Breeze AI in Practice: What Actually Works (And What Doesn't Yet)

Every CRM vendor is rushing to ship AI features. HubSpot's answer is Breeze AI — a suite of tools embedded across the platform that promises to automate content creation, sharpen lead scoring, surface insights, and act as a copilot for sales and service teams. After months of testing Breeze across multiple client portals, I can tell you exactly which parts deliver and which ones aren't ready for prime time.

This isn't a feature overview you can get from HubSpot's marketing page. This is what happens when you turn Breeze on in real portals with real data and real teams using it daily.

Content assistant: the clear winner

If there's one Breeze feature I'd recommend without hesitation, it's the content assistant for draft generation. Not because it writes publish-ready content — it doesn't. But because it eliminates the blank page problem that kills marketing velocity.

Here's how it works in practice: you give Breeze a topic, a target audience, and a tone. It generates a first draft in seconds. The draft is typically 60-70% of the way there. Structure is solid. Key points are covered. The language is generic but functional. A skilled marketer can turn that draft into a finished piece in 20 minutes instead of writing from scratch in 90.

Where it really shines is email sequences and social posts — shorter formats where the structure matters more than literary quality. In my experience, marketing teams typically see a significant increase in content output — often producing nearly twice as many email sequences — after adopting the content assistant. Not because the AI writes better emails, but because it handles the scaffolding so the team can focus on the messaging.

Setup tip: The content assistant performs significantly better when your brand voice guidelines are configured in Settings > Brand. Upload your style guide, set your default tone, and provide examples of content you consider on-brand. Most teams skip this step and then complain that Breeze writes generic copy. It writes generic copy because you gave it nothing specific to work with.

Where it falls short: Long-form content like whitepapers or detailed blog posts. Breeze tends to produce surface-level writing that lacks the depth and specificity these formats demand. For anything over 800 words, treat Breeze output as an outline, not a draft.

Predictive lead scoring: powerful in theory, demanding in practice

Breeze's predictive lead scoring analyzes your historical deal data to identify which contacts are most likely to convert. On paper, this is exactly what every sales team needs. In practice, the results depend entirely on the quality of your data — and most portals don't have the data quality this feature requires.

For predictive scoring to work well, HubSpot generally recommends having a meaningful volume of closed-won deals — typically in the hundreds — with consistent data across key properties: industry, company size, engagement history, source, lifecycle stage transitions. If your sales team has been inconsistent about logging deal properties — and let's be honest, most have — the model trains on noise and produces unreliable scores.

I've seen two outcomes. Portal A had clean data going back 18 months, disciplined sales reps, and well-defined lifecycle stages. Their predictive scores correlated strongly with actual close rates, and the sales team trusted the numbers within two weeks. Portal B had patchy data, inconsistent deal stages, and reps who rarely updated contact properties. The scores were essentially random, and the team abandoned the feature within a month.

Setup tip: Before enabling predictive scoring, run a thorough CRM audit. Run a report on property fill rates for closed-won deals. If key properties (industry, deal source, company size) are filled on fewer than 70% of records, fix your data first. Spend a month enforcing required properties on deals before turning on the AI. The feature is only as good as its training data.

Honest assessment: Predictive lead scoring is a genuinely useful feature — for portals with the data discipline to support it. If that's not you yet, don't force it. Use manual lead scoring with clear criteria until your data is clean enough.

Smart recommendations: promising but inconsistent

Breeze's recommendation engine suggests next best actions for sales reps: who to contact, when to follow up, which deals need attention. The idea is solid — reduce the decision fatigue that slows down sales teams.

In practice, the recommendations are hit or miss. About half the time, they surface genuinely useful prompts — a deal that's been sitting in a stage too long, a contact who recently visited the pricing page, a company that matches your ideal customer profile. These are the moments where Breeze feels like a real assistant.

The other half, the recommendations are obvious or irrelevant. "Follow up with this contact" when the rep already has a meeting scheduled. Flagging a deal as at-risk when the rep knows the prospect is on vacation. The system doesn't always have enough context about what's happening outside of HubSpot, and that creates noise that makes reps ignore the feature entirely.

Setup tip: Enable recommendations but set expectations with your team. Frame it as "check the suggestions once a day" rather than "follow every recommendation." The valuable signals are there — they're just mixed in with noise. As HubSpot improves the model, the signal-to-noise ratio should get better. For now, treat it as a supplementary input, not a primary workflow driver.

Copilot features: the interface that ties it together

Breeze Copilot is the conversational AI layer that sits across the HubSpot interface. Ask it questions about your data, tell it to draft a message, or have it summarize a contact's history before a call. This is where Breeze feels most natural.

The CRM data queries are surprisingly capable. "Show me deals closing this month over $10K" or "Which contacts in the healthcare industry opened our last three emails?" — these work well and save time compared to building reports or filtered views. For sales reps who aren't report-building experts, Copilot makes CRM data accessible in a way that traditional interfaces don't.

The contact summary feature is another standout. Before a sales call, ask Copilot to summarize a contact's full history — emails, meetings, page views, ticket history, deal progression. It produces a concise briefing in seconds. Multiple reps across my client accounts have told me this is the feature they use most.

Where it struggles: Complex multi-object queries. "Show me all companies with more than 5 contacts who have open deals and at least one support ticket from the last 30 days" — this type of query either returns incomplete results or fails entirely. The natural language processing works well for straightforward questions but hits its limits with complex data relationships. For context on structuring your pipeline data so Copilot queries work well, see my guide on building the perfect sales pipeline.

What delivers real value today

To summarize where Breeze stands right now:

Worth adopting immediately: Content assistant for short-form drafts and email sequences. Copilot for contact summaries and simple data queries. These features save measurable time with minimal setup.

Worth adopting if your data is ready: Predictive lead scoring — but only after a data quality audit and at least 150+ closed-won deals with consistent properties.

Worth watching but not depending on: Smart recommendations and complex Copilot queries. Useful in specific moments but not reliable enough to build processes around.

Practical setup checklist

If you're rolling out Breeze for the first time, here's the order I recommend:

First, configure your brand voice in Settings. Upload guidelines, set tone preferences, provide sample content. This takes 30 minutes and significantly improves every content-related Breeze feature.

Second, enable Copilot for your sales team and let them use it for two weeks without pressure. The contact summary and data query features sell themselves once people try them.

Third, roll out the content assistant to your marketing team with clear guidelines: use it for first drafts and short-form content, always edit before publishing, and provide feedback on what works and what doesn't.

Fourth, audit your data. If it passes the quality bar, enable predictive lead scoring as a pilot with your most data-disciplined sales team. Expand from there based on results. For a deeper dive on scoring methodology, read the lead scoring and MQL/SQL guide.

Breeze AI is not a magic button. It's a set of tools that reward thoughtful setup and realistic expectations. The teams getting the most out of it are the ones who treat it as an accelerant for good processes — not a replacement for them.