HubSpot AI Integration Guide 2026: Breeze AI and How to Connect AI to HubSpot
Guide to AI in HubSpot: what Breeze can do, when to connect other AI, four integration routes, data requirements, approval and GDPR. For Danish B2B teams.
Table of contents
What can HubSpot's AI, Breeze, do for a sales team?
HubSpot gathers its AI functions under the name Breeze. This includes, among other things, an AI assistant in HubSpot (Breeze Assistant), AI agents for selected tasks (for example customer service and prospecting) and data enrichment. The purpose is that the CRM itself helps write, summarise, enrich and suggest, so that salespeople type less. Which functions you actually have access to, and what they cost, depends on your subscription. Check with HubSpot or in your own account, because names, packages and prices change often.
TL;DR:
- Start by seeing what Breeze and your subscription already cover. That is the cheapest and quickest option.
- Connect other AI where HubSpot's own functions do not reach: your own language and processes, systems outside HubSpot, or special requirements for approval.
- Clean data and clear rules for what the AI may write in the CRM matter more than the choice of tool.
What is HubSpot Breeze in 2026?
Breeze is HubSpot's overall AI layer. Broadly, there are three kinds of functions:
- Assistance: an assistant that helps write emails, find information, summarise contacts and answer questions about data in HubSpot.
- Agents: AI agents for delimited tasks, for example answering customer enquiries from your knowledge base or preparing prospecting.
- Data: enrichment of company and contact information.
This is an area that is developing quickly. Use HubSpot's own documentation as the source for what is available, in which subscription, and in which languages.
Which HubSpot functions are most relevant for sales?
For a sales team the typical candidates are:
- Summaries of contacts and agreements, so a salesperson can quickly get up to speed on a case.
- Email drafts and follow-up based on what is in the CRM.
- Call recording and summary, if it is included in your subscription and you have the rules in order.
- Data enrichment, so fields such as industry and size are filled in automatically.
- Prospecting agents, which can find and prioritise leads according to your criteria.
In each case ask: What can the function do? What does it write in the CRM? Can we control what it may do, and can we see what it has done?
Why do CRM data fail, and what does it cost?
This is the recurring reason AI does not deliver. No AI can be better than the data it receives. Typical problems:
- Contacts who have left their jobs, or companies that have closed.
- Duplicates.
- Fields that are filled in differently by different salespeople.
- Agreements that are not updated, because it happens after meetings, when time is short.
- Information that exists only in emails or in salespeople's heads.
The consequence is that the AI builds suggestions on wrong assumptions. So clean up first. It is rarely exciting, but it is the investment that pays off most.
Which routes exist for connecting AI to HubSpot?
| Route | What it is | Suits | Watch out for |
|---|---|---|---|
| 1. HubSpot's own AI | Breeze and the functions your subscription includes | Standard tasks in HubSpot | Does not always cover your language and processes |
| 2. Apps from HubSpot's marketplace | Ready-made integrations from other suppliers | Telephony, transcription, single tasks | Each app has its own rules on data and price |
| 3. Workflows with external calls | HubSpot workflows can call other systems via webhooks and code actions (depends on subscription) | Teams with technical competence | Requires maintenance when systems change |
| 4. Custom AI layer via HubSpot's API | A solution built for your process that reads and writes in HubSpot through an integration | Special processes, several systems, requirements for approval | Requires that you can describe what you want |
The four do not exclude one another. Many start with route 1 and supplement with 3 or 4 where it does not suffice.
How do you set up AI in HubSpot: four steps
Step 1: Get the data ready
Remove duplicates, correct the most important fields and agree what each stage of the pipeline means. Also decide which contacts the AI may work with. It is especially important to be able to exclude those who have opted out of contact.
Step 2: Choose one task
For example notes and updates after meetings, or a draft follow-up. Describe what happens today, who does it, and how long it takes. That is your starting point.
Step 3: Determine rights and approval
What may the AI read? What may it write? Give it the fewest possible rights. Let it write as a note before it is allowed to change fields. Price, stage and probability should require a person's yes. Nothing that goes out to a customer should be sent without approval.
Step 4: Test, measure and expand
Use a test account or a small extract of data first. Compare with your starting point. Ask the salespeople who use it whether it helps. Expand when the first task runs without supervision.
How does this relate to Salesforce and other CRM systems?
HubSpot is not alone. Salesforce has its own AI layer, and the same goes for Microsoft Dynamics and others. The principles are the same: clean data, clear rights, approval and measurement. If you use several systems, or are considering switching, that argues for keeping the logic separate from any single CRM; we describe the thinking in the article on headless sales AI.
A rights matrix that makes it safe
Before anything is connected, write down what the AI may do in HubSpot. A simple matrix is enough:
| Action in HubSpot | The AI may do on its own | Requires a person's yes | Must never |
|---|---|---|---|
| Read contacts, companies and agreements | Yes, within the scope | ||
| Write a note or summary | Yes, marked as AI-written | ||
| Create a task | Yes | ||
| Fill in empty fields | Yes, until you trust it | ||
| Change agreement stage, price or probability | Yes | ||
| Send an email or message to a customer | Yes | ||
| Delete records | Yes, never | ||
| Contact those who have opted out of contact | Yes, never |
Adapt it to your own rules. What matters is that it exists, that it is written down, and that it can be seen by those who work in the system.
Five pitfalls
1. You start with the tool, not the task. Then you end up with a demo that impresses and a working day that is unchanged.
