Introducing Joule

Objective

After completing this lesson, you will be able to identify the fundamental functionalities of Joule within the SAP landscape.

Get to Know Joule: SAP’s AI Copilot

In this unit, we will introduce you to Joule – SAP’s AI copilot that truly understands your business. Joule revolutionizes how you interact with your SAP business systems, making every touchpoint count and every task simpler.

When we are speaking about Joule, we are speaking about one copilot for the SAP ecosystem, serving as a productivity tool to help our customers get their job done.

This means that we have a centralized framework for our cloud solutions that allows users to ask questions and interact with Joule wherever they are via natural language. Whether they are in SAP S/4HANA, using Joule to ask about open HR-requests in SAP SuccessFactors or they are in SAP Start, leveraging Joule to retrieve information from any other SAP cloud solutions.

This also means, Joule has one consistent User Experience which makes its usage as seamless as possible.

We offer Joule out of the box and natively integrated in our cloud solutions. That means there is pre-configured content that is active right away for our customers with limited configuration effort, which you’ll learn more about in this journey.

Joule is available in the first SAP cloud applications today, e.g. SAP SuccessFactors and SAP S/4HANA Cloud public edition & private edition (EAC), with many more SAP cloud applications to follow over the course of the year.

It will not be offered for on-premises.

As everything we are delivering with SAP Business, also with Joule we strive to provide:

  • Relevant content for our users → ensuring Joule understands a user's context. This includes roles and permissions. But also, which SAP products the user is using.
  • Reliable results → as all our AI offerings. Using LLMs means that responses are grounded on SAP knowledge and the business reality of the respective customer and end user.
  • Joule is responsible → this is very important to us, and we steer this together with our partners under a strict prohibition that any of our customers' data is not used for training purposes or any other usage for any 3rd party.

By integrating the power of LLMs and SAP’s rich business context, Joule helps users to easily navigate SAP systems, execute transactions, find information, and get analytical insights, making employees 80% more efficient.

Joule truly understands our customers’ businesses thanks to a host of unique capabilities. Joule:

  • provides contextually aware responses based on the user, specific SAP transaction, and customer business data.
  • maintains the user context and cross-application reasoning across SAP products, no matter whether Joule is accessed from SAP Start, the application’s UI, a mobile device, or somewhere else.
  • grounds result on business documents and links to the referenced sources by leveraging techniques like Retrieval Augmented Generation (RAG), extensible to sources like internal policies hosted on Microsoft SharePoint.
  • inherits user authorization and authentication rights, protecting sensitive data and ensuring compliance.
  • safeguards against inappropriate use, bias, and hate speech by employing guardrails.
  • is designed to put humans in the driver’s seat through preserving human oversight and agency.

Interaction Patterns

With Joule, we are offering four interaction patterns:

  • Transactional Interactions

    This provides our users a direct entry point to SAP, the backend system. Triggering and influencing business processes with the power of natural language and generative AI e.g., purchase orders which need to be reviewed and approved, job positions which are created, OR any other CRUD (Create, Read, Update, Delete) based interactions. All our Cloud products are currently developing such content packages for Joule for most relevant user interactions.

  • Navigational Interactions

    Sometimes, our users want to handle such business processes themselves in the relevant SAP screen. Also, not every business interaction is best served to be handled via natural language. Therefore, we allow users to navigate directly where they want to go. This is especially helpful for users who are not very familiar navigating SAP applications.

  • Informational Interactions

    With informational interactions we are providing our users knowledge-based results. These are, for example, policy related questions. We differ between two categories of informational Joule questions:

    • One is based on SAP owned content with the goal to have all our knowledge-base available via Joule, e.g. with the SAP-help documentation. This is managed by SAP and offered to every Joule user out of the box.
    • Additionally, we are enabling customers to upload customer owned content. This allows customers to upload their own HR policies, travel policies, information content, and others. We call this functionality Document Grounding.
  • Analytical Interactions

    Additionally, we plan to introduce analytical interactions. Coming later in 2024 we plan to enable analytical interactions based on JustAsk and SAP Analytics Cloud, enabling customers to leverage Joule for analytical queries. In 2025, we plan to integrate Joule into SAP Analytics Cloud and further evolve the analytical capabilities of Joule.

This image shows that Joule is designed to handle various tasks such as transactional, navigational, informational, and analytical queries. The interface displays employee selection, position details, approval processes, and sales analytics, emphasizing efficient and interactive user engagement.

Document Grounding

The Document Grounding capability in Joule can provide more comprehensive responses by drawing from business documents located in SAP and third-party repositories.

With Document Grounding, we enable customers to upload their own content (e.g. HR-policies, travel-policies, etc.). This enables their users to ask questions and receive answers based on these documents via Joule. Joule then finds and considers document extracts relevant to the user request and responds with a concise natural-language answer and includes its sources.

This capability deepens how Joule reliably delivers answers grounded in a customer's structured and unstructured business data. For example, in SuccessFactors, users leverage Joule to get answers to questions about their organization’s HR policy, saving time and enhancing the overall employee experience.

What is Grounding

"Grounding" is a process that combines generative Large Language Models (LLMs) with advanced information retrieval to improve the quality and accuracy of responses, without the time, complexity, and expense of training or fine-tuning an LLM with company-specific data. Grounding uses a customer’s own knowledge sources (e.g. travel policy, enablement content, service manuals, FAQs, etc.) to supplement an LLM's internal representation of information, which can help models be more accurate and reliable.

Grounding can also help models build trust by citing source documents, lowering ambiguity and reducing the possibility of hallucination.

How Does Document Grounding Work?

The image illustrates how AI-assisted document grounding enhances HR user experience. Jane enters an HR policy question into Joule. The system retrieves relevant text from the SAP HANA Vector Database, enhances the query with text chunks, and uses an LLM to generate a summarized response.

A customer can connect their document repositories via their own Microsoft SharePoint.

Document Grounding takes the relevant customer-owned documents via Microsoft SharePoint and creates so called, embeddings.

This means, we separate the text in different chunks and vectorize them. Embeddings refer to dense, continuous vectors representing text in a high-dimensional space. These vectors serve as coordinates in a semantic space, capturing the relationships and meanings between words.

When a user-utterance comes in, we find the most similar vectors. With this semantic search, SAP HANA Vector database and embeddings enable much granular concepts with meaning what a user is trying to say.

In the context of LLMs, embeddings play an important role in retrieving the right context for RAG.

We use this heavily in Joule to process a large amount of data, e.g. the SAP Help Documentation with all available languages consists of millions of pages.

Log in to track your progress & complete quizzes