316: The Practical Guide to AI in HubSpot

Hi everyone, welcome to HubShots Episode 316. This is the practical guide to AI in HubSpot. In this episode we're going to talk about AI approach, AI basics, some of the tools, and we're going to start with a discussion around AI and how it's evolved in the last year. We actually started this episode a year ago, we're going to talk about that. My name is Ian Jacob and with me is Craig Belly from Zen, and I'm from Sergerby Foud. Craig, how are you? Really well, and yeah, you're right. We started planning this a year ago and we kind of kept putting it off. I think why? We were just kind of underwhelmed with AI where it was promising a lot, but not perhaps delivering enough, and it was changing so rapidly, whereas things have kind of settled down now. So why don't we talk about the AI approach, Craig, because things have changed massively in the last year. I think a lot of the things that they spoke about a year ago have finally come to fruition. We've been on that journey with HubSpot, and I think even without HubSpot, all the other tools that we use in our day to day. So let's talk about AI approach and some of the expectations. I think you're right to highlight expectations. And by the way, listeners, just so you know, in this first shot, Ian and I are going to chat for a couple of minutes, just more around opinions and high level theory as opposed to the tools themselves. So if you're not interested in that jump ahead to shot two, because from shot two onwards, we're just talking about the features in HubSpot. But expectations, you're right. The start of the year, people will like, oh, FOMO, real hunger for AI show me a lot of interest. Get to the end of the year. They're almost jaded. There's so much AI slop. Social feed is just full of AI. It's kind of like, oh, is there any business value? Plus, I think people have been burnt enough by going to chat GPT and saying, hey, what workflows should I set up? What dashboard should I create? What lifecycle stages should I do? And getting bad advice and implementing it into HubSpot. Now they're like, oh, wow, I'm not sure what's going on. So there's a weariness around HubSpot. Well, maybe around HubSpot around AI. And people are asking, is it actually providing benefit? Not combined with the onslaught of updates. They're kind of what do I choose? How do I do it? And so there is a real need to be careful and selective in the AI. They put in place. But then there's also governance. And what are we talking about when we use the term governance? So when we're talking governance, we're talking about access for individuals, data access. We're talking about how to use things, what to share, what not to share. Policies around change management and how they utilize information. And I think these are actually very fascinating because I feel like if you're a smaller business, some of these things you just think don't apply. But I still think you need to be aware and understand what data you're sharing. I think when you're in a bigger organization, there is so much of protected data. Data is everywhere, right? So generally I find small to many businesses. They have their data in a few places. But they can generally the tools that are connected, then comes up to, well, what do we do? And then I think there's scrutiny around security in place with the larger businesses. And so this is all that drives governance and having some sort of policy and framework in place to do that. Totally great. And you went in the word policy. I'm hearing that from mid-sized businesses now, whereas previously it was just the enterprises. True. And at the start of the year, I remember big and small companies were handing out chat GPT accounts and basically saying, hey, pilot, you know the word pilot, go on pilot and see what kind of AI is for getting it's thrown to the wall and actually sticks. And we were seeing big businesses just kind of say, oh yeah, we connected a chat GPT account that people play. And now they've wise up and thought, oh, hang on, when we connected it, we're actually letting anyone access just about anything through chat GPT. And so they're locking it not all down again, which is good. So governance is rolling out, but it does sometimes limit things because before people were like, oh, yeah, I'll just have a play with an AI tool or feature. And now sometimes it's locked down on the flip side. Let's just talk about the great divide. And there's all the laggards that haven't yet embraced AI. They foolish, well, maybe, maybe not. Lots of people are in jobs, especially knowledge workers, they're just going to get their work done. They don't have time to play with AI. And you'll see these people and we might even think this is kind of unbelievable given the work we're in and our clients, but it's actually commonplace. People are forwarding on videos to me going, this is incredible. I'm like, you know, that's AI and they're what? Or they're just discovering that they can summarize an email reply or get help preparing emails. And they're like, okay, so we've all got to realize that just because where maybe using AI a