What Is the Customer Agent (and How Does It Work)?
The Customer Agent is HubSpot’s AI support agent that resolves customer conversations using your own content, knowledge base articles, website pages, files, or external URLs, and hands off to a human when its confidence is low. It’s built through a setup wizard (name, personality, content sources), trained with actions and guidelines, tested before going live, then deployed to specific channels: Facebook, WhatsApp, live chat, email, or calling (currently in beta). It requires HubSpot Credits and consumes them per resolved conversation.
If you’ve read what to automate vs. escalate or our 28-day rollout framework and want the underlying setup mechanics, this is that deep-dive.
How Do I Create a Customer Agent?
From Service > Customer Agent > Set up your agent:
- Enter an agent name.
- Choose a personality: Friendly, Professional, Casual, Empathetic, or Witty, or select “Use my brand voice” if you’ve already set one up.
- On the Content page, select what the agent can draw from: existing HubSpot content (knowledge base articles, website pages, landing pages, blogs), uploaded files (HubSpot supports .docx, .pdf, .csv, .xls, .pptx, and several other formats, with CSV/XLS optimized for structured data like product catalogs), or public URLs, including an option to auto-crawl and sync every page on a domain.
- Click Create agent.
One important detail: once created, a customer agent can’t be deleted, though it can be edited or paused at any time. Worth knowing before you spin up a test agent you don’t intend to keep.
What Happens After Creation?
Four configuration areas, each worth deliberate setup rather than defaults:
- Define: Identity (name, avatar, personality) and CRM permissions, whether the agent can access and update specific properties.
- Train: Content sources, plus actions it can perform (like resetting a password or qualifying a lead), and guidelines for how it should respond.
- Test: Preview behavior across email or live chat, run test questions or custom questions, test file attachments, and review Testing Insights, which shows why the agent responded a certain way, what triggered a given behavior, and which sources it cited.
- Deploy: Assign the agent to specific channels or route it through workflows and rule-based chatbots based on criteria like customer tier or issue type.
How Does the Agent Decide When to Hand Off to a Human?
This is the mechanic that matters most for trust. Based on its confidence level, the Customer Agent will do one of three things: generate a response backed by a verifiable source, ask a clarifying follow-up question, or reassign the conversation to a human agent. It’s a graduated response, not a binary “answer or don’t.”
You can customize this handoff process directly, including setting specific triggers (like certain keywords, customer tiers, or repeated unresolved turns) that force an escalation regardless of the agent’s own confidence score.
What Channels Can the Customer Agent Actually Run On?
Per HubSpot’s own deployment documentation, once an agent is created, defined, trained, and tested, you can assign it to: Facebook, WhatsApp, live chat, email, or calling, the last of which is explicitly labeled BETA by HubSpot. On a calling channel specifically, the agent facilitates a Q&A experience to help resolve the caller’s issue before routing to a live agent if needed.
You can also send a conversation to the Customer Agent from a rule-based chatbot mid-conversation using the Send to Customer Agent action, letting a simpler bot flow hand off to the AI agent at a specific point rather than routing everything through it from the start.
How Are Knowledge Gaps Identified?
When testing or reviewing live conversations, you can click Improve response to see how the agent should have answered, and from there create a short answer, add a handoff trigger, or otherwise correct the behavior. This is also where knowledge gaps surface, patterns where the agent lacks a confident source and needs either new content or a clearer routing rule. Reviewing this regularly is what turns the agent from “works okay at launch” into something that keeps improving.
What Does It Cost to Run?
The Customer Agent consumes HubSpot Credits per resolved conversation, HubSpot’s own credits FAQ specifically cites 50 credits ($0.50) per conversation resolved without human help. New setups typically get a 28-day free access period where the agent doesn’t consume credits at all, HubSpot sends a reminder four days before that period ends, and the agent automatically turns off (rather than starts charging unexpectedly) once it expires, until you re-enable it.
Worth flagging directly: this free-access period is a HubSpot billing mechanic, separate from any phased rollout strategy you might use operationally. If you’re planning a multi-week deployment approach, see our rollout framework, just don’t assume the two “28 days” are describing the same thing.
The FSM Perspective: Testing Insights Are the Real Configuration Step
The setup wizard makes creating an agent feel fast, name, personality, content, done. But the Testing Insights panel, where you see exactly why the agent responded a certain way and which sources it cited, is where the real configuration work happens. Skipping straight from creation to deployment means finding out about knowledge gaps in front of real customers instead of in testing.
This is the Process pillar of our Scalable Business Framework applied directly: the wizard sets up the shell, but the actual quality of the agent comes from iterating against Testing Insights and knowledge gaps before it ever reaches a live channel.
Frequently Asked Questions
Can I delete a customer agent once it’s created? No. It can be edited or paused at any time, but not deleted.
What content can the agent be trained on? Existing HubSpot content (knowledge base, website pages, landing pages, blogs), uploaded files in several supported formats, and public URLs, including full-domain crawling.
Does testing the agent use credits? No. Previewing and testing the customer agent doesn’t consume HubSpot Credits.
Which channels support the Customer Agent? Facebook, WhatsApp, live chat, email, and calling (currently in beta).
How much does each resolved conversation cost? 50 credits ($0.50) per conversation resolved without human help, per HubSpot’s credits documentation.
Not sure your content library is ready to train an agent on yet? Let’s talk it through.