Scale 24/7 support

The customer agent uses Breeze, HubSpot’s AI to automate routine requests, respond to customers instantly, and route complex issues to your team. By syncing high-quality content, clear processes, and defined handoff rules, you can scale support without increasing headcount.

The steps below outline how support leaders, admins, and operations teams can set up, train, test, deploy, and optimize the customer agent to manage customer conversations. By following this approach, you’ll provide 24/7 support, reduce ticket volume, and improve response times while maintaining a high-quality customer experience.

For example, Sprocket Supply Co., an ecommerce company selling sustainable office supplies, uses the customer agent to answer common questions about order status, shipping, and returns. This allows their support team to focus on complex issues like billing disputes or delayed deliveries while still providing fast, consistent responses to every customer.

Understand how AI powers the customer agent

The customer agent uses AI to interpret customer questions and generate responses based on your synced content sources, such as knowledge base articles, website pages, and internal documentation.

It combines:

• Content sources to answer questions
• CRM data to personalize responses and complete actions
• Actions to perform tasks like checking order status
• Handoff rules to escalate complex or sensitive requests to a human

The quality of your setup directly impacts performance. Well-structured content and clear rules allow the agent to resolve more conversations independently.

Identify your support goal

Before setting up the customer agent, define what you want it to achieve. Identify common customer questions and evaluate how your team currently handles them. This will help you determine which requests can be automated and which should remain with a human.

For example, Sprocket Supply Co.’s goal is to reduce Where is my order tickets by 40% and provide instant answers outside of business hours.

Define and analyze your existing support process

• Define ownership and governance: assign a clear owner responsible for setting up, monitoring, and continuously improving the customer agent. This ensures accountability for performance, content quality, and ongoing optimization.
• Review conversation/ticket management: identify where your team manages conversations, such as help desk, the conversations inbox, or an external platform. Review your existing setup, including user access and availability. Escalated conversations will be routed back to your team and managed via help desk or the conversations inbox. It’s recommended to use the help desk workspace for better ticket management, reporting, and AI-assisted insights. Learn more about setting up help desk or moving channels to the help desk workspace.
• Identify used channels: identify which channels you use (e.g., email, live chat, WhatsApp) and where you want the agent to respond. Start with live chat or a chatbot as your primary channel to validate performance before expanding to higher-risk channels like email. Live chat allows you to test safely by guiding users toward supported queries, monitoring conversations in real time, and quickly identifying gaps. Once performance is consistent, expand to additional pages and channels.
• Categorize customer queries: analyze ticket history to identify high-volume, repetitive requests (e.g., Where is my order?) versus high-complexity requests (e.g., billing disputes). Aim to identify 3–5 of the most common repetitive queries that would have the biggest impact if automated.

Define the queries the customer agent should handle

Use your existing setup and query analysis to determine which requests the agent should manage and when it should hand off to a human.

Example: Sprocket Supply Co., a sustainable workspace supply company, uses a customer agent to support both individual customers and wholesale buyers. Defined handoff rules ensure the agent resolves routine issues while protecting the customer experience in complex situations.

• The customer agent should handle:
• Order status questions (e.g., Where is my order?)
• Shipping information (e.g., delivery times, shipping costs, international options)
• Return and refund policies (e.g., Can I return this item?)
• Basic product questions (e.g., sizing, availability, materials)

• The customer agent should hand off to a human when:
• An order is delayed, lost, or marked as delivered but not received
• A customer requests a refund outside standard policy
• There is a billing issue or payment dispute
• The request requires accessing or changing sensitive account/order details
• The customer expresses frustration or asks to speak to a person

Prepare content sources

Once you’ve defined what the agent should handle, prepare the content it will use to respond. The agent’s accuracy is directly dependent on the quality of these sources.

1. Set up your content sources: follow customer agent best practices when setting up your content. You can sync knowledge base articles, website pages, landing pages, blogs, uploaded files, or URLs.
2. Use the knowledge base agent (BETA): use the knowledge base agent (BETA) to generate article drafts from support tickets. This helps you turn common questions into reusable, structured content for the customer agent.
3. Maintain content: ensure all policies (e.g., shipping, returns, etc.) are clear and up to date.

For example, Sprocket Supply Co. synced their knowledge base article containing their shipping and returns policies. Now, the customer agent can answer related questions without involving a rep.

Set up your customer agent

Once your process and content are defined, create and configure the agent:

• Create the customer agent: set up the agent and sync content sources. You can add or edit synced content sources at any time.
• Define the customer agent’s identity: assign a name, avatar, and persona consistent with your brand.
• Define the customer agent’s permissions: allow access to CRM data to retrieve or update information

For example, Sprocket Supply Co. creates an agent with a professional tone and gives it access to CRM data. This agent can then retrieve order details or update contact information during conversations.

Personalize and train the customer agent

Training determines how the agent responds, what it can do, and when it should involve your team. A well-trained agent can resolve more requests independently while ensuring complex or sensitive issues are handled by a human.

