Introduction

Setting up a great inbound voice AI agent starts with a few essentials: clear goals, a solid knowledge base, and the right tools to help your agent confidently handle every call. In this guide, you’ll learn how to use Oration to create an inbound support agent that consistently delivers helpful customer experiences and makes your team more efficient.

Whether you’re aiming to answer common questions, share real-time order updates, or automate simple troubleshooting, we’ll walk you through each step to design, launch, and continuously improve your inbound agent—no matter your starting point.

What You’ll Build

• Comprehensive Knowledge Base: Equip your agent with the information it needs to solve problems and guide customers.
• Custom Vocabulary: Train your agent with your company’s unique terms for smoother, more natural conversations.
• Real-Time Integrations: Connect to systems like order management for up-to-date answers.
• Performance Scorecards: Set up workflows to review and rate each conversation, so you know what’s working.
• Continuous Improvement Processes: Use post-call analytics to spot trends and strengthen your agent over time.

Let’s dive in!

Step 1: Create Your Inbound Support Agent

Let’s lay the groundwork for your support agent—clear and straightforward, no jargon or fluff.

Get Started in the Oration Dashboard

1. Log in to your Oration dashboard.
2. Click Agents in the left sidebar.
3. Select Create New Agent.

Set Up the Basics

• Give your agent a clear, descriptive name—like “Customer Support Agent” or “Inbound Support Agent.”
• For Initiation Type, choose AI Speaks First – Fixed.
• Add a friendly initial message, such as:
  “Hello, this is [Agent Name] from [Company Name]. How can I help you today?”

Step 2: Write a System Prompt that Guides Your Agent

The system prompt gives your agent clarity and confidence. It shapes how your agent communicates with customers and handles tough situations.

Prompt Structure: Keep It Customer-Centric

Here’s an approachable prompt template you can adapt for your inbound support agents:

You are Alex, a friendly and knowledgeable customer support specialist at [Your Company Name]. Your role is to help customers with their questions about orders, products, returns, and account issues.
 
## Your Personality
- Warm and approachable, like a helpful friend
- Patient and understanding, especially when customers are frustrated
- Clear and concise in explanations
- Professional but conversational
 
## Communication Guidelines
- Always greet customers warmly and ask how you can help
- Listen carefully to understand the full situation before responding
- Use simple language—avoid jargon unless the customer uses it first
- Acknowledge emotions: "I understand this is frustrating" or "I'm happy to help with that"
- Confirm understanding: "Just to make sure I've got this right..."
 
## What You Can Help With
- Order status and tracking information
- Product information and availability
- Return and exchange processes
- Account questions and updates
- General company policies
 
## What You Cannot Do
- Process refunds directly (explain the process instead)
- Make exceptions to company policies
- Access payment card details
- Make promises about specific delivery dates
 
## Handling Different Situations
 
**When you don't know the answer:**
"That's a great question. Let me look into our knowledge base to get you the most accurate information."
 
**When you need to transfer:**
"I want to make sure you get the best help possible. Let me connect you with a specialist who can handle this specific situation."
 
**When a customer is upset:**
Start with empathy, acknowledge their concern, then focus on solutions: "I completely understand why this is frustrating. Let's get this sorted out for you right away."
 
## Important Rules
- Never make up information—always check your knowledge base
- If uncertain, admit it and offer to find the answer or transfer to a human
- Always end calls by asking if there's anything else you can help with
- Be concise—don't overwhelm customers with too much information at once

Make It Yours

Personalize your agent’s system prompt by updating:

• Your company name
• The services you provide
• The top questions or issues your support team encounters
• Your unique brand voice and tone

Make sure every detail fits your business and how you want to sound to your customers.

Step 3: Fine-Tune Your Agent’s Settings

You can shape your agent’s behavior for a great customer experience. Head to the "Configurations" and "Advanced Settings" tabs to adjust key options.

Essential Settings

LLM (Large Language Model):

• Pick an LLM provider that aligns with your needs: OpenAI, Azure, or Oration OpenAI (our optimized version).
• Set the temperature to control how creative your agent’s replies sound—0.3 is usually a balanced starting point.

TTS (Text-to-Speech):

• Select a voice that feels friendly and fits your brand personality.
• Adjust the language, accent, and speech speed for clarity and accessibility.

