George B. Thomas opened this one with three questions for Max Cohen of hapily and Chad Hohn. Have you ever walked into a portal and suggested an audit? Walked in and thought, this needs an audit? Or just wished somebody had audited the thing? Chad's first answer set the tone for the whole episode: it depends on the definition of audit.
Here's where the crew landed. An audit doesn't have to be one giant project every few years. Small micro audits, one area at a time, can sit alongside a full portal audit, and AI can now do a lot of the reading. George showed how Sidekick's full audit went from weeks to under a day. And all three kept coming back to the same point: the humans who know the process, and the humans who live in the portal, still belong in the loop.
Bad Idea Buildup Happens to Every Portal
When Max did this kind of work for other people, it was mostly as an implementation specialist in brand-new portals. A blank canvas. It wasn't until he came to hapily, three or four years ago, that he lived inside a used, lived-in portal. The thing he notices all the time there is what he calls bad idea buildup. He admitted he just made that up. The cause, in his words: hapily doesn't have a dedicated admin. Everyone in there is really good at HubSpot, and no one is in charge of what happens. There's no king regulator or queen regulator of the portal.
“It's just a lot of super admins with a lot of good ideas that seem like really good ideas at the time.”
Then a couple of months or a year go by, and you're asking, "Oh, we don't do that anymore. That's still there?" Max described the artifacts and scars left by failed experiments, or by an idea implemented on the same call without a detailed plan: workflows, properties, a record layout nobody uses anymore.
George thinks everybody runs into this. Plenty of organizations, especially further upstream, have multiple super admins doing multiple things. Lots of chefs in the kitchen, lots of meals being cooked, and sometimes nobody's wiping down the counters. If that's your portal, start with our portal cleanliness guide for HubSpot super admins, and ask yourself whether you need a dedicated HubSpot Super Admin.
It Depends What You Call an Audit
Chad's point: sometimes a human just needs to get something done, even with a bit of a dumpster fire of data in there. You do your due diligence, try not to rebuild what's already there, and get them to time to value. But the more you look, the more you go micro.
Take a property audit: all of that junk is tied into reports, so you probably want to look at which reports matter first. If a report or dashboard hasn't been opened in six or more months, you could probably say goodbye, unless it's an annual report somebody only looks at once a year. Clearing those out frees you up for the other audits.
The micro audits Chad named on the call:
- Reports and dashboards nobody has opened in six months or more
- Properties: which ones matter and what they're associated with
- Email templates and sequences, which can now be inspected via API
- Workflows and views
- Duplicates: with Data Hub Enterprise you can flag them, pull them via API, and review them with an LLM
HubSpot has recently added new API scopes, and Chad said views are among them, so a view audit is on the table now. Or you go for the whole kit and caboodle.
George loved that Chad went micro. Over the last 30 days, he said, Sidekick has been living in that macro and micro world: a sequence audit that looks at templates, sequences, LinkedIn steps and tasks against a standard of being human, adding value, being content-driven and being AI-assisted. A workflows audit tool, because historically a workflows audit was a spreadsheet and a nightmare. And he and Jorge have been talking about a segments audit.
His questions to the crew: can somebody internal build that and run it through Claude and some APIs, or do you hire someone for the micro audits? And has anyone seen people selling an audit that isn't just the full-blown version? (If you want a picture of the full-blown version, here's the portal audit checklist every admin should review.)
Define the Process Before You Send the Robots
Chad does his own review and auditing. As new API scopes become available, his move is to take a manual process he does once a quarter and fold it in weekly, because if an LLM can do it weekly, he stays fresher and more hygienic. (For the manual side of that, see our CRM data hygiene guide.) But he wanted to be clear: to do it programmatically, you need at least some idea of what you'd have done manually.
“You can't just let an army of robots loose on a process or a problem if you don't know how to define it well.”
George agreed: "I think you have to have the process and the problems in place to then send the agent army out to do the things."
On tooling, Chad said the regular old Claude connector or ChatGPT connector isn't enough to do all of it (we've covered what the Claude connector's scopes can reach before). He'd add the HubSpot agent CLI, which in his words has "secret magical scopes that don't even exist at the API level," plus a service key with access to whatever new scopes come out. HubSpot's agent CLI developer docs cover the setup. George noted an audit could run read-only, with no write permissions. Chad's add: start with read, and if you want your agents to resolve things afterward, give them that next.
Chad's Agent Slacked the Humans First
Chad did exactly that at his organization recently. He had his agent find all the reports older than his cutoff and Slack each owner their reports and dashboards, with links, separated. The message: do you need any of these? If you don't, and you don't respond by November 11th, I'm going to blow them up. Most everybody said get rid of them. Two people said, "I want these two." George's takeaway: he loved that Chad communicated with the humans, and that he gave them time.
How George Used to Audit, and What Changed
Max asked how many portal audits George has done. George, who's been at this since 2013, started at 50 plus, moved to 75 plus, then said maybe in the hundreds, adding that he didn't want to hallucinate a number. Walking into a portal and wondering what's happening in there? Thousands of times.
So what did that look like before AI? George said that's part of why they were having this conversation. An audit historically took anywhere from two and a half to four weeks: the persona property, the rest of the properties, workflows, segments, any data cleanup, communication, email. He put the price at $5,000 to $7,500. It was crazy, he said, and there had to be a better way.
So he showed the crew what Sidekick built for the full audit, the big boy audit. The first test subject was his own portal, which he's had since 2013 or 2014 and called a hot mess. His reason for showing it: "you can leave your T-shirts, your underwear, and your socks laying around on the floor with the best of them."
