We Let HubSpot’s Breeze Prospecting Agent Loose on 1,000 Old Leads. Here’s the Honest Truth.

When HubSpot launched Breeze Prospecting Agent, I did what any reasonable person would do. I let it loose on our own data.

Specifically: 1,000 dormant contacts from the CRM at Major Tom, the full-service agency I co-founded 26 years ago. Not a client’s data, ours. Felt like the right way to kick the tires on an AI SDR before telling anyone else to do it.

The first month produced numbers I am still getting used to saying out loud. 24 meetings booked from a list of leads we had essentially written off. Over $750,000 in proposals out the door. Pipeline up about 30%. Sales activity more than doubled. Error rate under 1%. And then the campaign closed its first deal within the month: a global enterprise client with operations across multiple countries, $200,000, roughly triple what a typical new engagement looks like for us. I have repeated all of that in a few conversations now and people still ask me to say it again.

I want to be clear about what this means, because I think it is easy to read results like these and assume there is a catch. There is no catch. Any B2B company sitting on a database of old leads can point a properly configured AI prospecting agent at it and generate results like this. Not theory. Not a pilot. We ran it, it worked, and the numbers above are what actually happened.

For context on why I’m writing this on the Rivetline blog: Major Tom is a 55-person full-service agency doing digital marketing, sales, development, and creative work for B2B clients across North America. Rivetline is the specialist agency I started more recently, focused specifically on AI visibility and AI-powered sales systems. Breeze falls squarely into that second bucket, which is why the field test write-up ended up here. The campaign itself ran at Major Tom, on Major Tom’s own dormant leads.

What follows is the honest account. What worked, what surprised us, where we screwed up, and what we would do differently. If you want the setup walkthrough rather than the field test, see the Breeze configuration guide.

How We Ended Up Deep on HubSpot

Quick bit of backstory before the field test, because it matters for understanding why this worked. Major Tom has been on HubSpot for over a decade. We migrated off Salesforce and ran both platforms concurrently through the transition, which was not fun but gave us a real comparison. HubSpot won on every axis that mattered to us: usability, velocity of new features, the ecosystem, and the people. Dan Tyre was my rep for a long stretch of that and taught me most of what I know about using this platform properly.

We did not stop at CRM. Major Tom’s website is built on HubSpot CMS, not WordPress, which is genuinely rare for a full-service agency. We made that call to get better personalization and to keep our marketing stack unified. We were an early Clearbit customer for enrichment, and when HubSpot acquired Clearbit and rolled it into Breeze Intelligence, the transition on our side was effectively seamless. Today we run HubSpot across sales, marketing, and service, fully integrated as our RevOps layer. Eight Major Tom employees including me have attended INBOUND over the years. We are users, advocates, and beta testers whenever HubSpot ships something new.

The point is that by the time Prospecting Agent showed up, the soil was ready. This was not a brand-new HubSpot instance figuring out lifecycle stages for the first time. It was a decade-plus of compounding structure waiting for something to exercise it.

The Backstory: Why We Even Tried This

My co-founders and I had been talking for over a year about personally reaching out to our dormant leads. The plan was always the same: 20 contacts each per week across three of us, 60 a week, 180 in a month. Couldn’t be that hard, right?

What actually happened is we would each reach out to four or five close personal connections, get a coffee meeting, feel good about ourselves, and then get buried in client work. Rinse and repeat. Those personal reach-outs converted brilliantly because they were warm and they were us. But the approach was not scalable by any stretch, and after a year of talking about the bigger project, we had basically nothing to show for it.

So when HubSpot shipped Breeze Prospecting Agent, the idea of pointing it at our own dormant pool landed at exactly the right moment. We had 7,000 dormant contacts in HubSpot. We had three founders who had been failing to do this manually for over a year. We had a fresh AI SDR built into the CRM. It was time to stop talking and run the test.

What Is HubSpot Breeze Prospecting Agent?

Breeze Prospecting Agent is HubSpot’s native AI sales assistant, built directly into Sales Hub. Unlike third-party tools bolted onto your CRM, Breeze lives inside HubSpot, with direct access to your contact records, deal history, company data, and email sequences.

Its job: identify promising contacts, research them using enrichment data, and generate personalized outreach. Human-reviewed, AI-drafted. Think of it as an SDR that reads your CRM overnight and comes in the next morning with a shortlist and draft emails ready to go. Except this SDR also writes custom follow-ups across a multi-week sequence, never forgets anyone, and never gets distracted by a client fire.

The Setup

We were not starting from scratch. Major Tom’s HubSpot instance reflects 12+ years of structured use: a clean contact database, lifecycle stages set properly, company records linked to contacts, historical email engagement data, and a working sequence library. That maturity is why this test was possible in the first place.

