Customer Support Agent
In this tutorial, you’ll build a voice agent that handles inbound customer support calls. It can answer product questions from a knowledge base, walk customers through troubleshooting steps, process return requests, and log tickets to your CRM automatically.What You’ll Build
A phone agent that:- Greets callers and identifies their issue
- Answers product questions using your FAQ and product catalog
- Guides customers through step-by-step troubleshooting
- Processes return and refund requests
- Sends SMS confirmations for actions taken
- Logs every interaction as a support ticket in your CRM
Step 1: Create the Agent
- Go to your thinnestAI Dashboard
- Click Create Agent
- Name it:
Customer Support Agent - Set the description:
Handles inbound support calls for product questions, troubleshooting, and returns
Step 2: Choose the Model
Select Claude Sonnet as your model. Why Claude Sonnet? It follows complex instructions reliably, stays in character, and responds fast enough for natural voice conversations. It’s the best balance of quality and speed for support agents.Tip: Keep temperature low (0.2–0.4) for support agents. You want consistent, accurate answers — not creative ones.
Step 3: Add Knowledge
Upload two documents to give your agent the information it needs:FAQ Document
Create a document with your most common support questions and answers. Format it clearly:Product Catalog
Upload your product catalog or a summary document containing:- Product names and SKUs
- Key features and specifications
- Pricing information
- Common issues and their solutions
- Go to the Knowledge tab in your agent settings
- Click Add Knowledge Source
- Upload your FAQ document (PDF, DOCX, or TXT)
- Upload your product catalog
- Wait for indexing to complete (usually under 60 seconds)
Step 4: Assign Tools
Enable the following tools for your agent:Knowledge Search
This is enabled automatically when you add knowledge sources. It lets the agent search your FAQ and product catalog during calls.SMS
Enable the SMS tool so the agent can send confirmation texts to callers.- Go to the Tools tab
- Enable SMS
- Configure your sender number
CRM Webhook
Set up the CRM webhook to log support tickets automatically.- Enable CRM Webhook in the Tools tab
- Set the webhook URL to your CRM endpoint:
Step 5: Set the System Prompt
Copy and paste this entire prompt into the System Prompt field:Important: Replace[YOUR COMPANY NAME]and[AGENT NAME]with your actual company and agent name before saving.
Step 6: Assign a Phone Number
- Go to the Phone tab in your agent settings
- Select an available phone number or purchase a new one
- Set the call direction to Inbound
- Save
Step 7: Test Your Agent
Use the built-in test call feature or dial the assigned number from your phone. Run through these scenarios:Testing Checklist
| Scenario | What to Test | Expected Behavior |
|---|---|---|
| Product question | Ask “What’s the return policy?” | Agent searches knowledge base, gives accurate answer |
| Troubleshooting | Say “My device won’t turn on” | Agent asks clarifying questions, walks through steps one at a time |
| Return request | Say “I want to return my order” | Agent asks for order number, confirms eligibility, processes return, sends SMS |
| Unknown question | Ask something not in the knowledge base | Agent admits it doesn’t know, offers to escalate |
| Frustrated caller | Express frustration | Agent acknowledges emotion, stays calm, offers solutions |
| Multiple issues | Ask about two different things | Agent handles first issue, then asks “Is there anything else?” |
| Escalation request | Say “Can I talk to a real person?” | Agent offers to transfer, logs context |
What to Listen For
- Pacing: Does the agent pause naturally between sentences? Does it feel conversational?
- Clarity: Are answers concise and easy to follow over the phone?
- Accuracy: Does it pull the right information from your knowledge base?
- Confirmations: Does it confirm before taking actions?
- SMS delivery: Do confirmation texts arrive promptly?
Tips for Improvement
- Review call logs regularly. Look for questions the agent couldn’t answer and add them to your FAQ document.
- Add more knowledge over time. The more product and troubleshooting information you provide, the more issues the agent can resolve without escalation.
- Tune the temperature. If the agent sounds too robotic, bump temperature to 0.4. If it’s giving inconsistent answers, drop to 0.2.
- Set up post-call follow-up. Configure a follow-up SMS or email that goes out 24 hours after the call asking “Was your issue resolved?”
- Track resolution rate. Use your CRM webhook data to measure what percentage of calls the agent resolves vs. escalates. Aim for 70%+ first-call resolution.
- Add seasonal knowledge. Before holidays or product launches, update your knowledge base with relevant information (shipping cutoffs, new product details, sale terms).
Next: Appointment Booking Agent — Build an agent that schedules appointments and sends calendar invites.

