Skip to main content
Inbound Agents: Customer Support and Administrative Assistance

Create an AI Voice Agent for Customer Support

This guide explains how to implement an agent that handles customer service requests, answers frequently asked questions, and triages complex cases to the human team.

Written By Product

Last updated 9 months ago

When to use it

The customer support agent is ideal when you receive many repetitive calls: questions about schedules, prices, order status, standard procedures. It frees up your team for cases that really require attention.

Basic configuration

In Chat AI, describe the agent as follows:

"Create a customer support agent for [type of business]. It must answer frequently asked questions about [main topics], check order status, and transfer complex cases or dissatisfied customers to a human operator."

The AI will ask you to provide the knowledge base with answers to frequently asked questions.

Building the Knowledge Base

The agent is only as effective as the information you give it. Prepare:

Structured FAQs: the 20-30 most frequently asked questions with clear and concise answers

Operational information: hours, addresses, contacts, return policy, shipping times

Standard procedures: how to track an order, how to request a return, how to edit an account

You can upload documents to the AI Chat or enter the information directly into the conversation.

Manage order status

If you sell products, the agent must be able to check the status of orders. You have two options:

Direct integration: connect your e-commerce or management system via API with the agents' custom tool. The agent asks for the order number and retrieves the status in real time. Custom tools can be added by asking the agent to add a custom tool or from the "In call" step of the Advanced Builder. 

Manual fallback: if integration is not possible, the agent collects the data (order number, email) and promises a follow-up within X hours.

In Chat AI: "I want the agent to be able to check the order status. The customer provides the order number and the agent must say whether it is being prepared, shipped, or delivered."

Escalation to a human operator

Clearly define when the agent should transfer the call to a human:

Automatic triggers: angry customer, refund request above X euros, complex technical problem, explicit request to speak to an operator

How to handle the transfer: "I understand, this situation requires the intervention of one of my colleagues. I'll put you on hold and transfer you to an operator right away." Or, if no one is available: "My colleagues are busy at the moment. Can I have someone call you back within [time]?"

In the AI Chat, specify the escalation criteria: "If the customer requests a refund, is angry, or the problem is not in the FAQ, transfer to an operator."

N.B: This feature will soon be released in DeepAgent. 

Tone and handling difficult customers

Customer support requires empathy. Configure the agent with a patient and understanding tone.

For frustrated customers, the agent should:

  1. Acknowledge the problem: "I'm sorry for the inconvenience."

  2. Not argue: avoid blaming the customer

  3. Propose a solution or escalate quickly

In AI Chat: "If the customer is angry or frustrated, the agent should apologize for the inconvenience, show understanding, and propose a concrete solution or transfer to an operator."

Data to collect

For each support call:

Always: name, reason for call, outcome (resolved/escalated/callback)

If order: order number, specific problem

If complaint: details of the problem, what the customer expects

This data is used to improve service and identify recurring problems.

Metrics to monitor

  • First Call Resolution: % of issues resolved on the first call

  • Escalation rate: % of calls passed on to human operators

  • Average handling time: average duration of calls

  • Customer satisfaction: if possible, collect post-call feedback

An escalation rate above 30-40% indicates that the knowledge base is incomplete or the agent is not configured properly.

Common mistakes

Generic knowledge base: vague answers such as "check the website" frustrate customers. Give specific and actionable answers.

No way out: the customer must always be able to speak to a human. Never block this option.

Unrealistic promises: if the agent cannot resolve the issue, they should not make promises that the team cannot keep.

Ignoring sentiment: if the customer is clearly frustrated, insisting on FAQs will only make the situation worse. It is better to escalate immediately.

Estimated implementation time: 60-90 minutes (depending on the complexity of the knowledge base)