Introducing Joule-SAP’s Generative AI Copilot

Artificial Intelligence has undergone rapid evolution in recent years, propelled by advancements in Machine Learning, Deep Learning, and Generative AI, which constitute its foundational pillars. Generative AI has emerged as a transformative force, significantly streamlining the utilization of modern technological platforms like SAP.

At the heart of Generative AI’s impact within SAP lies its capacity to simplify the development process, facilitating seamless interaction between users and applications through natural language. 

Generative AI Trends

A McKinsey¹ study reveals that Generative AI contributes between $2.6 to $4.4 trillion annually to the global economy. Here are some notable trends observed across a diverse spectrum of organizations:

• 33% leverage Generative AI in specific business functions.
• 40% express intent to augment their AI investments in light of recent Generative AI advancements.
• Remarkably, 60% of organizations embracing Generative AI report increased revenue.

The Advent of SAP Copilot-Joule

SAP recently unveiled its own AI-driven Generative Copilot called Joule. It is poised to be integrated across all applications within the platform. Joule aims to empower even novice users to navigate the SAP ecosystem seamlessly, offering features such as automated code generation, analytics support, and content creation.

While SAP’s journey toward embedding AI capabilities into its solutions has been ongoing, the introduction of Generative AI Copilot Joule is poised to be a transformative game-changer.

Present iterations of SAP applications already have AI-enhanced functionalities across various domains, including source-to-pay, recruit-to-retire, lead-to-cash, and design-to-operate. SAP Copilot Joule represents a consistent effort to advance and simplify these capabilities.

Utilizing Natural Language Processing (NLP), SAP interacts with Foundation and large language models (LLMs). These models are similar to extensive datasets comprising billions of data parameters. These models, akin to neural networks found in the human brain, are either provided by SAP (planned), its partners, or custom-built. Notable LLMs currently utilized in the SAP environment encompass GPT-4, Aleph Alpha, and Azure OpenAI.

Advantages of Utilizing Joule Copilot for SAP

• Simplifies SAP tasks and offers support based on your role and requirements, enhancing efficiency.
• Provides instant, contextually relevant output with minimal input, expediting decision-making.
• Offers advanced analytical insights, facilitating better business process management and decision-making.
• Ensures high data privacy through built-in security measures, fostering a safe and compliant working environment.
• Generates cost savings compared to manual operations, optimizing resource allocation, and enabling staff to focus on more strategic tasks.
• Democratizes application usage and enhances human-machine interaction, incorporating sentiment analysis capabilities.

Given the above benefits and the growing adoption trajectory, transitioning to SAP Business AI with Copilot Joule presents an essential opportunity. This model offers a simplified, robust, dependable pathway towards leveraging Generative AI across various SAP applications. 

Generative AI Use Cases in SAP

• Supply Chain: Simplifies demand planning through data-driven insights, including proactive alerting and recommendations.
• Document Information Extraction: Expands document extraction capabilities across diverse business documents in multiple languages, enhancing operational efficiency.
• Transportation Management: Accelerates document analysis and processing, reducing manual effort and errors.
• Analytics Cloud: Enables real-time insights through natural language queries while ensuring data security and compatibility across multi-cloud environments.
• SuccessFactors:  Joule automates job posting generation, recruitment documentation, and approval workflows, enhancing HR processes.
• Customer Experience (CX) and CRM: These services streamline information extraction from customer interactions, provide sentiment analysis, and support response generation.
• SAP Build: Facilitates automatic code generation, editing, and data model creation, streamlining development tasks. 

Leveraging generative AI capabilities within SAP applications can help organizations optimize operations, enhance decision-making, and drive innovation across various functions.

Despite the efficiency and productivity benefits of Generative AI in SAP, it’s crucial to acknowledge its limitations. One limitation is restricting only relevant information up to the model’s creation. This leads to potential inaccuracies in output generation, including biases and hallucinations—seemingly accurate but ultimately incorrect results. To align Generative AI in SAP with business requirements and mitigate inaccuracies, various methods collectively termed “Grounding” can be employed. 

Aligning Generative AI in SAP with Business Context 

Prompt Engineering: Large Language Models lack prompt recollection features, necessitating users to provide task-specific information in prompts. Prompt engineering involves elaborating prompts with specific instructions, examples, references, or well-defined output structures to facilitate in-context learning for the model.

Embeddings and Retrieval Augmented Generation (RAG): Data is stored as vectors, numeric representations with semantic meanings. These vectors, stored as embeddings, can be recalled using vector similarity scoring to ground prompts with relevant information and output sources.

Orchestration Tools: Advanced tools can enhance prompt effectiveness by enabling LLM access to retrieve API specifications, connect to systems with relevant data, and generate output accordingly.

Fine-tuning: This method involves retraining foundation models on different datasets and examples of required input/output formats to improve domain-specific task performance. While computationally complex and resource-intensive, fine-tuning enhances task-specific efficiency, particularly for medium-sized foundation models.

Grounding: Employing these grounding techniques helps align Generative AI in SAP with business processes, mitigating inaccuracies and maximizing utility.

Adapting foundation models to specific needs is crucial to ensure accurate output aligned with your business context. Here are some best practices:

Grounding and Training: Generic Generative AI models must be grounded and trained on your business data to ensure accuracy. This process customizes the models to your requirements and improves output relevance.

Start with Simpler Processes: Begin grounding with simpler techniques like prompt engineering and Retrieval Augmented Generation (RAG) before resorting to more complex methods like Fine-tuning. This gradual approach allows for better understanding and optimization of model performance.

Test, Adapt, and Optimize:

• Continuously test different models.
• Adapt them to evolving needs.
• Optimize based on performance and cost considerations.

This iterative process ensures that the models remain effective and efficient over time.

Governance and Change Management: Establish proper governance and change management protocols within your organization to oversee the adaptation process effectively. This ensures alignment with business objectives and compliance with regulations.

The features offered by SAP Copilot-Joule promise to streamline business operations across the SAP suite. Generative AI in SAP empowers citizen developers, driving increased business value and cost advantages. Please contact us for further insights into how this can benefit your business.