Configure the following components to train the agent:

• Manage content sources: add, remove, or update content sources at any time to ensure responses remain accurate and up to date.
• Set up actions: define actions for common requests, such as resetting a password or checking an order status. This allows the agent to complete tasks instead of only providing information.
• Use the customer agent to qualify leads (BETA): configure the customer agent to ask qualifying questions, evaluate prospects, score leads based on your criteria, and route qualified opportunities to your sales team.
• Set guidelines: define how the agent should respond, including tone, structure, and level of detail. You should also specify topics the agent should avoid and provide reusable responses for common scenarios.

For example, Sprocket Supply Co. trains their customer agent using up-to-date knowledge base articles on shipping, returns and product details. They set up actions that allow the agent to check an order status and initiate return requests directly within the CRM.

They also configure the agent to ask qualifying questions, such as business size, and route high-quality leads to their sales team. To ensure a consistent customer experience, they define clear guidelines for tone and structure, and specify the agent should avoid handling billing disputes or sensitive account changes, which are escalated to a human.

Set up human handoff rules

Define when conversations should be transferred to a human. This ensures the agent handles routine requests while your team manages more complex or sensitive issues.

For example, Sprocket Supply Co. wants to ensure time-sensitive or high-frustration issues are handled by a human. They set up handoff triggers when a customer mentions refund, cancel, or can’t log in. The agent can either transfer the conversation immediately or continue assisting until a rep is available.

Learn more about setting up human handoff rules.

Test the customer agent

Before deploying the customer agent to live channels, test how it responds to different scenarios.

• Accuracy: confirm the agent pulls data from the correct content sources.
• Functionality: test that Actions (e.g., order tracking) execute correctly within the CRM.
• Logic: trigger handoff rules to ensure the agent transfers the conversation to the correct inbox or representative when required.

For example, test how the agent responds to questions about order status, returns, and refunds to confirm that it answers correctly or escalates when needed.

Deploy the customer agent

Once tested, deploy the agent. Deployment determines where the agent is used and which conversations it handles. It's recommended to start with a single channel or limit it to a small percentage of conversations. Then, analyze its performance before expanding to additional channels.

For example, Sprocket Supply Co. launches their customer agent on website chat, reviews 30 days of data, then expands to email.

When deployed, the agent can respond to conversations automatically, including outside of your team’s working hours.

Configure channel settings

Before deploying the agent, review channel settings to control how the agent interacts with customers:

• Determine the data the agent should collect: configure mandatory fields (e.g., requiring an email address before starting a chat).
• Manage threads: define whether the agent should reply to all contacts in a thread, or how it handles inactive threads.
• Manage the customer agent’s working hours: configure when the customer agent is active for each assigned channel. For example, the agent can be set to respond to conversations outside of your team’s working hours.
• Turn on reply recommendations: turn on reply recommendations to allow the AI to suggest responses for human reps to review before sending.

Determine the agent’s deployment scope

Decide which conversations the customer agent will be assigned to:

• Channel-based: assign the agent to all incoming conversations or a percentage of conversations on selected channels (Email, WhatsApp, etc.). Learn how to deploy the customer agent to channels.
• Workflows or rule-based chatbots (conditional): use workflows or rule-based chatbots to route specific queries. For example, you can build a workflow to route conversations to the customer agent based on properties such as Customer tier or Issue type. This allows you to control when the agent responds to specific conversations and tailor experiences for different customers. Learn how to deploy the customer agent to workflows and bots.

Analyze and improve performance

After deployment, continuously monitor and improve performance. For example, you could assign the agent to one channel and review 30 days of data before making changes to the customer agent or deploying to additional channels.

Analyze customer agent performance

• View performance reports and metrics: track key metrics, including the total number of resolved and deflected conversations, the number of conversations escalated to a human, agent workload and query volume. Learn more about customer agent performance reports and metrics.
• Identify and address knowledge gaps: review the questions the agent could not answer to identify gaps in your content. Update or expand your content sources, or add short answers to improve coverage. For example, if the agent frequently escalates refund-related questions, you may need to clarify or expand your refund policy content. Learn more about identifying and addressing knowledge gaps.
• Analyze content source performance: review which content sources are most frequently used, as well as which are underperforming or have low impact. Use this to inform what updates are made to content sources. Learn more about analyzing content source performance.

Use insights to improve performance

Use performance reports, key metrics and knowledge gaps to refine your content, guidelines, actions, and hand-off process before expanding the agent to additional channels.

For example, you could use the following benchmarks to help you evaluate and improve performance:

Once you consistently meet your benchmarks, you can expand to additional channels and allow the customer agent to handle a larger share of conversations.

Additional resources

• Customer agent best practices
• How to use the customer agent

Glossary

• Customer agent: AI-powered assistant that responds using synced content.
• Handoff: conversation is transferred from the customer agent to a human rep
• Deflection rate: percentage of conversations handled without creating a ticket
• Resolution rate: percentage of conversations resolved by the agent
• Actions: tasks the agent can perform, such as checking an order status or sending a password reset email.
• Guidelines: instructions that define how your customer agent communicates, including tone, response style and reusable scripts.
• Guardrails: defined boundaries for topics it should avoid and how it should respond to those topics if they come up.