Breeze AI, rolled out across all HubSpot platforms starting in 2025, is Service Hub's most powerful weapon.Breeze Customer Agent provides 24-hour primary support, Breeze Copilot assists with agent replies, and Knowledge Base Agent automatically generates KB articles.half an hour
Breeze Customer Agent (hereinafter referred to as ``Customer Agent'') is an autonomous agent that uses AI to directly respond to customer inquiries. Browse learning sources such as knowledge bases, websites, PDFs, videos, etc.Respond to customers in sentences equivalent to human agents. If the issue cannot be resolved, it will be automatically escalated and the conversation history will be transferred to the ticket. Available on all HubSpot platforms starting June 2025 (deepest integration with Service Hub Pro+).
For Customer Agents, ``what you are learning'' becomes ``what you can answer.''The type, quality, and update frequency of the learning source determine the AI resolution rate.. HubSpot supports four types of learning sources, each with different characteristics.
If you do not update the learning sources after updating the product, the Customer Agent will continue to respond with outdated procedures and obsolete features.Update related KB articles/web pages and re-upload PDFs every time a product is released.It is essential to agree on operational rules with the Product team. Most complaints that the AI gave a wrong answer arise from problems with the freshness of the learning source.
The most important design element of Customer Agent isWhen will it be handed over to humans?"is. The essence of handoff design is to clearly distinguish between inquiries that should continue to be handled by AI and those that should always be handled by humans, and to automate this process. If the handover is too slow, customer frustration will increase, and if it is too early, the purpose of setting up AI will be lost.
When handing over from AI to humans, the worst experience for customers is for agents to have to "relisten from scratch."The ticket that is automatically created at the time of handover includes the entire conversation history, the resolution steps attempted by the AI, the customer's contact information, and the CRM health score.Make sure that.
| Transfer information | automatic/manual | Setting points |
|---|---|---|
| Full history of conversations with AI | Old tickets are lined up at the top and I have to look for the latest status. | The full text will be recorded in the ticket's "Conversation" tab. Agents can check how far they have tried from here |
| Summary of solution steps attempted by AI | Automatic (summarized by Copilot) | Breeze Copilot automatically generates an “AI-enabled summary” as an internal memo when creating a ticket |
| Contact/Company information | Automatic (CRM linking) | When a customer enters their email address in chat, they are automatically linked to a contact in your CRM. |
| Health score/contract plan | Automatic (CRM linkage) | Branding → Primary color |
| Classification of escalation reasons | Automatic generation + manual confirmation recommended | The escalation reason (keyword trigger/unresolvable/customer request) determined by AI is recorded in an internal memo. |
If the AI makes an escalation decision outside of business hours, the ticket will remain at the "New" stage because there is no agent in charge.“Escalation outside business hours → Workflow to notify manager at 9:00 the next business day”By setting this, we create a system in which important tickets that arrive after hours are processed with top priority the next morning.
Breeze Knowledge Base Agent (Enterprise only) explained in Chapter 4 can be used in combination with Customer Agent.Self-reinforcing cycle of “ticket → KB article generation → AI resolution rate improvement → ticket reduction” (AI flywheel)Realize. The goal is for the two AIs to work together, allowing KB to continue to grow autonomously without the hands of the operations staff.
