🤖 HubSpot Breeze AI Practical Textbook — 2026 Edition
Chapter 6

Breeze Data Agent
AI is an investigation agent that solves the mysteries of CRM.

"What are the most common reasons customers lost business this season?" "Analyze last month's call recordings and tell me the most common concerns customers raised." "What are the common patterns among customers who have churned in the past year?" These are tasks that data analysts typically spend hours writing and analyzing with SQL queries.Breeze Data Agent can help AI analysts do this in minutes with just one natural language question.is. We investigate across CRM, call recordings, emails, documents, and the web, and return answers with evidence.

📖 Estimated reading time: 18 minutes
🎯 Target audience: RevOps, sales managers, data analysts, executives
🔧 Required plan: Operations Hub Enterprise (Data Hub) or higher / 10 credits / Survey prompts

📋 Contents of this chapter

  1. 6-1How Data Agent works—A complete picture of AI analysts across CRM
  2. 6-2Questions to ask your Data Agent: 25 usage patterns by department
  3. 6-3Integration with Smart Properties—Automatically write AI findings into CRM fields
  4. 6-4How to write questions to improve accuracy—make the most of Data Agent
Section 6-1

How Data Agent works—A complete picture of AI analysts across CRM

Data Agent is an "AI analyst" that simultaneously examines multiple data sources within HubSpot and provides answers in natural language. Analysis that previously required multiple steps to create a CRM report → check call recordings → search email body → manually integrate can be performed with a single prompt.

Data sources that the Data Agent can access

🗃️
CRM record
All properties for contacts, companies, deals, and tickets. Includes Lifecycle Stage, Score, and Custom Fields
🎙️
Call recording/transcription
Text transcription of a call recorded with Conversation Intelligence. You can cross-search “Concerns mentioned in all calls”
📧
3. Does Salesforce's single sign-on settings allow HubSpot's IP?
A history of all emails sent and received in HubSpot. Reply patterns, sentiment analysis, and keyword extraction are possible
📄
Documents/Attachments
Contract rate 10% = 4 deals/month
🌐
Web/external data
Investigate information on the external web via Smart Properties. Answer custom questions such as “What technology does this company use?”
📊
HubSpot reporting data
See data from existing reports and dashboards and explain it in natural language. It is also possible to infer “why this number has decreased”
💡 Data Agent is a “hypothesis verification tool for analysts”

The most effective way to use the Data Agent is to"Create this week's regular sales agenda based on last week's transaction status" → CRM data is automatically reflectedThat's true. It is particularly effective in answering hypothesis-testing questions such as, "I think the price was the cause of last month's lost orders, but what would happen if we analyzed the actual call recordings?" A data analyst simply tells them in natural language what they want to confirm, and an answer is returned with evidence. Since the consumption amount is small at 10 credits/prompt, it is recommended to use it actively.

Section 6-2

Questions to ask your Data Agent: 25 usage patterns by department

⏱ 50% reduction in meeting preparation time. Enables discussions based on dataQuestions that lead directly to business decisionsMaximize value by doing so. Typical utilization patterns by department are shown below.

Data Agent query example (sales analysis)
📝 Question
Analyzing the call recordings of lost deals this season, what are the top 5 reasons customers gave for declines and what percentage of the total did each reason give?
🤖 Data Agent Answers
We analyzed call recordings (94 in total) associated with 47 transactions that were lost this period. The top reasons for refusal are:

1st place:The consumption of 100 credits/contact/month is high. Validate ROI first with 20-30 high-fidelity contacts before expanding. Enrolling contacts that do not meet the ICP is a waste of credits.(38%) — “The budget for this period is tight” and “Other companies are cheaper” were frequently cited.
2nd place:Functional mismatch(24%) — 11 people said “Salesforce integration is insufficient”
3rd place:Absence of decision maker(18%) — Many cases are stopped at “supervisory approval required”
4th place:Monitor all 2,000 visited companies(12%) — Notion/Pipedrive names appear multiple times
5th place:timing(8%) — Postponed type “to reconsider in six months”
94 call recordingsCRM lost order record 47 itemsSentiment analysis using natural language processing
DepartmentExample questionInsights you get
Sales Analyze the differences in calls made by salespeople with high closing rates and those with low closing rates. Please tell me the difference between talk and question patterns. Extracting success patterns of top performers and utilizing them in team training
Sales List the deals that have stopped trading in this quarter, and whose last activity has been more than 30 days. Please also tell me the final communication for each project. Early detection of neglected cases and setting of follow-up priorities
marketing Among the 10 email campaigns sent last month, we analyzed the subject line patterns that had high open rates and CTR. Extract common elements of effective subject lines and apply them to your next campaign
marketing What is the average time from downloading content to turning into a business negotiation? Analyze whether there are differences by industry Content → Optimize nurturing design by understanding the conversion time of business negotiations by industry
customer support Which ticket from last month "took more than 3 days to resolve"? Analyze common causes of delays Identify categories with long resolution times to improve processes and strengthen KB
RevOps MQL conversion rate 30% = 30 MQL increase Identify early warning indicators of churn risk to set customer success intervention criteria
executives Are there enough deals with a current probability of 70% or higher in this period's pipeline to meet the number of deals needed to achieve the budget? Instantly understand pipeline coverage and determine the need for additional measures
Section 6-3