lot more than others doesn't necessarily mean it's the norm, that combined with the fact that there is governance and things getting locked down, often people want to play with AI in their workplace and they're not allowed to anymore. So there's this whole kind of malaise of AI unknowns that people are grappling with. And so I think as we just before we get any of the features, let's just talk about the path forward. I mean, what should we do? We all want to be AI first companies. What's the kind of approach we can maybe consider and what's fascinating and we went to a, I'm going to throw this in here, can AI event, which I think they're holding quite regularly here in Sydney. It was fascinating to see that the people that were there, the different viewpoints. And I think I quite enjoyed having a different perspective because I think we live in a very hubspot-centric world or maybe Google-centric world, etc. When we do paid advertising and how AI is affecting that, etc. And so on. So I think one of the things where you would suggest is choose a scenario in your daily process that's manual and consuming you time that you want to focus on and that can actually solve a problem for you, set time aside each week or each day to train yourself with all of this is about getting good and understanding and testing new things. One of the things I remember hearing in you to head to when we're on at that event was a founder of, I think it was the founder of, was it relevant, say I? And he was talking about one of his discoveries was in some incremental change. He had a really good analogy of a car. And a brake, how previously people that didn't drive when they didn't have antilog brakes, you know, they had to press the brake pedal to the point where they felt it lock up. And then he's off the brake, right? And he kind of described AI a bit like that. It's not this incremental pushing of the brake or acceleration. It is like you've got to go, where's that limit till something, like until you kind of come to a grinding halt, but then kind of ease back to find that that position. And I feel like sometimes you've got to do that. And I feel the more immersed you become or the more things you try, the better you will gain knowledge and understanding of what's actually going on. And as the funny thing is, bottles keep changing every month. And so even that you just see such the, the not even incremental, it's exponential features and things that it's that can be done. So I think think about that and we'll touch some of this stuff as we're going through the scenario. And then the final things would be check your HubSpot updates regularly and return to tools that you've tried previously that you might have been a little bit so say about and have a look and try them again. I think one of the things here is to don't be afraid to go back and try what you tried before. Because I think you might be pleasantly surprised. I think they're good tips. I want to pick up on one of them though that you choose a scenario that actually solves a problem. I think a lot of the mistakes that businesses have made with their pilots is they get outside kind of suggestions on what to work on. And so they're not doing these AI pilots that the people doing it don't really see any value on. And it's like, oh wow, that's nice. Wow, fun. It's not actually solving a problem. So pick something that actually does solve a problem. I'll give you an example. You know the company research agent. We're going to talk about this a bit later in the show where you can basically use AI to go off and research a company. Now if you're in sales preparing for a sales call, that's really valuable. But for just about everyone else in the company, there's very little value. And so you get these kind of pilots who are like, oh wow, check this out. And it is pretty amazing. It goes off and does all this research and does in a minute or two what used to take hours, days or weeks, right? So you look at it and you go, oh, that's amazing. But if that's not actually solving a problem for you, then what's the value? It's fun and you know, it gets touted as this amazing feature, but has it actually given you business value? That's what we've got to ask. And so that's why we always go back. Does it solve a problem in your business, in your process? So pick one, think about it and focus on it. And to your point about checking a HubSpot updates regularly, without filter in mind, when you go through the updates, you're going to hang on, this is the problem I've got. Does this actually help? Let's think about a common business problem. Oh, I get 10 lots of meetings. And then I've got a right notes afterwards. Okay, that's time consuming. Oh, I have someone to help me. Oh, that's a business problem. Okay. AI is the obvious answer there. You have a meeting recorder come in, record the call, use AI to summarize, pull out pain points, action items, etc. Perfect use case. Okay. So then when you're looking at the HubSpot updates and you see the. meeting recorder beta, you're going, right, that's actually solving a business problem. We will focus on that and we will basically