Outline the Agent’s Work
Learning Objectives
After completing this unit, you’ll be able to:
- Explain what business process mapping is.
- Describe why process mapping is a useful tool when designing AI agents.
- List some questions to ask about each business process affected by an Agentforce use case.
Trailcast
If you'd like to listen to an audio recording of this module, please use the player below. When you’re finished listening to this recording, remember to come back to each unit, check out the resources, and complete the associated assessments.
Translating a Use Case into an AI Agent
The Coral Cloud team is making great progress with their AI agent. They defined the Agentforce use case, considered the project’s technical requirements, and developed some risk mitigation strategies. But how does the team turn their autonomous AI dream into reality?
If you’ve taken any of the hands-on challenges in the Become an Agentblazer Champion trail, then you know exactly where to whip up an AI agent. That’s right: in Agentforce Builder.
Nora and her project team experiment with Agentforce Builder and prototype solutions in their sandbox. They deeply consider the work their AI agent will do and how to get the agent to execute that work reliably, accurately, and safely. In the same way that Coral Cloud wouldn’t hire a new employee without onboarding them first, the resort shouldn’t deploy an AI agent that can’t follow its business processes.
Your Work Is the Foundation
When the Coral Cloud project team defined their autonomous AI use case, they identified the main job to be done: reservation management. They also listed the tasks and processes related to that work.
- Look up reservation details.
- Resend itinerary or confirmation.
- Book and modify reservations.
- Add a special request, such as early check-in or adding a cot to the room.
- Cancel a reservation.
- Process a refund for cancellation.
But if the Coral Cloud team wants to delegate that work to an AI agent, then those processes must be well-defined. As you can imagine, if the organization can’t describe how that work gets done, it’s nearly impossible to configure an AI agent to do it. One way for Coral Cloud to define the work is to map all the business processes related to reservation management.
What Is Process Mapping?
Process mapping is a visual representation of the steps in a business process that shows how it's done from start to finish. Typically, a diagram (flow chart, swimlanes, or some other visualization) is used to depict the steps and their sequence. And key pieces of the process are represented in the map, such as when and how steps start and end, and what inputs and data are required.
Organizations use process mapping for things like enabling continuous improvement, encouraging transparency and communication, and complying with auditing regulations and standards. But process maps also happen to be a useful tool for designing the behavior of AI agents.
Process Mapping for AI Agents
As you outline the business processes you want to delegate to an agent, consider these questions.
- What goal is the user trying to accomplish?
- How might the user express their goal or intent? How would they phrase their questions or requests?
- What processes are involved in achieving the goal?
- When does each process begin and end? What steps need to be taken? In what order?
- What company policies, rules, and guidelines apply to the process? What guardrails are relevant to each step?
- At what points in the process are decisions made and how?
- What clarifying or follow-up questions might the agent need to ask?
- When would a user need to confirm something before a step proceeds?
- What data is required? What’s optional?
- What input do you need to collect from users? Does that input need to be in a specific format or syntax?
- What are the outcomes of each step or process? How is the output used? How is success measured?
- Are there situations when specific language should be used when communicating with users?
- What questions or paths of conversation should never be pursued?
- Under what conditions should the conversation be escalated to a live service rep?
The Coral Cloud team decides not to create formal diagrams for each of their business processes. But as they prototype their agent, they try to answer the previous questions so the project team has the information they need to design and iterate on the solution.
If you want help visualizing the business processes affected by your Agentforce implementation, be sure to watch this video (4:04 min) for inspiration.
The Building Blocks of an Agent
Now that Coral Cloud has a better understanding of the business processes related to its Agentforce use case, Nora and her team can iterate on the building blocks of the AI agent in Agentforce Builder—its topics, actions, and instructions. With these building blocks, Coral Cloud can translate its business processes into the agent’s setup.
How Agents Work
Before Coral Cloud dives deeper into the prototyping process, here’s a brief review of how agents work. Agents are made up of topics, which define the different jobs an agent can do. Topics include natural-language instructions that tell the agent how to make decisions and what it should and shouldn’t do.
Topics also contain a set of actions, which are the tools the agent can use to do its job. For example, an Order Management topic might include these actions.
- Get order by order ID.
- Get order by email address.
- Look up return status.
- Create return label.
But how do those building blocks work together in a conversation with an AI agent? When a user enters a question or request, the agent selects the most relevant topic and then launches one or more actions included within that topic. That keeps the agent focused on tasks and data that are best suited to the current conversation.
If you’re unfamiliar with the building blocks of an agent, check out this video (4:38 min) or explore the documentation in the Resources section.
Standard Versus Custom
Salesforce provides some standard topics and actions for Agentforce out of the box, so you can get up and running quickly. But you can also create custom topics and actions to give your AI agent additional abilities, so it can perform tasks specific to your business.
Nora reviews the library of standard topics and actions to see if any out-of-the-box functionality is available for Coral Cloud’s reservation management use case. She doesn’t find anything that fits the bill, so her team must go the custom route for their autonomous AI project.
In the next unit, follow along as Coral Cloud designs and iterates on its custom actions and topic.