Under the hood, it uses all the API reads George can give it, the Claude connector, and the Claude Chrome extension to fill in gaps, all run through Claude Code with the API scopes. And it takes less than a day.
Here's what the human gets. A short audio overview in George's AI-generated voice (Max called it a snack in a backpack). Filters for bad, medium, good, not checked and security issues. Progress tracking, so a second or third audit shows what was fixed versus what's still sitting there. And every item links into HubSpot and to knowledge articles with best practices. Yes, his own portal had findings, including deals missing associations and unengaged marketing contacts. The audit also has a section on what you're doing right. George explained why.
How the Human Feels When the Audit Lands
“One of the things that I don't think a lot of us have paid attention to historically is how the human feels when they actually get the audit delivered to them.”
Picture it: after two and a half to four weeks, you deliver the findings, and the humans feel totally bad about themselves and their organization. How did we get here? Deliver it so they can see they're doing good stuff, and the next question becomes, what do we need to fix next?
Humans Review It, and Humans Answer Questions Mid-Audit
Before an audit goes out to anybody, George said, it goes to a human: him, or usually Jorge, who double-checks the numbers and makes sure it actually makes sense. Chad added a check of his own: that the links aren't hallucinated, pointing to a page where the AI agent thought a page was. Then they tell the agent to go read every comment the human left and fix the issues.
The bigger change is a midsection. They start the audit, then ask the agent for the top 10 to 12 questions to ask the human, because some of what people are doing is on purpose. The answers go back into the system, and then it finishes the audit. George asked: historically, when has an audit ever had a middle point, or human context on what you're telling them not to do? It works, he said, because Sidekick has the process. Since 2013, they've looked at portals hundreds if not thousands of times, so the bots have a plan and a human side to follow.
“There's a new way to use a new set of tools to do the work that is the audit.”
Whether you build it yourself or bring someone in, he said, the audit can become monthly or quarterly, not something that happens every three years when the termites have eaten the foundation. Audits always felt like a website project to him: 90 days of work, then nobody touches it for years. He asked who wouldn't say yes to an audit in one day for under 1,000 bucks instead of five, six, $7,000, and then added, "This isn't a sales pitch." His point was the new way to deliver value.
Can You Just Ask Your Portal What's Wrong?
“I can't wait for the day you can walk into a HubSpot portal and ask it what's wrong with it, and it can tell you in plain language.”
That's been Max's line on this show for years. George's answer: we're there. With Breeze on its own, he said, we're partway there. A human asking Breeze has to know which questions to ask to get the story. You can't just say, tell me everything that's wrong with my portal. But with some ingenuity and historical understanding, he said, you can point a system at a portal and ask for everything you need to know right now.
And for client work, he called the audit a massive part of the context layer, alongside email and transcripts. Map their properties and objects, run the audit, and tie it all back into a second brain wiki for that client.
A Portal Doctor Inside Breeze
Chad's thought while George talked: what if you had a markdown file for custom context and a saved prompt you could give somebody's Breeze? Could Breeze do the audit? That, he said, is closer to where Max wants to be. George's version: ask his own system for a prompt that lets Breeze do the same thing in HubSpot, and have it name the gaps Breeze would have. Chad saw it as a pair, a context file and a prompt, in the librarian methodology he's talked about before.
Chad pointed to the connectors Breeze has now, naming Notion, Google Drive, Slack, Google Docs, Sheets and Slides, Linear, Dropbox and Box, so Breeze can do more long-running work. (See our Breeze Assistant projects update and HubSpot's guide to creating and managing Breeze projects.) The one thing he wasn't sure of: whether Breeze handles the agentic loop well, compacting when its context runs out and keeping a scratch pad. "Hey, Breeze team."
Once it's done once, Chad said, it's repeatable. Put it in a Breeze project with a scheduled Breeze task, and you've got a portal doctor. (Scheduled prompts in Breeze Assistant reached public beta in August 2026.)
Human Context at the Front and in the Middle
Max confessed he had done audits for some side work after all. The big thing for him was context from the humans living in the portal: what do you know it's doing, what do you think it's doing, what should it be doing? A workflow might look simple and still touch something in the outside world, or change a property that triggers another workflow you'd never know about unless you asked. Chad added the middleware question. If there's middleware involved and you can't audit it, you're missing a world of context.
Chad's answer to how you layer that in: a grill-me step first, where the system asks the questions it needs before jumping in. His markdown file is the audit's instructions with no context. Then a 30-minute pre-audit meeting feeds the human context in. His agents could notice the meeting got completed, read the notes, run the audit, and come back with a draft and four questions, then fold the answers into the rest of the audit and change some assumptions.
“You have human context on the front, additional questions in the middle, and maybe asking for more access to middleware or other systems that could be doing work.”
George shared a real example. During one client audit, the system came back with a question: on this meeting date, you said you handled transactions this way, but we found these invoices in HubSpot. Can you explain why they're there? The client's answer: that was just a test, and those can go away. The question tied straight back to the meeting where they named the tool they actually use.
Audit Your Own Portal, One Area a Month
For the closing takeaway, Max said you don't have to be a HubSpot partner or consultant to do an audit. Audit your own portal. A nice little mini audit, maybe monthly, is worth doing.
Chad's version: pick one thing to audit. Workflows, views, reports. George took it further: imagine each month had a couple of areas, workflows and emails this month, segments and views the next. Then it's digestible, and once you're through it, Chad said, it's a lot easier to keep up. George: "Ain't that the truth with life in general?"
If this episode has you eyeing your own portal, pick your first area and work through it. The free HubSpot Portal Audit Checklist gives you a place to start.
Be a happy, helpful, humble human and do some happy HubSpotting.