ignore many of the others. So that's the way to think about it. And to your point around returning to the calls, absolute, to the beta features, the tools, absolutely because they are improving a lot. For example, the meeting recorder, we haven't switched to the Hubswatt one entirely. We still use a VIMO. And I know a lot of companies use others because the meeting record is not up there. It will be soon. And so we keep checking and our goal is to just use the Hubswatt one. Same for other tools. You go back to them and you realize, wow, they've improved a lot. Sometimes just in the space of weeks. So keep going there. And that brings us to this idea of betas because don't you remember the good old days in when enabling betas was just something that the geeks and birds could look after? Yes. And this uses as a cycle, well, we don't need to look after betas. That's, you know, that's the future. Now I say to everyone, you've pretty much got to be installing some of the betas. Be across the product updates in Hubswatt, selectively, strategically, enabling many of them and many of them are betas. And well, while they call betas, well, just because they're optional really, maybe there's a level of quality control that hasn't been applied yet, but really you've got to be enabling them. The platforms we use, Hubswatt of course, being the main one, just get more and more complex. Combined with AI being such a turning point in the industry, of course, AI features are getting added to Hubswatt and it's becoming more and more complex. And in some ways, the irony is people are so confused and overwhelmed by the tool that Hubswatt then says, oh, well, here's some AI support to help you work out how to use all the damn AI features in the product. And it just strikes me that AI was supposed to make our lives easier. Ian, but it just seems to make them more complex. Not necessarily a bad thing, but just an observation. All right. Enough of the theory. Should we dive into the actual features now, Ian? Yes, let's do Craig. And let's go into the AI basics. Now, the first thing is data enrichment, right? Now, this is something that, as of inbound, or let's say as of September 2025, is now included in every Hubswatt portal. The enrichment pulls from data inside Hubswatt's data set for contacts and companies. What I do like about it is that you've got some really cool things in there currently, is to what you want to enrich and then even things like out of office, if people get replies for out of office, etc. It'll keep that in there. I have seen Craig this week a large number of enrichment data deletion requests come through on different portals, and I don't know whether you've experienced any. I've never seen them before. Okay. And so they're coming through. So my take on that is that people requesting must be somewhere in Hubswatt that any enrichment or data for that contact doesn't get done and see it's basically, there is a whole separate section that looks with data request and it'll say that it needs to be, so you basically review it, press the button, remove the information that was enriched by Hubswatt. Interesting. I've seen that across a few portals this week, which has kind of been surprising for me because I've had to do a deep dive and go, hang on, why are we getting these all of the sudden? And this is what's happening, right? Data enrichment happening across the platform, which is then opening up other doors for other things that need to be done. So if you're not using it, use it inside Hubswatt, scan your portal, it'll tell you, now in Australia, we don't have a great contact data enrichment. Companies are not so bad. I said getting better by the day or by the week. So definitely be using that to help you with some of that data that you could utilize in your Hubswatt. And I'll tell you why this is important. If you're looking at segmenting well, and it had a lot of conversation with people around segments and about email opens and about how they communicate with people, if you're enriching your data, you can actually segment well if the data is there, right? And so think of this as another way to slice and dice your information and get a better result. Totally agree. Industry group on the company levels, the one that comes to mind. It's really quite often it's there and can be used for segmentation. But as you said, data enrichment is now included with paid plans. So it doesn't consume credits, but it brings us to the topic of credits and the fact that most of the AI well useful AI features in Hubswatt are starting to consume credits. And if they aren't yet, because they're still in beta, then they probably will in the future. Fortunately, Hubswatt does have a reasonably good way to manage your credits or see how your credits are being used. And there should be settings that allow you to limit it. We've talked about that previously. We won't go into all of that now, but you can control to make sure your credits don't kind of blow over and push you to another tier. That control is with you. And you will see little icons next to some of the features throughout Hubswatt that say whether