

• Once the new scenario is published, you can find it in the the intelligent scenario management app. Access the scenario and click on train.

• Activate the model version

• After deployment, the model exposes predictions through apply views. These can be integrated into any visualization or application making it easy to compare predicted versus actual values. The apply view names can be found in the setting tab of the scenario.

Although building custom scenarios requires additional development to integrate them into applications, the process aligns with standard CDS view practices. In this way, predictions can be fed into business flows to make them actionable to end users.

Side by side machine learning with inference

In a side-by-side configuration, the machine learning model is trained and hosted external to the S/4HANA system, typically on systems like SAP BTP AI Core or any environment that can expose a secure REST API. ISLM serves as the orchestration layer within S/4HANA: it identifies the smart scenario, makes the runtime call to the external model, and marshals the returned prediction back into the SAP business process.

This setup is especially useful when faced with the high-level models, intricate ML designs, or when the entire data science pipeline already exists outside SAP. It natively accommodates real-time inference applications within transactional processing without requiring the model to be placed inside the ERP system itself.

However, for simpler predictive tasks such as binary classification or structured business data regression, embedded scenarios using APL are simpler to implement and keep up. Side-by-side versus embedded architecture is a function of model complexity, infrastructure landscape, and the degree of integration ML pipeline must have with SAP processes.

Conclusion: standard scenarios as quick wins

Custom Machine Learning scenarios offer tailored solutions but typically require additional development efforts. As a contrast, typical intelligent scenarios are already installed in SAP systems and may be simply enabled to improve business processes and yield concrete value to organizations embarking on their AI journey for the first time.

ISLM enables companies to evolve from experimentation with AI to executing AI natively on SAP solutions. Standard scenarios provide immediate benefits with little effort, while custom and side-by-side deployments offer flexibility for complex use cases. The embedded standard scenarios are a good way to quickly achieve business value, while side-by-side options allow to scale towards more complex cases as the company becomes more mature with the use of AI and machine learning.