Exploring Joule & Joule Agents

The key differentiator for Joule, compared to general-purpose AI like ChatGPT, is its business context. Joule doesn't just have access to a large language model; it is deeply integrated with and "grounded" in SAP's business data and logic. This means Joule understands:

• Roles: It knows if the user is a sales manager, an HR representative, or a supply chain planner and tailors its responses accordingly.
• Revelant Company Data: It has secure access to real-time data within the SAP system (e.g., sales figures, employee records, supply chain status).
• Business Processes: It knows the steps required to create a purchase order, approve a leave request, or analyze a sales pipeline.
• User Permissions: It respects all existing security and data privacy roles, ensuring users only see and interact with data that they are authorized to access.

Joule functions as a conversational chatbot that can perform a wide range of tasks:

• Answering Questions & Retrieving Information: Users can ask complex questions in natural language, and Joule will query the system to find the answer:
  • Example: "Show me all overdue invoices for my top three customers in North America."
  • Example: "Which of my team members have not completed their mandatory compliance training?"
• Summarizing Complex Data: Instead of navigating through multiple screens and menus, users can command Joule to perform actions:
  • Example: "Summarize the sales performance for the DACH region for the last quarter."
  • Example: "Give me the key points from the latest supply chain disruption report."
• Executing Tasks & Transactions: Instead of navigating through multiple screens and menus, users can command Joule to perform actions:
  • Example: "Create a purchase order for 100 units of product X from supplier Y."
  • Example: "Start the onboarding process for our new hire, Jane Doe."
• Generating Content: Joule can create new content based on templates and data within the system:
  • Example: "Draft a job description for a Senior Financial Analyst based on our standard template."
  • Example: "Write an email to my team congratulating them on exceeding our sales target."
• Navigation and Guidance: For new or infrequent users, Joule acts as a guide:
  • Example: "Where do I go to approve travel expenses?"
  • Example: "Take me to the customer relationship management dashboard."
• Coding and Development (for developers): In SAP BTP services such as SAP Build Code and also in ABAP Development Tools In Eclipse, Joule can assist developers by generating code, explaining code snippets, and accelerating the development process:
  • Example: "Generate a CAP application to process sales orders being received from a business partner."
  • Example: "Generate a CDS view with the following fields..."

As mentioned earlier SAP's AI strategy is centered around AI being everywhere both visible and behind the scenes. Joule is available in a variety of products across the SAP portfolio with more products being added in the future.

Joule in the SAP BTP Cockpit enables administrators to use AI to perform a variety of tasks such as (but not limited to):

• Adding a new user
• Listing global accounts or subaccounts and see details
• Viewing available Cloud Foundry environments

This is the most fundamental interaction pattern, similar to how you would use a search engine or a traditional chatbot. The user's goal is to retrieve information or get an explanation.

• Core idea: Find and present existing information.
• What it does
  • Answers "who," "what," "where," and "when" questions.
  • Retrieves specific data points from the system (e.g., a purchase order status, an employee's start date).
  • Explains business terms or processes based on SAP documentation or company-specific knowledge.
  • Finds relevant help documents.
• Example
  • User: "What is the status of purchase order 4500018543?"
  • Joule: "Purchase order 4500018543 was delivered on November 15, 2023. The corresponding invoice has been paid."
• Why it's important: This pattern reduces the time users spend searching for information, navigating complex menus, or needing to know specific transaction codes. It democratizes access to data.

Similar to the informational pattern the navigational pattern also gets information for the user but instead of retrieving the information and bringing it back to the user Joule instead brings the user to where the information is located. In other words Joule interprets if the users want to navigate to a business application and then navigates them directly to where they want to go. This pattern can be very useful for casual users of applications who do not have extensive experience navigating the systems in question.

• Core idea: Take me to the application that will provide me with the answer I'm looking for.
• What it does
  • Navigates the user to the application that will answer the question.
  • Hides the details of the underlying navigational details.
  • Allows the user to "take over" once the application is visible to the user.
• Example
  • User: "Show me all open production orders."
  • Joule: Joule will show a list of open production orders. It will also provide a link that redirects the user to the SAP Cloud ERP Manage Production Orders Fiori application.
• Why it's important: This pattern reduces the time users spend trying to figure out the navigational details of applications.

This pattern moves beyond just finding information to actively doing something in the system. The user asks Joule to perform a business transaction on their behalf.

• Core idea: Execute a business process or action.
• What it does
  • Creates, updates, or deletes business objects (e.g., creating a sales order, updating a customer's address).
  • Submits requests for approval (e.g., submitting a leave request, a travel expense report).
  • Initiates a business process that may have multiple steps. Joule can guide the user through it, asking for necessary information along the way.
• Example
  • User: "Create a purchase requisition for 25 standard office chairs from our preferred supplier, 'Office Supreme'."
  • Joule: "Okay. I've drafted a purchase requisition for 25 chairs from Office Supreme. The total cost is $3,750. Do you want me to submit it for approval?"
• Why it's important: This pattern is a major efficiency driver. It streamlines workflows, reduces manual data entry and clicks, and empowers casual users to perform tasks that might otherwise require extensive training.