2. You write too much too fast. An AI that may change fields from day one can make many errors before anyone notices. Start with notes and suggestions.
3. You forget to measure. Without a starting point you cannot see whether it helps. Measure the time spent on the task before you build.
4. You think AI solves a missing work process. If the salespeople do not use the CRM consistently, you are automating a gap. Solve how you work first.
5. No one owns it afterwards. Systems and models change. Someone must be responsible for ensuring that it keeps working, and for correcting the instructions when something goes wrong.
GDPR and compliance for AI in HubSpot
AI in the CRM processes personal data. The most important questions are:
- Basis and information. Why are you allowed to process the data in this way, and are the people informed?
- Suppliers. Do you have data processing agreements with HubSpot and with all external tools that get access? Where is data processed, and who are the sub-processors?
- Transfers. Does data leave the EU/EEA? Which transfer basis is used?
- Who sees what. Limit access so that the AI sees only what it needs.
- Recording. If conversations are recorded and analysed, special requirements apply; see the guide to call recording in Denmark.
- Training. Is your data used to train the supplier's models?
HubSpot offers, among other things, a choice of data region. Check what your account is set to, and have a lawyer assess your overall setup.
From conversation to CRM: the flow in practice
Many salespeople spend the evening typing into the CRM what they did during the day. The flow we build is simple in principle: the conversation, if it may be recorded, or the salesperson's short status, is written out in Danish. The AI produces minutes and proposes fields and the next step. The salesperson sees it, corrects it and approves it, and only then is it written into HubSpot.
On the use cases page you can see it as "The meeting is over. The CRM work begins". The agent may write the minutes and update the fields itself, but asks first before anything goes out to the customer. It can be connected to HubSpot, Salesforce, Pipedrive and Microsoft Dynamics, as well as telephony and calendars. This only makes sense if the CRM is already used consistently. The examples use invented names and figures and show the structure, not a result you can count on.
If an AI is also to call by itself, that is a separate track. Here we work with JesperAI, and the rules for automated calls must be assessed legally before anything is set in motion.
What can Agent360 contribute?
We build AI solutions in the customer's own systems, including on top of HubSpot. We contact your suppliers and help with access, and you are copied on everything throughout. The first thing we build is completed in full before we go further. The price depends on the task, and we give an estimate after a short conversation. If you would rather learn to build with AI yourself, there is a personal mentoring programme.
Frequently asked questions about HubSpot and AI
Q: Is Breeze enough? For many standard tasks it may be. If your process is special, if you use several systems, or if you have requirements for approval and Danish language that Breeze does not cover, a supplementary solution may make sense. Start by seeing what you have.
Q: What does Breeze cost? It depends on subscription and usage, and it changes. See HubSpot's current pricing page, and ask HubSpot what is included in your subscription.
Q: Can the AI write directly into our CRM? Yes, with an integration. We recommend that it first writes notes and suggestions, and that changes to stage, price and probability require a person's approval.
Q: How do we make sure the AI does not contact those who have opted out? By recording it in the CRM and letting the AI layer read the marker before any draft is written. Nothing should be sent without a person's review.
Q: Can we use Breeze and a custom solution at the same time? Yes. Many do. Agree who does what, so that they do not write to the same fields.
Q: What is the difference between Breeze and Salesforce Einstein? Both are the CRM suppliers' own AI layers, with different functions, subscriptions and prices. Compare them on the tasks you actually need to do, not on the headlines.
Conclusion: From systems that store to systems that help
A CRM becomes more valuable when it does not only store, but also helps write, summarise and remember. Start with what HubSpot already offers, clean up your data, choose one task, and decide what the AI may and may not do.
Book 20 minutes of free advice if you want to hear whether your task in HubSpot is worth building. We will also tell you if it is not. See also examples of tasks we can help with.
Sources
- HubSpot, Breeze
- HubSpot, Developer documentation
- HubSpot, Cloud Infrastructure and Data Hosting
- The Danish Data Protection Agency (Datatilsynet), rules and guidance
#hubspot#breeze#ai#crm#integration
Contents
Should your CRM keep itself up to date?
We build the layer that moves data into the CRM you already use, without anyone having to type it in. We look at your current set-up first.
Book a reviewShould your CRM keep itself up to date?
We build the layer that moves data into the CRM you already use, without anyone having to type it in. We look at your current set-up first.
How we built it · CRM & customer overview
Customer CRM that updates itself
An internal page per customer with contacts, the agreement, the projects and a timeline where emails, meetings, email drafts and tasks land by themselves. Above the pages sits the board with all customers, owner and latest contact, and the system suggests next steps based on fixed rules.
- Email, meeting, task
- Lands on the customer’s page
- You take the next step
Related articles

MFA: multi-factor authentication for your business – how to get started
Learn how to implement MFA securely in your organisation. Guidance on methods, costs and best practices for protecting company data.

English-language invoices: requirements for invoices to foreign customers
Invoice in English to foreign customers legally. Danish regulations, EU and international VAT rules, zero-rating, reverse charge, and automation.

Public Procurement in Denmark: how to bid for government contracts
How Denmark's Public Procurement Act works for suppliers. Threshold values, tender procedures, exclusion grounds and AI solutions for proposal management.