Optimizing Rainmaker Prompts: Best Practices for AI Message Handling

Rainmaker Prompts define how ACE’s Rainmaker AI interprets and responds to inbound text, email, and Virtual Agent messages. Well-structured prompts ensure that messages are correctly evaluated and routed to the appropriate workflows.

In ACE, Rainmaker Prompts can be configured and maintained in two areas, depending on the scope of automation:

1. Rainmaker Condition Flow Actions – For prompts that evaluate specific inbound messages and determine whether a workflow follows a true or false path.

2. Rainmaker Preferences – For global or context-aware prompts uploaded through a custom text file (rainmaker_custom_prompt.txt) that influence Rainmaker’s responses across all communication channels.

This article explains how to structure, test, and refine prompts—whether managed in a Condition Flow Action or through Rainmaker Preferences—to improve message accuracy, reduce false evaluations, and enhance overall automation performance.

Structuring Effective Rainmaker Prompts

Each Rainmaker Prompt should clearly define both the language that should trigger a condition and the language that should not. This balance improves accuracy and reduces false positives across all Rainmaker evaluations—whether configured in a Condition Flow Action or in Rainmaker Preferences.

• A Positive Prompt – Specifies the language that should trigger the condition.

• A Negative Prompt – Defines messages that should be ignored to reduce false positives.

Example: Evaluating for a Wrong Number

• Positive Prompt: "Look for phrases like ‘wrong number,’ ‘wrong person,’ or ‘that’s not me.’"

• Negative Prompt: "Ignore generic inquiries like ‘Who is this?’ or ‘What company is this?’"

This format helps Rainmaker identify wrong number messages and initiate the appropriate workflow while using negative prompting to ignore generic inquiries, preventing them from being added to the wrong workflow.

Example: Using a Rainmaker Prompt File to Automate Responses and Actions

Prompts stored in the rainmaker_custom_prompt.txt file (uploaded through Setup → Rainmaker → Rainmaker Preferences) define how Rainmaker responds across all communication channels. These prompts can guide Rainmaker’s replies, enforce policy boundaries, or even initiate workflows through Action Paths.

When discussing credit reporting or credit bureaus and the customer is discussing account payment, payoff, or resolution, 
provide the following information as stated here:
"Once your account is paid in full, we will notify the credit bureaus that your account is paid and request its removal 
from your credit report. Please note that the credit bureaus may take 30 to 60 days to update their records."

If the consumer asks to "discuss", "connect", or "talk" about the account, 
respond with ${RAINMAKER_FUNCTION::ActionPath({ActionPathID:12345})} 
and "We'll call you as soon as possible."

Example: Dispute Condition

• Positive Prompt: "Detect phrases like ‘I don’t owe this’ or ‘I dispute this debt.’"

• Negative Prompt: "Ignore messages like ‘I need more details’ or ‘I have a question about my account.’"

By distinguishing true dispute messages from general questions, Rainmaker prevents irrelevant inquiries from being mistakenly added to the dispute workflow, ensuring that only genuine disputes are escalated for appropriate handling.

Managing Rainmaker Prompts

Rainmaker Prompts are managed differently depending on whether they are tied to a specific workflow or applied globally across communication channels.

1. Rainmaker Condition Flow Action

Prompts defined in Rainmaker Condition Flow Actions evaluate inbound text, email, or Virtual Agent messages.

They determine whether a workflow should follow the “true” or “false” path based on message content.

For configuration details, refer to [Configuring Rainmaker Condition Flow Actions for Inbound Messages].

2. Rainmaker Preferences

Prompts configured in Setup → Rainmaker → Rainmaker Preferences provide a broader layer of control.

From here, administrators can:

• Upload a custom text file (rainmaker_custom_prompt.txt) containing AI instructions and triggers.

• Include references to Action Paths to initiate automated workflows when certain phrases or conditions are detected.

• Use variables (e.g., ${customer.companyName}, ${debt.currentBalance}) to personalize responses.

These prompts enhance Rainmaker’s understanding of context across all communication types—chat, text, email, and Virtual Agent interactions.

Iterating & Refining Prompts Over Time

AI-driven automation requires continuous improvement to maintain effectiveness. Agencies should test, refine, and adjust prompts based on real-world results.

1. Start with a Basic Prompt – Define clear detection criteria.

2. Observe Real-World Results – Identify undesirable results.

3. Refine Prompt Language – Adjust based on observed outcomes.

4. Re-Test and Iterate – Conduct real-time testing using manual checks and sample accounts to validate results. Continue refining until Rainmaker consistently evaluates messages as expected.

How to Test Rainmaker Prompts for Wrong Number

1. Use a Sample Account

Select a test account for controlled evaluation.

2. Create an Action Path

1. Navigate to Setup → Workflow → Action Paths, New.

2. Right-click on the newly created Action Path Container and select Edit.

3. Enter the following details:

   1. Label: Rainmaker_Condition_Flow Test for Wrong Number

   2. Description: Processes inbound text/email conditions for Rainmaker AI

   3. Class Name: (Auto-fills by default)

   4. Bundle Action: False

   5. Email Manager Action: True

   6. Text Manager Action: True

4. Click Save.

5. Add Rainmaker Condition

   1. Right-click on Rainmaker_Condition_Flow Test for Wrong Number and select New.

   2. Enter the following details:

      1. Class Name: Rainmaker Condition

      2. Prompt:

         1. Positive Trigger: Detect phrases such as "wrong number," "wrong person," or "that’s not me."

         2. Negative Prompt: Ignore generic inquiries like "Who is this?" or "What company is this?"

   3. Label: Rainmaker Condition = Wrong Number

   4. Click Save.

6. Add True Actions for Wrong Number

   1. If the condition is met (True), add the following action:

   2. Account Flag Add: Rainmaker Evaluation Wrong Number

3. Test the Prompt

Send inbound messages using various phrases to evaluate Rainmaker’s effectiveness.

Additional Testing Considerations

• Use different wording variations to validate prompt results.

• Conduct real-time and manual testing with sample accounts.

• Refine prompt language and re-test as needed.

• Once satisfied, add additional flow actions to further automate workflows and improve efficiency.

Ideas for Rainmaker Prompts & Actions

Below is a table with suggested prompts, example inbound messages, and recommended actions for different scenarios. These are starting points, and agencies should refine them over time to improve results based on real-world interactions.

Combining Prompts with Flow Actions

Once Rainmaker tests are meeting your expectations, prompts can be combined with automated actions to streamline workflows.

Final Takeaways

• Use clear, simple language for prompts.

• Include both positive and negative prompts to refine AI detection.

• Test and iterate prompts based on real interactions.

• Start with flagging and manual review before automating other actions.

• Monitor AI performance and adjust prompts as needed.

By following this approach, you can ensure that Rainmaker consistently interprets inbound messages, minimizes manual intervention, and supports effective workflow automation.