Setup for Breeze took about two hours. Most of the time went into feeding the Prospecting Agent enough context that it would actually sound like us:

1. Brand kit: we uploaded our voice and tone guidelines, positioning, messaging pillars, and all of our brand materials so Breeze could represent Major Tom accurately rather than write generic agency-speak
2. ICP criteria: detailed profiles of our ideal customer, including industry, company size, revenue range, services they tend to need, and the job title, age range, and interests of an ideal prospect within those companies
3. Services data: everything we sell, how we position each service, and what problems we solve for which kinds of clients
4. Enrichment and approvals: connected data sources so Breeze had real external signals, and a human approval workflow so nothing sent without a rep signing off

One thing worth flagging: we did not let Breeze loose on all 7,000 dormant contacts. We segmented down to 1,000 who had previously hit MQL or SQL status, so these were previously qualified leads who had simply gone quiet, not random cold data that happened to be in the CRM. That segmentation choice matters a lot for interpreting the results.

I was originally planning to launch the following Wednesday. I had a whole rollout doc in progress. Then on a Thursday, I looked at it and thought, “What am I waiting for?” Our sales reps had capacity. The config was ready. Perfect is the enemy of done. I pulled the trigger that afternoon. Worst case, we’d catch issues in week one and fix them. Best case, we’d be five days ahead. We ended up being five days ahead.

What Actually Happened

The quality surprised us. Not perfect, but genuinely usable. Breeze pulled company context, read the contact records and prior chat history, referenced relevant triggers like recent hiring activity and industry signals, and framed the outreach around those inputs. The emails it generated were better than what three distracted founders had been writing ad hoc.

It also did not just send one email. Each contact went into a multi-touch sequence with customized, unique follow-ups. First email, then a follow-up a few days later, then another. Business days only, business hours only. Over a 30-day window, about five touches per contact. Across 1,000 contacts, that works out to roughly 5,000 emails, each of them individualized based on the prospect’s record.

By the end of the first month, the campaign had more than doubled our overall sales activity and increased active pipeline by roughly 30%. Not 30% over some theoretical baseline. 30% over what our team was already doing. That was the number that made us pause and recalibrate how we thought about AI SDRs in general.

The Bounce Trick

This is the part I want every agency owner and sales leader to internalize, because it produced some of our best outcomes and almost nobody does it.

Standard practice when an email bounces with a “no longer with the company” auto-reply is to mark the contact as gone, note it in the CRM, and move on. What we did instead: we treated the bounce as a signal. The sales rep would find out who had replaced that person, add them to HubSpot, and reach out manually with something like, “Hey, I understand Mark Johnson is no longer with the company. We were discussing a project with him a few years ago. Who’s picking up that thread now, and how is your marketing going in general?”

About 10% of our outreach bounced. In a lot of those cases, the follow-up message got forwarded internally, often landing in the inbox of the new CMO, and sometimes the founder or CEO. We ended up opening fresh top-of-funnel relationships with people we would never have reached through a cold list. Bounces stopped being a cost and started being an opening.

What It Did to Our Database

One outcome we did not anticipate was what the campaign did for our CRM itself. When Breeze flagged contacts with thin or outdated data, we started reviewing those records more carefully. In a number of cases, we found the original contact had left the company entirely. Same pattern as the bounces, just caught earlier in the process. We reached out to acknowledge the change and ask who had taken over the relationship. Several re-engagements came directly from what would otherwise have been dead-end records.

If your CRM has sat idle for a year or two, running Breeze on it may tell you more about the health of your data than any manual review would.

The Mistakes We Made

A few things did not go smoothly. Worth being honest about them.

Our sales reps called us on some of the targeting early, and they were right. Breeze was going after some of the wrong people for the wrong reasons. When we dug in, most of those cases traced back to our own CRM: job titles that had changed, companies that had restructured, contacts who had moved on. Data hygiene problem, not a Breeze problem. But it would have been a real problem if the reps had not caught it.

We also got feedback from one or two prospects that the initial cadence felt too frequent. We had Breeze reaching out every two days at one point. We pulled that back to a maximum of twice per week, and the complaints stopped. Lesson there: AI will do whatever you tell it to, including things you would never do yourself if you were paying attention. Watch the cadence.

We also hit situations where the messaging was off-base for certain segments. Breeze works from the inputs you give it, and some of our inputs were too broad. It took a few rounds of negative keywords and exclusions before the output tightened up. If you deploy this, expect to iterate.

We paused mid-campaign to recalibrate. When we restarted, some contacts queued right before the pause went out with stale timing. A small number slipped through before anyone caught it. If you pause a Breeze campaign, review the queue before you restart. That lesson is now part of our playbook.

Overall, errors, misfires, and negative responses landed well under 1% of contacts reached. For a campaign this size on a dormant list, that is a strong number. Less than 1% is still real people, though, and a handful of those interactions required personal follow-up. Go in with that expectation.

