Understanding Terret's AI Prompts for Deal Pulse

Where to go to update your Deal Pulse AI Prompts 

1. You'll need to be logged in as a sales operations user with access to the settings menu.  Click on the settings cog in the bottom left and then navigate to Deal Pulse under Company Settings. 
2.  There you'll see the list of Deal Pulses you've already configured.  Click the one you'd like to update.  Next to each Milestone (each field) on the right side, click the pencil icon. 
3.  In the edit milestone pop up page, the last text window titled 'Additional prompt instructions' is where you can add your AI prompt instructions.  Please reference the below rest of the guide for best practices and tips on how to update your prompts for the Deal Pulse milestones

How Default Prompts Work

Terret's AI comes with a sophisticated default prompting system that handles most common sales scenarios automatically. Here's what's built-in:

Base Agent Configuration

Every AI interaction starts with these default settings:

• Role: The AI operates as a "Senior Sales Analyst"

The default context automatically includes:

• All CRM data from the opportunity/deal.
• Complete transcripts of calls linked to the opportunity/deal.
• Email conversations linked to the opportunity/deal.
• Calendar events linked to the opportunity/deal.
• Current date for time-sensitive analysis.

Built-in Intelligence

Without any customization, the AI will:

• Focus on facts from the available data
• Handle relative dates correctly
• Update previous answers with new information
• Format outputs to match your CRM field types (boolean, picklist, date, text)
• Remove unnecessary IDs and timestamps
• Provide confidence scores and source attribution

How Prompt Customization Works

Custom prompts add a layer on top of the default system, allowing you to tailor the AI's behavior for your specific needs.

The Layering System

Think of prompts as layers:

1. Base Layer: Default agent behavior (always active)
2. Your Custom Layer: Additional instructions you provide
3. Output Layer: Automatic formatting based on CRM fields

Your custom prompts become "additional instructions" that enhance, not replace, the base behavior.  Please note that we have a size limit on custom prompts set to 3K characters. 

What You Should NOT Include

Since these are already handled by the default system:

• ❌ "Review the call transcripts and emails" - Already automatic
• ❌ "Return true or false" - Output format is auto-detected
• ❌ "You are a [different role]" - Don't redefine the agent's role
• ❌ Data source instructions - The AI already gathers all available data
• ❌ Basic accuracy instructions - Built into the base behavior

What You SHOULD Include

Focus your custom prompts on:

• ✅ Specific criteria unique to your evaluation
• ✅ What to look for in the data
• ✅ Custom definitions of success
• ✅ Industry-specific considerations
• ✅ Override instructions (if you need different default behavior)

Best Practices for Custom Prompts

1. Frame as Questions

Instead of complex instructions, use simple questions:

• Good: "Is the prospect actively evaluating our solution?"
• Less effective: Complex multi-paragraph instructions

2. Add Specific Criteria

Be clear about what constitutes a positive signal:

Focus on determining whether the prospect is actively considering OpenAI based on:

- Direct mentions of evaluation or POC

- Budget discussions related to AI tools

- Timeline for implementation decisions

3. Keep Instructions Focused

Since the AI already handles the basics, your prompts should be laser-focused on what's unique:

Look for these specific buying signals:

- Mention of board approval needed

- Discussion of implementation timeline

- Questions about integration with existing system

3. Specify Output Format

You can format the output with simple instructions

Example: Generated output should start each next step on a new line prefixed with the date by when the step needs to be executed.  

If you would like to append the AI recommendation to the beginning of the text field and keep the rest of the text that's already there you can use the following instruction.  This is helpful for appending fields such as next steps with the latest update.

Append any new next step to the existing <user_answer>, but don't modify the content of the current <user_answer>

Competitive Intelligence

Identify if [Competitor Name] is being evaluated by looking for:

- Direct mentions of the competitor

- Discussion of their features or pricing

- Comparison requests between solutions

Deal Qualification

Determine if this opportunity meets our ICP by checking for:

- Company size mentioned (needs to be 500+ employees)

- Specific use cases: data analytics, reporting automation

- Technical requirements discussions

Custom Scoring

Evaluate the strength of the champion relationship:

- Strong: Multiple 1-on-1 calls, introduces us to other stakeholders

- Medium: Regular communication but no introductions yet

- Weak: Only responds when we reach out

Understanding Prompt Interaction

Your custom prompts work WITH the default system:

1. The AI starts with its base understanding
2. Your custom instructions are added as additional context
3. The combined instructions guide the analysis
4. Output is automatically formatted for your CRM

Think of it as giving the AI a specific lens through which to view the already-gathered data, rather than telling it how to do its job from scratch.

Testing

To test your custom prompts, in the milestone configuration where you enter your custom prompts, theres an option in the bottom left titled 'Test Prompt'.  After clicking that you'll be able to search for any opportunity to run a test for that prompt against.  We recommend ideally selecting an opp that you believe would  have the call recordings and context for the prompt you're trying to answer.

Troubleshooting

If Results Aren't What You Expected

1. Remove redundant instructions - If you're telling the AI to do something it already does
2. Check for conflicts - Ensure you're not contradicting the base behavior
3. Simplify - Often shorter, clearer prompts work better
4. Test incrementally - Add one instruction at a time to see its effect

If the AI Seems Confused

This often happens when custom prompts try to redefine fundamentals:

• Remove any role redefinition
• Remove output format instructions
• Focus only on what to look for, not how to look

Quick Reference

✅ DO

• Write clear, specific criteria
• Focus on what's unique to your process
• Use simple, question-based prompts
• Add industry-specific context

❌ DON'T

• Redefine the AI's role
• Specify data sources to review
• Include output formatting instructions
• Add basic instructions about accuracy

Remember: The default AI is already quite sophisticated. Your customizations should enhance its capabilities for your specific use case, not rebuild them from scratch.