STT (Speech-to-Text):

• Choose the STT provider (for example, OpenAI or a preferred vendor) that offers the accuracy, speed, and language support your customers need.
• Consider enhanced models if your audience includes people with strong accents or who use industry-specific vocabulary.

Conversation Settings

• Max Call Duration: 10–15 minutes is a good call length for most support scenarios.
• Silence Timeout: A pause of 5–7 seconds before gently following up works well.
• Interruption Word Threshold: Try a setting of 3 for a good balance between natural flow and responsiveness.

Advanced Options

• Turn on noise suppression to help conversations sound crisp and professional.
• Want agents to feel more “human” and less robotic? Subtle background noise—like soft office sounds or a low call center murmur—can make conversations feel warmer and more like a real support call.

Step 4: Post-Call Analysis (PCA): Learning and Improving with Every Conversation

Post-call analysis gives you clear, actionable takeaways from every call—so you can understand how things are going, spot trends, and keep making your support better.

How to Enable Post-Call Analysis

1. Go to your agent’s page and click the "Post Call Analysis" tab.
2. Switch the Enable toggle to “on.”
3. Add clear instructions (a prompt) that tell the AI what you want to learn from each call. Not sure where to start? Click "Generate Prompt" for a pre-built suggestion, or write your own to focus on what matters most.
4. Set up a schema: define tag names (like category, sentiment, etc.) and describe what each one will track.

Once you save your PCA settings, the AI will automatically analyze each new conversation. You’ll find results in the History tab on the left navigation—just open a call and select the Analysis section.

Reviewing Results

• Each analyzed call shows up with structured tags and summaries in the Analysis panel.
• To get PCA insights on older conversations, just click "Generate Analysis" on those calls.

Getting the Most from PCA

• Keep tweaking your prompt and schema until you’re seeing the details and insights that matter to you—common topics, sentiment, escalation triggers, and more.
• Use PCA output to spot patterns and needs, refine your knowledge base, and coach your team or tune your AI agent’s tone.
• Make it a habit to regularly review PCA findings, so your support keeps getting smarter, faster, and more helpful for your customers.

Step 5: Build Your Knowledge Base

A great knowledge base is the heart of your agent—it’s where your agent looks for accurate, up-to-date answers for your customers. The stronger and more relevant your resources, the better experience you’ll deliver.

Gather Your Content

Start by collecting your most helpful resources:

• FAQs from your website
• Support documentation
• Product information
• Company policies (returns, shipping, etc.)
• Common troubleshooting guides

Double-check that everything you upload is accurate, current, and matches what you’ve written in your system prompt. Consistency between your knowledge base and your prompt helps your agent deliver clear, confident answers every time.

Uploading in Oration

1. In the left navigation bar, go to "Knowledge Base".
2. Click "Create Knowledge Base" and choose a clear, descriptive name (for example: "Customer Support Knowledge").
3. Use "Add Documents" to upload your content.
4. Save. Repeat as needed for each topic or area.
5. When you’re ready, go back to the Agent Details page. Scroll down to add the knowledge base you created to your agent.

Step 6: Define Your Custom Terms

Clear, consistent language helps your agent sound natural and on-brand. Creating custom terms is an easy way to make sure product names, industry lingo, and preferred phrases always come across correctly.

What Are Custom Terms?

Custom terms allow you to:

• Set the proper spelling and pronunciation for product names
• Standardize company-specific jargon
• Clarify acronyms or abbreviations for your audience
• Reinforce brand names
• Teach your agent technical or nuanced terms

How to Create Terms

1. In the left navigation, select "Terms", then click "New Term".
2. Enter the word or phrase you want to adjust, along with the replacement you prefer.
3. If you want the replacement used every time—no exceptions—check "Strict replace."
4. When you’re finished, update the term status from "Pending" to "Approved".