| step | operation | in charge | effect |
|---|---|---|---|
| ① Ticket accumulation | A certain number of solved tickets in the same category (estimate: 5 to 10) will be accumulated. | Old tickets are lined up at the top and I have to look for the latest status. | KB Agent analysis triggers are generated |
| ② Article draft generation | KB Agent analyzes ticket content and automatically generates drafts of article titles, instructions, and SEO meta descriptions. | KB Agent（AI） | Article creation time reduced by approximately 60% |
| ③ Human review/publication | The person in charge checks and supplements the draft and publishes it on KB. Immediately reflected in Customer Agent learning sources | Person in charge (human) | Addition of quality assurance and latest information |
| ④ Improve AI resolution rate | New articles added to Customer Agent learning sources to increase self-resolve rates for tickets in the same category | Customer Agent（AI） | Escalation rate in the same category has decreased |
| ⑤ Decrease in tickets | By increasing self-resolution, the number of tickets sent to agents decreases, giving more time to improve the next KB. | health | global marketing |
KB Agent is limited to Enterprise, butEven in Professional, you can manually run the cycle of "Weekly ticket analysis → KB article creation → Customer Agent accuracy improvement". By setting up a routine where every Friday, I select one question that has come up three times or more this week and write an article about it, my KB grows at a pace of 4 to 5 questions a month, and after 6 months, the customer agent's resolution rate has significantly improved. Regardless of whether it is automated by AI or not, the design philosophy of the flywheel itself remains the same.
Breeze Copilot is an AI assistant that provides real-time assistance to agents while they are on the ticket screen. While Customer Agent is used for first-line response, Copilot isImproving the quality and speed of “tickets handled by humans”used for Always visible in the right pane of Help Desk Workspace, agents can take advantage of it without having to open another tool.
Traditionally, it took several months for a new agent to become fully capable of dealing with clients.By using Breeze Copilot's reply suggestions, KB suggestions, and next action suggestions, even newcomers can provide answers of a certain quality from the first day.Become. By strictly adhering to the rule that ``do not send Copilot proposals as is, always check and revise them in your own words before sending,'' productivity will be improved while preventing skill stagnation due to AI dependence.
Breeze Customer Agent consumes 100 credits (approximately $1) per conversation. For organizations with a large number of monthly inquiries, credit costs become too large to ignore.While managing credits as “costs”, accurately measure ROI based on the difference between “reduced agent effort”This is essential when deciding whether to continue investing in AI.
| Management items | Setting location | Recommended settings |
|---|---|---|
| Setting a monthly credit limit | Settings → AI → Credit management | The upper limit is set at 120% of the estimated monthly consumption. Send overage alerts to CFO/CS manager |
| Behavior when credits are exceeded | Settings → AI → Behavior when exceeded | Changed the default "Auto Continue" to "Stop and Notify". Preventing unintentional mass consumption |
| Monitoring AI resolution rates | Reports → Customer Agent → Resolution Rate | Check the three indicators of “AI resolution rate”, “escalation rate” and “credit consumption” on the dashboard every week. |
| Credit allocation by channel | Settings → Breeze → Channel settings | If you use Customer Agent across multiple channels, such as live chat and WhatsApp, understand consumption by channel and focus on high-priority channels. |
Customer Agent (primary response), Copilot (agent support), and KB Agent (article generation)—the three AIs have a complementary relationship. It is practical to prioritize the Customer Agent settings first, enable Copilot in parallel, and add KB Agent after the Enterprise contract.
If you do not update the KB, web page, and PDF every time a product is updated, you will continue to guide customers to outdated information. The Product/CS team should agree on a learning source update routine that matches the release cycle.
Refunds, cancellations, complaints, VIP customers, customer requests—these cases set keyword triggers that the AI immediately escalates to humans. Escalations outside business hours should be designed in conjunction with the next morning notification workflow.
It is particularly effective in improving the quality of responses for new agents, but the habit of sending proposals as is can lead to skill stagnation. ``Refer to Copilot's proposal and confirm it in your own words before sending'' as a team rule.
Credit cost of $1 per conversation vs. labor cost of $8 to $15 per human interaction - ROI is the number of cases solved by AI x the differential cost. The three indicators of "AI resolution rate, credit consumption, and man-hours saved" are tracked on a monthly basis on the dashboard.
The more KB the customer agent has, the higher the customer agent's resolution rate, and the KB agent generates the next article. Once this cycle begins, you'll accumulate data assets that your competitors can't keep up with just by introducing tools.