I would start with a task the team repeats and can judge: preparing for a call, answering a common support question or reviewing an outreach draft. That gives you something concrete to test before expanding your use of AI.
Breeze covers an evolving set of assistants, agents and AI features in HubSpot. This guide focuses on everyday assistance, support, content, prospecting and data enrichment. Check each feature's requirements separately; sharing a brand name does not mean sharing access or billing rules.
Breeze Assistant: start with a task you can check
HubSpot's current documentation calls the conversational tool Breeze Assistant. Older material may refer to Breeze Copilot. The assistant can summarize records, prepare for meetings and help generate content. Its actions depend on user permissions, AI settings and connected tools. Check the Assistant requirements.
For a small pilot, ask for a summary of a familiar account or a draft follow-up based on a known conversation. Compare the response with the underlying records. Check dates, missing activity and any claims the draft makes before using it.
I would judge whether it reduces the work of finding and checking information. A fluent summary is useful only if it reflects the right customer context.
Customer Agent: check the sources and human handover
Customer Agent can use knowledge base content, website pages, imported URLs and uploaded files. It is not restricted to a HubSpot knowledge base. HubSpot explains how to manage its knowledge sources.
Start with a defined category of questions and maintained answers. Include questions the agent should hand to a person, then test whether that handover reaches the right team. Track incorrect answers and repeat contacts alongside the number of conversations handled.
Source selection matters: HubSpot notes that private source content can still appear in customer-facing responses even when the source itself is not cited. Review what is appropriate to expose before adding it. Give someone responsibility for updating the answers as policies change.
Content Agent: use the context, then edit the draft
Content Agent can draw on account context and brand voice settings to produce content. The current documented beta lists Marketing Hub Professional or Enterprise as a requirement for its features across tools. Review its requirements and supported content types, rather than assuming every content feature is included with Content Hub.
I would try it on an outline or a variation of content you already know. Check the facts, positioning and promises, then read it against your own writing style. Brand voice settings help provide direction; they do not replace editorial judgement.
Measure the work through approval, including corrections and rewriting. The time taken to generate a draft is only part of the task.
Prospecting Agent: review the audience and the message
Prospecting Agent supports prospect research and outreach through configured plays. Its email review settings distinguish reviewing messages before sending from sending without review.
I would begin with review enabled and a narrow audience. Confirm the fit criteria, existing customer relationships and contact ownership before enrolling records. Read the research and messages against what sales actually knows about those accounts.
Then measure accepted recommendations, relevant replies and corrections needed. An activation count or a high volume of generated emails does not tell you whether the targeting works. Agree who handles replies and removes unsuitable contacts from the process.
Data enrichment: check eligibility and overwrite behavior
Treat the older Breeze Intelligence label as a starting point for finding the current enrichment documentation. HubSpot now states that data enrichment does not consume HubSpot Credits, while smart properties do. The enrichment overview explains that distinction and the subscription requirements.
Coverage still depends on the record. Contacts need a business email and companies need a domain; available information varies. Automatic, manual and continuous enrichment also differ in how they fill or replace values. Check enrichment behavior before using it on existing records.
I would test a relevant sample and review the fields that matter to your process. Decide which source owns each value, especially where another system also writes to HubSpot. A headcount or database-size threshold alone is a poor reason to enable or reject enrichment.
Check access and usage costs before the pilot
HubSpot Credits apply to specific usage-based features. Availability also depends on subscription and seats. Check the current consumption rules, included allowance and overage settings in your account; a quoted dollar price from an older announcement is not a complete cost estimate. HubSpot's credit guide links to the current product catalog.
Before testing, write down what counts as a useful result and how you will stop or limit the pilot. Include the cost of review and maintaining source content. For support, also distinguish a resolved conversation from one that returns later or needs a person.
Keep the data and review process in scope
Choose a task with a clear owner, usable source information and an observable result. Keep the pilot small enough to inspect the outputs and investigate exceptions before increasing its reach.
My data quality case shows how I separate a detected mismatch from the decision to correct a record. The lifecycle stage guide covers the qualification definitions that prospecting and handovers depend on.
Questions about a particular setup? You're welcome to compare notes.