Examples for Brand Voice

Here are a few sample terms you might set up for a friendly, professional tone:

Term: Thanks
Replacement: "Thank you"
Usage: Always use "thank you" instead of "thanks" to maintain a professional tone
 
---
 
Term: Yeah
Replacement: "Yes" or "Absolutely"
Usage: Avoid casual "yeah"—use "yes" or "absolutely" instead
 
---
 
Term: Problem
Replacement: "Challenge"
Usage: Reframe negatively. Say "Let's resolve this situation" instead of "Let's fix this problem"

Step 7: Adding Real-Time Tools

Tools enable your agent to fetch up-to-date information during a conversation—like checking an order status or retrieving account details. Let’s walk through how to set up a tool and define its actions so your agent can lend a helping hand, right when it’s needed.

Getting Started with Tools

1. In your Oration dashboard, select "Tools" from the left-hand navigation.
2. Click "New Tool" and give it a name that’s easy to recognize, such as "Inbound Support Tool."
3. You’ll be prompted to create actions within this tool.

Building an Action

Every tool can include one or more actions. For instance, your support tool might have actions like "Get Order Status," "Cancel Order," or "Check Product Availability." Here’s how to add an action:

Click "Create Action" to open the setup form, which includes fields like Description, URL Template, and Schema. You’ll also see optional sections (Headers, Body) if the integration requires them.

When completing each field, use clear and meaningful descriptions. Pull the necessary details—such as required parameters, dynamic variables, authentication, and request formats—directly from the API documentation for the service you’re connecting.

A Few Friendly Pointers:

• Use descriptive action names that clearly signal their purpose.
• Mark parameters as required when the API needs them—this helps your agent gather the right information from callers.
• Write descriptions that make it obvious to the agent (and your team) how each action should be used.

Connecting the Tool to Your Agent

After setting up your tool and its actions:

1. Return to your Agent Details page.
2. Scroll to the "Tools" section.
3. Click "Add Tool," then select the tool you just created.
4. Save your agent.

That’s it—your agent now has new abilities to pull information in real time.

Useful Actions for Support Agents

Depending on your support flow, you might want to set up actions like:

• Order Status Lookup: Track shipments and confirm delivery details.
• Cancel Order: Help customers cancel orders quickly.
• Failed Payment Inquiry: Assist with payment issues or errors.
• Product Stock Check: Verify inventory status for requested items.
• Return Initiation: Start the return process on a customer’s behalf.

Prompts and Best Practices

In your System Prompt, be specific about when and how the agent should use tools. For example:

If a customer asks about an order status, use the Get Order Status tool. 
Ask for their order number first. If they don’t know it, request their email address 
and the approximate date of the order.
 
If the tool can't find an order or returns an error, say: 
"I'm having trouble finding that order. Let me connect you with a specialist who can help."

Handling Errors Gracefully

When a tool fails or returns incomplete data, keep the conversation supportive and proactive. It’s a good idea to teach your agent to say something like:
"I’m having trouble accessing that information right now. Let me connect you with someone who can help."

That way, your agent always keeps the customer informed—and reassured that the next step is already in motion.

Step 8: Testing Your Agent

Before you send your AI agent out into the world, give it a thorough test drive. Real customers rely on your agent for help, so it’s worth taking the time to iron out any wrinkles and make sure everything feels natural, helpful, and reliable.

How to Test Your Agent

1. Go to Preview in the top right corner of your agent dashboard.
2. Try out different modes—chat, web call, and phone call—to interact with your agent as your customers would.
3. Simulate a range of scenarios. Check common requests, rare questions, and unexpected inputs to see how your agent responds.

Agent Testing Checklist

A little structure goes a long way. Use this checklist to make your testing process clear and consistent:

Basic Functionality

• The agent greets promptly and appropriately.
• Voice sounds clear and natural.
• Uses accurate terms from your set terminology.
• Handles silence well (doesn’t speak over people).

Knowledge Base

• Responds accurately to FAQs.
• Gives customers information from your knowledge base.
• Says “I don’t know” (or similar) when something’s genuinely unknown.
• Never invents answers.

Tool Use

• Successfully accesses and uses support tools (like order lookup).
• Handles unexpected or invalid inputs gracefully.
• Explains tool results in a natural, conversational way.
• Has a fallback response when a tool fails.

Conversation Flow

• Asks follow-up questions when something’s unclear.
• Remembers context and doesn’t lose track mid-call.
• Transitions smoothly between topics.
• Wraps up conversations in a helpful, professional way.