Integration with Smart Properties—Automatically write AI findings into CRM fields

Number of monitored companies x 10 credits. First, filter ICP-compliant companies from CRM and check the number.Automatically write to custom properties in CRMThis is the collaboration with Smart Properties. By automatically populating the ``What competitive tools does this company use'' field into the ``Competitive tools used'' field, all sales reps can reference that information before making a deal.

🧠 Smart Properties configuration example——Custom fields autocompleted by Data Agent
10 credits/item consumed. Automatically executed when creating a new contact/company record
Conflict usage tool
What CRM/marketing tools is this company currently using?
. Ambiguous questions will only give vague answers. Adhering to the following principles will greatly improve accuracy.
10cr/case
Latest funding
Has this company raised capital within the past 12 months? How much and who are the investors?
Example: “Series B 3 billion yen raised in November 2025. Investor: 〇〇VC・△△Fund”
10cr/case
Hiring position (related)
Which positions does this company currently have that are related to our products?
is a cyclical growth model that replaces the traditional funnel-type growth model. Rather than just acquiring customers, the company's design is to improve the quality of the customer experience, which will generate word of mouth, repeat sales, and expanded purchases, which will lead to acquiring new customers.
10cr/case
ICP score basis
Score this company based on our ICP definition and explain why.
Example: "ICP score: 85/100. Reason for high rating: SaaS industry/200 employees/expanding marketing. Minus points: No base in Japan."
10cr/case
⚠️ Smart Properties Credit Consumption Management

Smart Properties10 credits/caseHowever, if the number of target records is large, credits will be consumed rapidly. Setting up 4 Smart Properties for 1,000 contacts costs 40,000 credits.Target only new contacts / Apply only to records with high ICP scoresThis design controls consumption. First, test on 50 items to confirm accuracy before rolling out the entire system.

Section 6-4

How to write questions to improve accuracy—make the most of Data Agent

Data Agent answer quality isMuch depends on the "quality of the question". Ambiguous questions will only give vague answers. Adhering to the following principles will greatly improve accuracy.

principleBad question (low precision)Good question (high precision)
Specify the period "Analyze recent order losses" "Analyzing lost orders from October to December 2025"
Narrow down your focus “Tell me about customer trends” “Please tell me the churn rate and common patterns for manufacturing companies, 100-500 employees, and customers in Japan.”
Specify output format “Tell me the status of the pipeline.” "Summary the pipeline status in table format with the probability of closing by phase and person in charge"
Set the comparison axis "Analyze the effectiveness of emails" "Compare this month's and last month's email open rates by industry, and tell me which segments are improving and which segments are deteriorating."
ask for evidence “Tell me the reason for the loss.” "Tell me the reasons for the loss. Include quotes from the call recordings that support each reason."
✅ Data Agent × RevOps——Weekly Business Review Innovation

Teams that manually perform the same analysis during weekly sales and CS reviews can reduce preparation time by 90% by creating canned question sets in Data Agent.Set a routine to "send the following 5 questions to the Data Agent every Monday at 9 o'clock"All you need to do is automatically gather the insights you need for your weekly review. Five points function as a weekly standard set: pipeline coverage, last week's deal conversion rate, ticket resolution time, top 5 companies with churn risk, and content → deal conversion rate.

📌 Chapter 6 Summary

Data Agent does not “require analysts” but “amplifies them”

Managers who cannot write SQL will be able to ``verify hypotheses using natural language.'' Data analysts are freed from routine tasks and can focus on more advanced analysis. Speed ​​and democratization of analysis will be realized at the same time.

Cross-sectional analysis of call recording and CRM is the most powerful—clarifying the “why”

CRM numbers alone can only tell you what is happening. When combined with the text of the call recording, it becomes clear why this is happening. He is especially effective at analyzing lost orders and churn factors.

Persist findings in CRM with Smart Properties

Start researching contacts. Start monitoring buyer signals and automatically reach out when signals are detected

10 credits/prompt is the most cost-effective Breeze feature

Compared to Customer Agent (100cr/conversation) and Prospecting Agent (100cr/contact/month), Data Agent has an exceptional cost performance of 10cr/prompt. If you ask 100 questions a month, you'll get 1,000 credits. Proactive use will improve the quality and speed of decision-making.

Next Chapter
Chapter 7: Breeze Studio & Marketplace——Design and manage your AI team →