This final pattern focuses on helping the user understand data by summarizing, comparing, and visualizing it to uncover trends and insights. It's about answering the "why" and "what if" questions.

• Core idea: Analyze data to provide summaries, trends, and insights.
• What it does
  • Answers comparative questions (e.g., "Compare sales in Germany vs. France this quarter") .
  • Summarizes large datasets into key performance indicators (KPIs).
  • Identifies trends, anomalies, or correlations in business data.
  • Generates charts, graphs, and other visualizations to make data easier to understand.
• Example
  • User: "Show me the top 5 selling products in the EMEA region for the last quarter and visualize the monthly sales trend for each."
  • Joule: "Here are the top 5 products in EMEA for Q3. Product 'X-100' was the top seller. I have generated a line chart showing the monthly sales trend for all five products. You can see a significant spike for 'X-100' in September." (Joule would display the chart)
• Why it's important: This makes data analysis accessible to non-analysts. Managers and business users can get immediate answers to complex business questions without needing to run custom reports or rely on a separate BI team.

Just as GenAI gave rise to Joule, Agentic AI has given rise to Joule Agents. Joule Agents represent the next logical enhancement beyond Joule acting as intelligent, proactive, and autonomous entities capable of performing complex, multi-step tasks across various SAP solutions. In essence, SAP Joule Agents are SAP's vision for the future of enterprise interaction – moving from clicking through menus to simply telling the system what you need, and letting the intelligent agent figure out how to get it done.

At their core, SAP Joule Agents are AI-powered automation workflows that live within the SAP Joule copilot. They are designed to understand user intent expressed in natural language, plan and execute a series of actions across different SAP systems (like SAP Cloud ERP, SAP SuccessFactors, SAP Concur, SAP Fieldglass, etc.), and provide intelligent responses or complete tasks on behalf of the user.

Joule Agents provide organizations with a complete Agentic AI solution. Some of their main features are:

• Intelligent & Context-Aware:Understands nuances, remembers past interactions, and applies relevant business context.
• Proactive & Autonomous:Can anticipate needs, suggest next steps, and execute tasks without constant manual oversight.
• Multi-Step Task Automation:Handles complex workflows involving multiple applications and data points.
• Seamless Integration:Works natively across the entire SAP portfolio (SAP Cloud ERP, SAP SuccessFactors, SAP Ariba, SAP Concur, etc.) and potentially with third-party systems via extensibility.
• Natural Language Interface:Users interact using conversational language, similar to speaking with a human assistant.
• Secure & Compliant:Operates within enterprise-grade security and governance frameworks, respecting data privacy and access controls.
• Extensible:Organizations can train and customize agents or build new ones to fit their specific business processes and unique requirements.

SAP Joule Agents leverage a combination of techniques:

• Natural Language Understanding (NLU) and Generative AI LLMs:When a user types a request into Joule, the NLU engine interprets the intent, and LLMs help in reasoning, planning, and generating human-like responses.
• Tool Orchestration:Agents aren't just LLMs; they are integrated with a vast array of "tools" or "skills." These tools are essentially pre-built connectors to specific SAP APIs, business processes, or data points. For example, a "create purchase order" tool, a "check inventory" tool, or a "generate report" tool.
• Planning and Execution Engine:Given a user request, the agent's underlying intelligence plans a sequence of steps (which tools to use, in what order, with what parameters) to fulfill the request. It then executes these steps, interacting directly with the relevant SAP backend systems.
• Context Awareness:Agents maintain context throughout a conversation, remembering previous interactions and understanding the current state of the user's work, leading to more relevant and seamless interactions.
• Learning and Adaptation:Over time, agents can learn from user interactions, feedback, and new data to improve their performance and accuracy.

Looking at the graphic above in conjunction with the techniques just mentioned we can summarize Joule Agents as follows:

When a user interacts with a Joule Agent, their input serves as the starting point. The agent then leverages its Goals & Instructions module—guided by Large Language Models (LLMs)—to plan and reason through the task at hand. This involves dynamically analyzing different possibilities to achieve the desired outcome. Once the approach is determined, the agent enters the Execution phase, where it uses tools to perform specific operations required to complete the task. These operations may include accessing external systems such as databases, applications, or even the web to retrieve or transmit information. The agent's ability to use memory ensures that ongoing interactions and decisions are informed by past experiences, enhancing its effectiveness and adaptability over time. This seamless integration enables AI Agents to handle complex scenarios and deliver solutions tailored to specific business needs.

• Human Resources:
  • "Process John Doe's onboarding, send him welcome package."
  • "Approve all outstanding leave requests for the marketing department."
  • "Find all employees eligible for promotion in Q4."
• Procurement:
  • "Create a purchase order for 100 units of widgets from Vendor X, delivered by next Friday."
  • "Check the status of PO 12345."
  • "Find alternative suppliers for raw material Z with a delivery time under 5 days."
• Finance:
  • "Generate a sales report for Q3 comparing actuals against budget."
  • "Process this expense report and flag any discrepancies."
  • "Summarize cash flow for the last month and highlight anomalies."
• Sales & Service:
  • "Update customer ABC's contact information and log our last meeting notes."
  • "Create a new sales opportunity for product X for account Y."
  • "Provide an update on order #67890."