Getting Started

Autonomous AI is transforming the way organizations operate, and Salesforce’s Agentforce is at the forefront of this revolution. The product was made generally available by Salesforce in October 2024. Whether you want to streamline case management, enhance lead nurturing, or delight customers, Agentforce empowers businesses to accomplish more with fewer resources. In this post, we’ll share five practical tips to help you successfully implement and use Agentforce. 

Feeling anxious about diving all in with Agentforce? Contact the Sercante team for an Agentforce readiness assessment. That way, you can be sure you’re getting set up for success before you implement Agentforce in your org.

Understanding Agentforce

Before diving into the tips, let’s take a closer look at what Agentforce is.

Agentforce enables autonomous AI agents to perform tasks without human intervention, acting as digital workers within Salesforce or external customer channels. These agents enhance productivity by automating routine tasks and assisting with complex ones. With tools like Agent Builder, you can customize agents using pre-built topics and actions or create entirely new ones tailored to your organization’s needs.

Agentforce integrates seamlessly across the Salesforce platform, leveraging Data Cloud for reasoning and learning. Out-of-the-box agents include Service Agents for case deflection, with more capabilities to be released in December 2024, such as SDR and sales coaching agents.

Unleashing the Power of Agentforce: Five Steps to Get Started

Follow these five steps to get started on the right foot when you dive into Agentforce.

Tip 1: Identify Use Cases

Start by identifying where Agentforce can deliver the most value in your organization. Ask yourself:

• How are you using your CRM today?
• What are the current pain points in your processes?
• Are there routine tasks that could be automated to free up team capacity?
• Are there new processes you’ve avoided due to resource constraints?

Examples of use cases include automating FAQ responses for service teams, generating campaign briefs for marketing, or assisting sales reps with lead prioritization and moving deals faster.

Then for each use case, think about what would be needed to transition to an agent:

• What job should they do?
• What actions will they need to take?
• What actions should they NOT take?  (This is just as if not more important to make sure you have defined the lane where an agent should operate within that use case)

Your responses to those questions are going to help you to understand the level of effort involved in use case. This in turn is going to help you to prioritize based on the level of effort and potential value

Tip 2: Define Success Metrics

To gauge the success of your Agentforce implementation, establish clear goals and KPIs. 

Questions you can ask:

• What does success mean? How will we know we have addressed our problem? 
• What metrics are we tracking today that we want to see improvement on?
• Are there additional metrics that will let us know we are seeing success?

For example:

• Reducing average case handling time by 20%
• Improving lead response times
• Increasing campaign ROI by automating content creation

Ensure you have baseline data for comparison and that the necessary measurement tools are in place to help you track success.

Tip 3: Assess Your Data

Your AI agents are only as good as the data they access. For the use cases identified, evaluate your data readiness:

• What data do need? 
• Where is it located? Is it in your CRM, Data Cloud, or other external systems?
• Is it accessible from your CRM? If it’s stored in an external system, do you have APIs in place to get that information?
• Is the data clean, accurate, and up-to-date?
  • Follow this blog post for tips on how to keep your imported Pardot prospect data clean.
• Do you have a single view of the customer across systems?
  • Get tips for creating that single source of truth in this blog post that covers marketing asset reporting, and this blog post that covers customer data platform reporting.
• Lastly, are knowledge bases and metadata structured for easy access? 
  • Agents need knowledge to inform how they will operate and answer questions. This is all of the background configurations your agents actions will rely on — flows, prompts, and Apex for example — they need to also be clearly identifiable and accessible. 
  • When you add actions to your topics, it uses the descriptions to help fuel the instructions. The naming conventions of your resources will also make it easier to determine what the inputs and outputs need to be.

Data and metadata are the backbone of AI performance, so this is an important area to pay attention to.

Tip 4: Start Small

After completing the previous steps, you may have more than one great use case to start with. Here’s where you ask yourself: What are the quick wins that we can get started on that can move the needle and that we can expand on as we mature?

It’s really easy to get caught up on how this can solve ALL the things. There are many challenges to starting a complex process all at once. If a lot of effort is required to get the data in place or to get the actions set up, it will be more difficult to roll out, not to mention making it potentially disruptive and prone to issues 

Avoid the temptation to tackle complex processes right away. Instead, focus on a simple, high-impact use case to pilot Agentforce. There are many out-of-the-box topics and actions that make getting started easier. For example, automating a single FAQ response or generating summaries for sales reps.

Starting small helps build confidence, momentum, and organizational buy-in, and it also reduces the risk of missteps.

Tip 5: Nail Down Clear Instructions

When designing agents, clarity is key. Use the Agent Builder to create and test well-defined topics and actions:

• Topics – Include precise instructions for classifying user requests, setting guardrails, and outlining scope.
• Actions – Clearly define what the agent should do, including required inputs and expected outputs.

Salesforce Agentforce Topic Instruction Best Practices

Instructions are the foundation for grounding how agents perform. They set the guardrails for how the agent should behave and give the agent the context it needs to do its job. 

Here are a few best practices for writing Agentforce topic instructions:

• Start simple
  • Start with the main use case first to ensure the agent is performing as expected. Then, add in more detail to address edge cases. Be sure to test existing instructions for any conflict. You don’t want to confuse the agent! 
• Use plain language
  • Use concise natural language to describe what your action does. Keep it to 1-3 sentences, and it can include the goal of the action, any use cases, and the objects or records it uses or modifies. 
  • In general, the more relevant detail you include in your instructions, the easier it is for the agent to differentiate between actions. Also, be sure to vary the words you use. For example, use a mix of “Get,” “Find,” “Retrieve,” or “Identify” for actions that will query records.
• Avoid industry or company jargon
  • Write like you are instructing someone who doesn’t know your business. Even terms like ‘qualified lead’ could mean something different from one organization to another. Give context where necessary, and reference clear criteria using the data it will have access to. 
  • For example, instead of vague terms like “qualify lead,” specify conditions such as “lead status equals MQL.” 
  • The agent isn’t not going to know your business processes either, so be explicit about the sequence of instructions or any conditions a conversation must meet for an agent to apply an action.
• Think of all the paths
  • You want to go through every possible permutation to determine the actions required. For example: a customer reaches out because they didn’t receive their order. 
  • First think about the order status ( Order Shipped, Delayed, Not Found, Processing). If the status is Shipped then there could be different tracking statuses (In Transit or Delivered for example). If the order is showing as Delivered, was it delivered to the customer’s correct address? Was it stolen? …and so on.
• Remember the Guardrails
  • Keep the Agent in its lane by providing clear instructions on what the agent should not do to prevent unwanted responses. In cases where the agent is customer-facing, be sure to also give clear direction on when an interaction should be routed to a human.

Test these instructions thoroughly in the Agent Builder’s testing environment to ensure your agents behave as expected.

Ready to Explore Agentforce?

Agentforce offers an exciting opportunity to enhance productivity and streamline operations. By identifying the right use cases, preparing your data, and starting with manageable projects, you can set your organization up for success.

Want to learn more? Check out Salesforce’s Agentforce Trailhead and virtual workshops to get hands-on experience. Need expert guidance? Contact the Sercante team for an Agentforce readiness assessment.