Skip to main content

Introducing Joule, the SAP Generative AI Copilot

Artificial Intelligence has evolved rapidly over the past years, with Machine Learning, Deep Learning, and Generative AI being its core capabilities. Generative AI has made a profound impact in simplifying the usage of modern technology platforms such as SAP. The crux of Generative AI in SAP lies in its inherent ability to simplify the development process by enabling easy interaction between users and the application in a natural language.

Trends in Generative AI

As quoted by SAP, a McKinsey1 study shows Generative AI adds $2.6 to $4.4 trillion annually to the global economy. Here are some trends from a diverse sample of organizations.

  • 33% use some form of Generative AI across a specific business function.
  • 40% want to increase their AI investments based on the latest advancements in Generative AI.
  • The best part is that 60% of organizations adopting Generative AI had increased revenue.

The SAP Copilot-Joule

SAP has recently announced its own AI-powered Generative Copilot, called Joule, set to be deployed shortly across all applications in the platform. It aims to enable even novice users to seamlessly use the SAP platform, with automated code generation, analytics, and content creation, to name a few. While SAP’s foray into infusing AI capabilities in its solutions has been a part of its evolution in the past years, Generative AI Copilot Joule will undoubtedly be a game changer.

The current versions of SAP applications come with AI-enriched capabilities across various functions, such as source to pay, recruit to retire, lead to cash, and design to operate, amongst others, and the SAP Copilot Joule aims to take it further and more straightforward.

SAP uses Natural Language Processing (NLP) capabilities to interact with Foundation and Large Language Models (LLM). A model can be defined as a vast dataset containing billions of data parameters, provided by either SAP (planned) or its Partners or custom-built. They have an infrastructure similar to the neural networks that a human brain contains. Current LLMs used in the SAP environment include GPT-4, Aleph Alpha, and Azure OpenAI.

Benefits of Using Joule Copilot for SAP

  • Streamlines tasks within the SAP application and support your activities with a tailor-made approach that understands your role and required outputs.
  • Users can get instantaneous and contextually relevant output with just a simple prompt.
  • Delivers enhanced analytical insights that help you better manage your business processes and move to an enriched decision-making process.
  • The SAP Copilot Joule scores high on data privacy with built-in security systems that enable users to work in a safe and highly regulated environment.
  • Benefit from cost savings when compared to manual operations. Users also save on resource utilization and move your employees to more complex roles.
  • Generative AI in SAP with Joule democratizes the use of applications in your organization and simplifies the interaction between man and machine. It also includes sentiment analysis features.

With the above statistics and the accelerated adoption path, now is the time to make an imperative move. SAP Business AI gives you a simplified, robust, and reliable model to move towards Generative AI across its various applications using the Copilot Joule.

Use Cases for Generative AI in SAP

SAP Business Technology Platform (BTP) will have generative AI capabilities that simplify the development process for coders and citizen developers. The SAP Generative AI HUB is a centralized environment for using SAP Generative AI and includes Model Grounding, Prompt Editor, and Prompt Management.

Let’s look at some use cases where Generative AI in SAP can simplify your operations.

Supply Chain: Generative AI in SAP simplifies demand planning by accessing various parameters and data to drive insights, including proactive alerting and recommendations.

Document Information Extraction: While SAP currently has a document extraction feature for some widely used scenarios, Generative AI in SAP takes it forward by expanding the scope for almost all business documents in over 40 languages.

The SAP Transportation Management Application, which has many documents that need to be analyzed and processed, can use Generative AI. You stand to benefit from reduced time spent on manual extraction, improving your operational efficiency and reducing errors.

SAP Analytics Cloud: Within the SAP Analytics Cloud, you can use natural language prompts to query the database and get real-time insights. Using Large Language Models (LLMs) ensures secure retrieval per your data policies. You can leverage workflow compatibility across multi-cloud models and connect to external data sources.

SAP SuccessFactors: With the SAP Copilot Joule, you can automate the generation of job postings by simply using a natural language prompt. You can use multiple parameters, generate supporting descriptions, and create approval workflows. Joule also enables the creation of interview questions and other recruitment documents.

SAP Customer Experience (CX) and CRM: Joule simplifies extracting information and action items from customers, such as emails, support requests, and workflow-based tickets. You can get a quick summary of relevant information, along with sentiment analysis, and can also create responses.

SAP Build: With the SAP Copilot Joule, you can automatically create code, edit your existing code fragment, and create related data models per your requirements.

The infographic below shows the announced Generative AI capabilities across the SAP ecosystem, as highlighted in blue.

SAP Business AI

Limitations of Generative AI in SAP and Ways to Mitigate Inaccuracy

With the above features that make Generative AI in SAP more efficient and productive, the solution comes with its limitations, such as only having relevant information until the creation of the model. There are also several scenarios where the output will be inaccurate, such as providing biased data and generation of Hallucinations (wrong but seemingly accurate results), which calls for a need to align them with business requirements.

Take a brief look at the various methods to help align Generative AI in SAP to your business processes and data, collectively termed “Grounding” of your foundation models.

Aligning Generative AI in SAP to Your Business Context

Prompt Engineering

Large Language Models are stateless and do not have prompt recollection features. This challenge requires users to give task-specific information in the prompt, and that’s where rapid engineering comes into place.

Prompt engineering involves elaborating the prompt, giving specific instructions, examples, and references, or making a well-defined structure of the desired output to give in-context learning to the model.

Embeddings and Retrieval Augmented Generation (RAG)

Data can also be stored in the model using vectors, which are numeric representations of information with a semantic meaning. Using a machine learning algorithm, vectors are stored as embeddings in the LLM and can be recalled by referencing the required numeric code using vector similarity scoring. Embedding grounds the prompt with the required information and references the output’s source.

Orchestration Tools

Using orchestration tools is an advanced technique to augment the effectiveness of prompts. The LLM accesses the tools that retrieve the required API specification, uses it to connect to the system having the relevant data, and creates the output.

Fine-tuning

Fine-tuning is a widely accepted grounding solution for enhancing output quality. It retrains a foundation model on a different data set and examples of the required input/output format. This practice improves the domain-specific task performance with the required solutions. Inherent drawbacks of this approach are being computationally complex and expensive in collecting the vast training data.

However, Fine-tuning is well suited for improving task-specific efficiency for medium-sized foundation models but is not recommended as a first approach to ground your foundation models.

Best Practices for Adaptation of Foundation Models

  • Generic Generative AI models might not be accurate in giving output as per a specific business context. To ensure accuracy, you must ground and train these models on your particular business data.
  • Ensure that the grounding begins within simpler processes such as prompt engineering and Retrieval Augmented Generation (RAG) rather than Fine-tuning.
  • Test, adapt, and optimize different models based on performance and pricing.
  • Ensure proper governance and change management within your organization.

The above features of SAP Copilot-Joule will greatly simplify business operations across the SAP suite. Generative AI in SAP puts the power of application usage into the hands of citizen developers, driving greater business value and a definite cost advantage. To learn more about how this can benefit your business, contact us; we will gladly help.

References:

  1. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-in-2023-generative-ais-breakout-year

About the author

New Era Technology

New Era Technology creates powerful digital, data, and technology solutions that keep companies moving forward in a rapidly changing world. Our 500+ consultants use data insights, experience design, and tech innovation to help you reimagine your business for whatever comes next.

Ready To Talk?

Let us know how we can help you out, and one of our experts will be in touch right away.