Review and Iterate

• Listen back to call recordings under Call History.
• Note any rough spots, misunderstandings, or awkward phrasing.
• Make edits to your system prompt or knowledge base to improve the customer experience.
• Re-test after making changes until you're happy with the flow.

Step 9: Setting Up QA for Your Agent

Quality assurance (QA) lets you review interactions and ensure your AI agent consistently delivers support that meets your standards—every time.

Creating a Scorecard

Scorecards give you a simple, objective way to measure how your agent is doing in real customer conversations.

To create a scorecard in Oration:

1. Head to Scorecards in the left sidebar, then click Create Scorecard.
2. Choose a clear name—something like Inbound Support Quality Scorecard works well.
3. Add a description, such as "Evaluates agent responses for service quality, brand consistency, and successful issue resolution."
4. Set your Passing Points (for example: on a 50-point scorecard, require a minimum of 40 to pass).

Writing a Scorecard Prompt

Let the AI know what to look for when scoring calls. For example:

You’re an expert QA agent. Evaluate the conversation between a user and an AI agent using the criteria provided. If a criterion doesn’t apply, give full points for that item.

Defining Criteria and Questions

For each criterion, add a question with these details:

• Question: What are you evaluating? (Example: “Did the agent greet the customer professionally and introduce themselves?”)
• Description: Explain what a good answer looks like.
• Max Points: How many points is this worth?
• Evaluation Type: Choose "Score" for graded answers or "Pass/Fail" for yes/no.
• Fatal Question: If missing this is deal-breaking, mark the question as fatal (for auto-fail).

Sample Criteria:

Proper Greeting
   Did the agent greet the customer professionally and introduce themselves?  
   Points: 10 (Pass/Fail)

Scorecard Tips

• Early on, consider running QA on every call. As you gain confidence, you can switch to sampling (e.g. 50% at steady state; 10–25% for ongoing monitoring).
• After making your scorecard, assign it to your agent under the Quality Assurance tab and set your preferred review cadence.

Regular QA is your path to consistently excellent support. With just a bit of ongoing attention, you’ll keep your agent representing your brand the way you want—warm, accurate, and dependable.

Troubleshooting Common Issues

Even with careful setup, a few bumps in the road are normal. Here’s how to spot them and get your agent back on track.

If your agent gives inaccurate information

• Why this might happen: Some knowledge base entries might be unclear or give conflicting details.
• How to address it:

  • Review and clarify any confusing KB entries.
  • Make sure each topic has one clear, definitive answer.
  • Add simple examples to help illustrate the right response.
  • Test by asking sample questions, just like your customers would.

If your agent sounds robotic or stiff

• Why this might happen: The system prompt could be too rigid or formal.
• How to address it:

  • Add conversational examples to your prompts—the more natural, the better.
  • Include helpful filler phrases, like "Let me check that for you…"
  • Look for spots where the language is overly formal and adjust for a warmer, friendlier tone.
  • Try tweaking voice settings for a pace and energy that suits your brand.

If tools frequently fail or time out

• Why this might happen: There may be an API issue or a configuration hiccup.
• How to address it:

  • Check that the API endpoints are up and reachable.
  • Double-check authentication credentials.
  • Test the API outside of Oration to confirm it’s working as expected.
  • Update your prompts to add more helpful error handling.
  • Set timeout limits that balance speed and reliability.

If customers keep asking to speak to a human

• Why this might happen: The agent may not be meeting their needs yet.
• How to address it:

  • Read transcripts to understand where customers get stuck or frustrated.
  • Grow your knowledge base to cover any gaps.
  • Improve your tools so the agent can handle more requests smoothly.
  • Fine-tune the greeting to clearly set expectations.
  • Consider being up front: add a friendly, "I'm an AI assistant" introduction if needed.

Next Steps

You’re well on your way! Here’s how to keep building momentum:

1. Launch gradually: Start with a soft launch—route only a small portion of calls to your agent.
2. Monitor closely: Listen in on every call for the first week to catch opportunities to improve.
3. Iterate often: Make small, regular changes as you notice patterns.
4. Measure what matters: Keep an eye on stats like handled call volume, customer satisfaction, and savings.

Building a truly helpful AI agent is a journey, not a finish line. As you learn from your real callers and continue to refine, your agent will grow smarter, warmer, and even more valuable to your customers.