AI-Assisted vs. Manual Outreach: What We Measured

Factor	Manual Outreach	Breeze AI-Assisted (Our Campaign)
Research time per contact	5 to 10 minutes	~30 seconds (review only)
Contacts engaged (first month)	Limited by rep bandwidth	1,000 previously qualified dormant contacts
Meetings booked	Typically a handful on a dormant list	24
Pipeline value generated	Bounded by volume	Over $750,000 in proposals
Deals closed from campaign	N/A	1 enterprise deal, $200,000, within first month
Total sales activity lift	Baseline	More than doubled
Active pipeline lift	Baseline	~30% increase
Error or complaint rate	Varies, opt-outs routine	Under 1% of contacts reached
Personalization quality	High when done well	Good, higher than expected
Consistency across reps	Varies significantly	Consistent baseline quality
CRM hygiene byproduct	None automatic	Surfaced dozens of stale records
Weak spot	Speed and consistency	Stale data, review bottleneck

10 Things to Know Before You Try It

1. Feed it everything about your brand and your ICP. The output quality is a direct function of how much context you give it. Brand voice, positioning, ICP details, services, ideal prospect profile, all of it. Generic input, generic output.
2. Segment before you unleash. Running Breeze on every dormant contact is not the move. We segmented to leads who had previously hit MQL or SQL, so we knew they had been qualified at some point. That one filter is probably why our conversion numbers looked the way they did.
3. Keep a human in the approval loop, at least at first. Even good AI output needs review. Our sales reps pushed back on some of the targeting in week one, and they were right. If your team cannot review at the volume Breeze generates, start with a smaller segment.
4. Start small, smaller than we did. We started with 1,000 and thought that was conservative. If we were doing it again, we would start with 100, validate the output over a week, refine what needed refining, then scale.
5. Mind the cadence. Our initial setup reached out every two days and one or two recipients told us that felt too aggressive. We pulled it back to twice per week and the friction disappeared. Whatever your default is, watch for signal from actual responses.
6. Treat bounces as opportunities. When an email bounces because someone has left, reach out to their replacement. Roughly 10% of our outreach bounced, and following up with the new contact created real top-of-funnel relationships, often landing in front of CMOs and founders.
7. Enrichment is not optional. The personalization that actually works comes from enrichment data. Connect your sources before launch.
8. Review velocity becomes your bottleneck. If Breeze generates 80 drafts a day, someone has to review 80 drafts a day. Plan for that capacity before you scale.
9. Your platform investment compounds. Breeze worked on day one because our HubSpot instance reflected 12+ years of structured use, including a Salesforce migration years ago and full integration across sales, marketing, and service. If your CRM is six months old and half-configured, plan on a longer runway before you see numbers like these.
10. Measure what changes, not just what sends. Track reply rate, meeting rate, and pipeline contribution, not just emails sent. Our headline result was 24 meetings and over $750,000 in proposals from a list we had written off, but the metric that shook us was active pipeline up 30% and overall sales activity more than doubled. And then a $200,000 deal closed. From a dormant list. In the first month.

Frequently Asked Questions

The Bottom Line

HubSpot Breeze Prospecting Agent is the most practical AI sales tool we have tested because it lives where your data already lives. The personalization is not magic, but it is good enough to be genuinely useful, which is more than we can say for most AI tools in this space.

The headline result for us was not any single metric. It was that the agent finished a project that three founders had been failing to finish for over a year. 24 meetings booked, over $750,000 in proposals, a $200,000 enterprise deal closed in the first month, pipeline up 30%, sales activity more than doubled. A list we had written off produced the single biggest inbound lift of the quarter. And additional opportunities from the same campaign are still in progress.

If your HubSpot data is clean, your ICP is defined, and you have someone senior to review output, Breeze is worth deploying. If you are hoping it will fix a messy CRM or replace a sales process you have not built yet, it will not. If you run Zoho rather than HubSpot, Zoho Zia Agents offer a comparable capability at significantly lower cost.

One Last Thing

I’m not a paid advocate for HubSpot. I’m just a long-time customer who has grown my agency on top of their platform for over a decade. The first time I went to INBOUND, I snuck out after the conference wrapped and watched the Red Sox play from the top of the Green Monster at Fenway Park. Fell in love with Boston so hard I went back the next day to watch the Yankees play there too. I’ve been back for INBOUND multiple times since.

That’s the context I bring to this field test. I want HubSpot to win. I also want to be honest about what this product does and does not do, because that is the only version of a case study that is actually useful to another operator reading it. The numbers above are real. The mistakes are real. The bounce trick works. The cadence matters. And Breeze Prospecting Agent, used properly, is the closest I have seen an AI tool come to earning its seat on a sales team.

Keep Reading

---

Major Tom is a 55-person full-service digital agency that has been running B2B marketing, sales, and CRM systems for over 26 years. Rivetline is the specialist sister agency focused specifically on AI visibility and AI-powered sales systems. If you want the Rivetline team to configure an AI SDR system inside HubSpot, Zoho, or another CRM, see how the HubSpot setup works or get in touch. For full-service digital work across the broader stack, visit majortom.com.