AI in sales

AI in HubSpot: what already works for a B2B sales team

Everyone is talking about AI agents. For a sales team of ten or twenty people the question is simpler: where does it save time today, without mistakes reaching customers?

Thovar · Updated 21 October 2026 · 5 min read

Lots of promises, little practice

Every software vendor talks about AI agents that find leads, write emails and handle customer questions on their own. For a B2B company with a sales team of ten or twenty, the question is much simpler: where does it save us time today, without mistakes reaching customers?

The answer is less spectacular than the demos, but useful. AI in HubSpot is mostly good at work that keeps coming back and that nobody enjoys: summarising, writing a first draft, filling in data. And it only works if the CRM data is right.

What AI HubSpot has today

HubSpot calls its AI Breeze. Roughly, there are two parts:

• Breeze Assistant. An assistant in the CRM that summarises records, suggests emails, prepares meetings and answers questions about your data. It's in every package.
• Breeze agents. Components that carry out tasks on their own, such as a customer agent that answers customer questions, a prospecting agent that researches companies and suggests outreach, and a data agent that enriches records. Most agents need a Professional or Enterprise package.

Agents are paid for with HubSpot Credits. Each package has a monthly allowance, around 3,000 credits a month on Professional. Unused credits expire at the end of the month. According to pricing published in October 2026, a conversation resolved by the customer agent costs around 50 credits, and a lead for which the prospecting agent recommends outreach around 100 credits. With heavy use, the allowance runs out quickly.

Five uses that already work

1. Prepare a visit in two minutes

Before a customer visit, a salesperson wants to know what's going on: recent emails, open quotes, who has changed roles since the last meeting. The assistant summarises the company record. That saves a quarter of an hour of searching per visit, and the salesperson walks in better prepared.

2. Call notes that end up in the CRM by themselves

Typing up notes after a call is the first thing salespeople skip. AI can turn notes or a recording into a short summary, with agreed next steps as tasks. Always tell people a call is being recorded, and agree who checks the summary.

3. A first draft of the follow-up email

A quote that's been open for a week deserves a short, personal email. AI drafts one from the deal, the salesperson edits and sends it. The key word is draft: nobody but the salesperson presses send.

4. Enrich and clean data

Filling in missing industries, websites or company sizes, spotting duplicate companies, turning free text into fixed values. That's work AI is good at and people hate. Do spot-check what changes before you let it loose on thousands of records.

5. First-line customer questions

A customer agent can answer common questions in chat or by email, for example about delivery times, documentation or order status. That only works with a good knowledge base behind it. Without one, the agent guesses, and your customer notices straight away.

Where AI doesn't help (yet)

• A messy process. If your deal stages mean nothing, AI won't turn them into a better forecast. It just summarises the mess faster.
• Bad data. AI reads what's in the CRM. If that's wrong, you get confident answers that are wrong. That's worse than no answer.
• Decisions about customers. Which customer gets priority or what discount is possible stays with people who know the customer.
• Emails nobody has read. Don't let AI send messages to customers or prospects without someone checking them. One strange email costs more trust than a hundred good ones earn.

Data first, then AI

Most AI features are only as good as the data they work on. Before you switch anything on, it's worth checking:

• Does each company and contact appear only once?
• Are the key fields, such as industry, owner and stage, filled in on most records?
• Do deal stages mean the same thing to everyone?
• Are emails and calls logged in the CRM, or do they sit in individual inboxes?
• Is there a knowledge base with current answers, if you want to use the customer agent?

If the answer is often no, start there. How to approach that is in Why salespeople don't use the CRM, and what does work.

GDPR and the EU AI Act

AI in the CRM means customer data passes through AI models. That isn't forbidden, but it calls for a few agreements:

• Know what's switched on. In HubSpot's AI settings you decide whether AI features may use CRM data. Only switch on what you actually use.
• Check the data processing agreement and hosting. HubSpot has a data processing agreement and offers hosting in a European data centre. Check which AI components fall under it.
• Be open with customers. Under the EU AI Act, people must know when they're talking to an AI system. Those transparency rules apply from August 2026. A chatbot posing as an employee is out.
• Ask consent for recordings. Tell participants in advance that a call is being recorded, and why.
• Agree internal rules. Which data may and may not go into a prompt, and who checks what AI produces before it reaches a customer?

This isn't legal advice. If you're unsure about your situation, check with your privacy lawyer or data protection officer.

How to start

1. Pick two tasks. Tasks that come back often and take time, such as call notes and visit preparation.
2. Check the data they need. Clean up what the AI will read first.
3. Switch it on for a small group. Three salespeople who want to try it, not the whole team at once.
4. Measure after a month. How much time does it save, and how often did someone have to correct something?
5. Expand what works. And switch off what doesn't pay off. That saves credits too.

That way AI doesn't become a project in itself, but a few concrete improvements your team notices every day.