Empower Your Service Reps With Customized Prompt Templates for Service Replies

As Salesforce Admins, you play a key role in delivering a smooth experience for internal teams at your company by customizing and maintaining the tools that make their day-to-day easier. One powerful tool you can leverage is Service Replies, which uses generative artificial intelligence (AI) to assist your service reps by suggesting responses directly within the service console as they chat with customers. 

Users were initially limited to responses produced by out-of-the-box templates, which often felt generic and disconnected. Now, with Service Replies integrated into Prompt Builder, you can take control of how responses are shaped—customizing them to reflect your brand’s voice and ensuring reps can communicate in a way that feels natural and personalized.

Personalize responses for your team with Prompt Builder

Admins now have the ability to modify the prompt templates used for Service Replies directly in Prompt Builder. This helps you reflect your organization’s tone, values, and style, creating consistency and keeping the process efficient.

You can modify both the Contextual and Grounded templates to reflect your brand’s identity.

• Contextual: Based on the ongoing conversation
• Grounded: Drawing from your company’s knowledge base

Here’s a quick overview of how to get started.

1. Open Setup and search for “Prompt Builder”.
2. Select a Contextual or Grounded Service Replies template depending on your org’s needs.
3. Use the Prompt Template Workspace to modify the templates. Click Save As, and then select New Version.

Once you save your changes, you have the freedom to modify the prompt template. You can even test it with your own input and create multiple versions for different use cases. The brand voice in the standard prompt template is defined as “concise and friendly,” but you can modify this instruction to whatever brand voice you choose.

If you’re unsure of your company’s desired tone and messaging, refer to your company’s style guide and engage with the marketing and communications teams so that your prompts align seamlessly with your brand.

For example, a playful brand might instruct the large language model (LLM) to use emojis and friendly language, while a more professional brand might emphasize concise and formal responses. One good practice is to list a few examples of good content for the LLM when crafting new instructions. Here’s an example for a brand that values playfulness in its support interactions.

Respond in a friendly and encouraging tone, using clear and simple language. Imagine you are a supportive friend who is guiding the user through their journey. Avoid technical jargon and maintain a positive and playful attitude, even when addressing problems. Use emojis and exclamation points to reinforce the enthusiastic and approachable tone.

For example: “Oh no! It sounds like you’re having trouble with the audio. Don’t worry, we’re here to help! Let’s try a few things to get those sounds working again. First, could you try…”

By tailoring prompt templates, you ensure service reps deliver responses that feel authentic and personalized—without having to waste time rewriting or adjusting generic replies. For admins, this means driving efficiency while empowering your users.

Enhance Service Replies with more data

The out-of-the-box Service Replies templates have no idea who you or your customer are! Fortunately, you can bring that information into the conversation by including key customer data in the templates and underlying instructions to give service reps more context. You can do this through merge fields, as well as by using the resource picker in the Prompt Template Workspace to bring data in through flows and Apex classes.

For example, you can automatically include your rep’s name and company information in the greeting or pull in key customer details like their first name and case history. This makes the conversation more personal, efficient, and effective. Access these fields from the Current User object in the resource picker:

Your name is {!$User.FirstName} {!$User.LastName}. You work in the Customer Success organization at {!$User.CompanyName}. Always introduce yourself with your full name, {!$User.FirstName} {!$User.LastName}, and specify that you are part of the Customer Success team at {!$User.CompanyName}.

Because the Conversation object (either MessagingSession or LiveChatTranscript) is used as an input to the prompt template, you have access to all of its related objects like Account, Contact, and Case. Let’s add some information about your customer using the fields from Conversation object > Contact in the resource picker. You can add specific fields one at a time or make use of the Record Snapshot, which provides an overview of the record. For example:

Information about the customer is here:
{!$RecordSnapshot:conversation_object.EndUserContact.Snapshot}.
When appropriate, address them by their first name, {!$Input:conversation_object.EndUserContact.FirstName}.

It’s important to note that when using merge fields in your prompt templates, you should aim to avoid null values. It’s easy to guardrail against them by providing default values in the prompt. Combining these two small modifications with the brand voice instructions we added above, we’ve transformed our opening message to Ryan’s inquiry about a broken solar panel.

Before: “Could you please provide more details about the issue you’re experiencing with your solar panel?”

After: “Hi Ryan, I’m John from Ursa Major Solar. I’m sorry to hear that your solar panel isn’t working. Let’s figure this out together! Can you tell me a bit more about the issue? For example, is there any error message or unusual behavior you’re noticing with the panel?”

Use best practices for prompt engineering

Admins will continue to play a crucial role in optimizing AI-driven interactions. When refining your prompts, keep in mind that clear instructions are essential for generating high-quality responses. Here are a few tips to ensure your prompts deliver consistent and accurate replies.

• As your prompt templates get longer, consider creating sections through ### or “”” when adding new content.
• You can also control the length of the output by adding an instruction to make the response “within ___ characters”.
• Use “must” instead of “should”.
• Use “strictly” for strong commands.
• Format guardrails as “you” + “must” + verb. 
• Use “always” when you want consistency.
• Use strong verbs like “don’t” and “stop”. 
• Use “never” for absolute guardrails.
• Avoid using words like  “indirectly”, “subtly”, or “convey”—LLMs will usually ignore them.

Be sure to apply any basic instructions to both the Contextual and Grounded prompt templates to ensure consistency across responses. Remember, prompt engineering is an iterative process, so regular testing and adjustments will help you fine-tune how the AI supports your service reps over time.

Drive impact with personalization

As an admin, you’re in the driver’s seat, accelerating your service reps toward success. Incorporating these customizations can save time for service teams while delivering a more consistent and personalized experience across interactions. It’s up to you to work closely with your service team and fine-tune custom prompt templates that fit your unique business. You’ve got the power to turn this into a game-changing feature your users will love. For additional tips on configuring Service Replies for your org, take the Einstein Service Replies module on Trailhead. 

Resources

• Salesforce 360 Blog: Give Your Reps an AI-Powered Boost with Service Replies
• Salesforce Help: Einstein Service Replies
• Trailhead: Einstein Service Replies