Architect the Right Solution

Learning Objectives

After completing this unit, you’ll be able to:

Define a data strategy for your Agentforce project.
Develop a solution for connecting the AI agent to channels.
Specify criteria for escalating conversations from AI to live service reps.
Plan the security permissions for the Agentforce project.
List some Salesforce-specific factors that can influence your Agentforce implementation.

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.

From Goals to Solutions

In the previous unit, Coral Cloud defined the business goals for its autonomous AI use case and identified the work the AI agent will do and its jobs to be done. With objectives in place, Nora can kick off the discovery process so that the organization can gather and document the project’s requirements.

A Note About Developing Agents

Although thorough planning is crucial for the success of any project, building AI agents is unique and requires a more hands-on approach. Unlike traditional software development, where extensive upfront planning and discovery is common practice, AI agent development requires prototyping and continuous refinement. You can’t plan an AI agent solely with design documents and spreadsheets.

As you define your project requirements, simultaneously build and test your AI agent in a sandbox environment. Try different approaches, figure out what works and what doesn’t. The key is to use testing and iteration to guide your planning and development efforts.

Think About the User Journey

It’s important to take a user-centered approach when designing any technology solution, and Agentforce is no exception. Nora and her team visualize the high-level interactions between the agent and all the users who engage with the agent.

Thinking about the user experience can help you understand the project requirements for your AI agent. For example, internal users often have broader and more complicated queries for an agent than external users, whose needs are often more predictable.

Nora and her team sketch out a rough journey map to highlight the stages and touchpoints in the user experience for reservation management. If you’re unfamiliar with journey mapping techniques, check out Journey Mapping.

Gather Technical Requirements

After the Coral Cloud team spends time thinking about the desired user experience for their AI agent, they can gather the technical requirements. These are the main factors they explore.

Data
Channels
Routing and escalation
Security controls
Salesforce considerations

Define the Data Strategy

In the previous unit, the Coral Cloud team evaluated the data readiness for each of their use-case ideas. One reason they chose the reservation management use case is because their data can support it. And that’s vital because having the right data vastly improves the accuracy and reliability of an AI agent.

Now, Nora needs to establish the data requirements for the project. Coral Cloud uses Salesforce to manage reservations at all of its resorts using a data model with custom objects. So she knows she can use data from those reservations and any relevant external data and uploaded files. She makes sure to audit the reservation data in Salesforce to assess its completeness, quality, understandability, cardinality, uniqueness, accessibility, recency, security, and governance. Then she defines the project’s data strategy.

If you’re not sure how to create a data strategy for your AI project, check out AI + Data: Project Planning. You can also explore free tools, like Cuneiform, on AppExchange, to help you back your intuitions with a technical analysis of your data.

Specify the Channels

When planning its Agentforce solution, Coral Cloud also needs to consider the channels where the AI agent will operate. How will the agent engage with internal or external users? How will the agent be presented to users? What’s the desired experience?

A stylized illustration showing an AI agent interacting with an end user on a website.

For the first autonomous AI use case, Coral Cloud starts with its messaging channels: the company website, WhatsApp, and Facebook Messenger. And then down the road, Coral Cloud can add phone and email.

Coral Cloud has an Experience Cloud website, and the resort is using Messaging for In-App and Web. Messaging for In-App and Web supports all the channels where it plans to deploy its AI agent, so Coral Cloud is in good shape for its Agentforce implementation.

Keep in mind that you can prototype an AI agent in your sandbox before configuring the channels if you want to test out your ideas for the agent’s design. But you need to establish your channel strategy before deploying to production and thoroughly test the agent’s performance on each channel.

When Do You Escalate?

Nora is excited about the new AI agent being able to interact with customers on Coral Cloud’s website and chat channels. But she also knows that the AI agent sometimes needs to escalate to a live service rep in certain situations. Some reasons for escalating to a service rep can include company policy, brand requirements, security measures, risk management, or regulatory compliance.

During the planning process, Coral Cloud defines all the ways in which human decision-making and oversight will occur during the course of the AI agent’s work. Clear guidelines about when and how service reps should step in will make it easier for Nora and her team to configure the right guardrails as they prototype the agent. (You learn more about guardrails in the next unit.)

Channel Routing

Speaking of routing from an AI agent to a service rep, Coral Cloud also needs to think about how its Agentforce project affects the resort’s current channel-routing solution. The organization is using Omni-Channel for intelligent routing and automation, so Nora’s already thinking about how the team can build an integration between the AI agent and Omni-Channel.

For the integration, Nora is following the steps in the Connect Service Agent to Customer Channels documentation. Keep in mind that the way you set up your channel-routing solution depends on your use case and the configuration of your Salesforce org.

Establish Security Controls

As the head of business technology, Nora keeps security top of mind, and Coral Cloud consistently follows security best practices when they build technology solutions on the Salesforce Platform. But this isn’t just any technology project—it’s an Agentforce project, which is powered by generative AI.

Nora found information online about how Agentforce uses the robust Salesforce security infrastructure, including the Einstein Trust Layer. The Einstein Trust Layer uses a secure gateway and zero-retention agreements to protect company data. That puts her mind at ease about the security of the new AI agent. She decides to learn more by downloading the Agentforce and Einstein Generative AI Security white paper.

Next, Nora considers the security controls that need to be in place to make sure Coral Cloud’s AI agent is trustworthy. When you start planning your own AI agent, consider permissions and access for admins, employees, customers, and the AI agent itself.

Admin Access

To create and manage AI agents in Salesforce, you need the Manage AI Agents user permission and the required permissions for your agent type, or the Customize Application user permission.

Employee Access

If your use case is internal and your employees interact with the AI agent in Salesforce, the agent runs in the context of the currently logged-in Salesforce user. The standard Salesforce access controls—such as licenses, permissions, field-level security, and sharing settings—determine what each user can access. See Trust and Agents for more information.

Customer Access

If your agent is deployed to external channels, such as your website, there are probably certain actions you don’t want an AI agent to perform on behalf of customers unless they’re verified. For example, maybe anyone who visits your website can ask the agent questions about your products. But if someone needs help with an order, the user first has to authenticate.

The way you design your authentication solution for Agentforce depends on the security and identification requirements for your particular use case. Check out Maintain Trust with Agentforce Actions to learn more.

Agent Access

The final piece of the security puzzle is the agent user. Some AI agents, such as service agents, operate as an agent user, and the agent user has a dedicated user profile and role in Salesforce. The actions the AI agent can take depend on the permissions it’s assigned.

By default, the agent user has a limited set of permissions. And from a security perspective, that’s great because it complies with the principle of least privilege. But it also means that when you set up your AI agent, you must specifically grant the agent all the permissions it needs to do its work, or else it won’t function properly. See Best Practices for Agent User Permissions to find out how to control what your AI agent can do and what data it can access.

When you’re configuring the agent’s permissions in Salesforce, you probably won’t get it right the first time, so be sure to test the security controls in your sandbox when you’re prototyping your agent. To watch a demo about agent access, check out the Set Up Agent Permissions (6:03 min) video.

Salesforce Considerations

If you’re an existing Salesforce customer, you’re not building your AI agent in a vacuum. Your AI agent needs to coexist with your current Salesforce configuration and architecture. Here are a few Salesforce-specific factors that can influence your Agentforce implementation.

Licensing and Provisioning

Each Agentforce project is unique, so the Salesforce products and features you need for your project vary depending on your use case. It’s best to check with your Salesforce account executive to confirm the licensing requirements for the specific AI agent you want to build.

Billing and Consumption

Agentforce uses a consumption-based pricing model. For more information on how usage is billed, see the documentation linked in the Resources section. You can also refer to your contract or contact your account executive.

Requirements

To use Agentforce, these products and features must be enabled in your Salesforce org.

Lightning Experience
Einstein Generative AI
Data Cloud

Orgs and Environments

Regarding your orgs and environments, here are a few things to think about.

How many Salesforce orgs are affected by this Agentforce project?
Is Data Cloud enabled in your sandbox environment?
If you’re deploying your AI agent to messaging and in-app channels, what’s your current chat solution?
Do you want to start developing agents in a sandbox or production? In most cases, a sandbox helps you save on costs when you’re creating and testing.

Einstein Bots

Have you implemented Einstein Bots in your Salesforce org? If so:

How are the bots being used today?
How is performance measured?
Are there specific business outcomes you’re trying to improve with generative AI?
Are you considering turning your bot into an AI agent?

Existing Automation

With Agentforce, your AI agents use agent actions as tools to accomplish specific tasks. Those agent actions are built on top of existing Salesforce Platform technology, such as flows, Apex, and prompt templates. Start cataloging any existing automations related to your use case that you might be able to repurpose for your AI agent.

Nora has done a thorough job considering the requirements for Coral Cloud’s new AI service agent. Next, her organization evaluates the risks associated with the project and thinks about guardrails and governance.

Tempo stimato

Argomenti

Hai bisogno di aiuto?

Risorse per Agentforce