they use credits or not. But just be aware, this is another reason why you should come back and check or sometimes they don't consume credits currently because they're in beta, but they will in the future. So potentially some surprises there. Anything you want to say about credits in? It keeps getting better in terms of the management of it. So on the dashboard, when you look at the features, you can click through to those features and see on a daily basis what the credit consumption is. So you can understand I have been caught out with the intent feature and credits, credit consumption. And it's been an interesting few weeks with that and just understanding how it works. And I think the tools are getting better, but just be aware, I'm always now on the back foot. Well, when I say it, it probably should be on the front foot. We credit utilization and how it consumes it, especially around intent and adding companies and so on. And what you do because you can run things in work for actions that can all of a sudden chew up all of your credits. So just be aware, know your numbers, know what's going on. And I think you should be right. Yeah, absolutely right. We've put a screenshot in the show notes. So is this one of your clients and they've racked up 3000 credits just on in the intent? Yes, correct. So yeah, it is easy to run away at times. Make sure you have that option to cap your credit so it doesn't blow over. All right. Let's keep it going. Because we're still just in AI basics, inbox automation, the legacy way through your inbox connection, you can just turn it on. But if you've got data enrichment and a new conversational enrichment beta, I think it's still in beta. I'm not sure if that's rolled out yet, but you can turn that on. It's part of your data enrichment settings. You mentioned this earlier. It's got the return to office date detection. It's quite nice. Looks for that just in email replies or signatures and can put that against the contacts. So that's good. That in place, what else do they got? AI summaries. They're kind of left and center throughout all the objects in HubSpot. I don't think those AI summaries consume credits, do they? No, they don't. They're just, yeah, seeing that quick summary. I think one thing I want to tell people and you'll see wherever these AI summaries is that are in the help desk, is it on the contact company, etc. If it's not right, I would encourage people to put the thumbs down on the summary because that is indication back to their system or to the AI that this is not right. The better we can have this feedback loop, the better the system will become. If you see something, do something is my little motto at the minute and just make your life better and make everyone else's life better. Look at that often. I agree. I also like how the summaries include the sources. Yes. So you can just go, what? What do I get this for? I'll click to, oh, that's where I got it. Correct. All right. Next to mention, we're just talking about the basics here still is breeze assistance. So just the assistant at the top menu, by you can pop that down. It's a little chat interface that appears from the side. Yes. Later, we'll talk about you can choose which assistant because there's the default breeze assistant, but you can actually plug in your own assistance from the HubSpot Marketplace. And it's really got access to all your data. You just chat away and talk. I have to minute, I don't use it much. Do you use this much, Ian? Not as often as I would like to crave. And I guess I was going to ask you the question because you know, you've got the breeze app on your phone. This will be very similar to that. Would it not? Yeah, it is. It's the same. The breeze is the app on your phone basically taps into the same thing because you do connect your portal and stuff like that. Yes. Yeah. I think I, I don't know. It's hidden mess. I think it will get better. I know those tools that I've used and just had a really lackluster experience and kind of ignored. Yeah. And I need to train myself to come back because it does keep getting better and better. It does. You almost need this indicator that says, hey, Craig, I know I was really terrible before, but give me a ton of go. But yeah, it is getting better. But I tend to use some of the other tools. Correct. And I think these are all incremental Craig. Like if, when I was talking to some of the product managers when I was away at Inbound, for example, there was no memory. So you'd start a chat. You'd kind of come back or you'd ask a following question. It was, it's kind of lost all sort of idea of what it was doing. Now that's much better now. And so I think. like we were discussing. Incidentally, day after day, week after week, it just keeps getting better. And maybe what you should consider doing is have one prompt that you run and look at the output, just do it every week and compare the output, right? And then you might discover something and then maybe make it two prompts with the same kind of in the same line to just see is the memory holding in place. I mean, one thing I do like is you can add your apps in there, like your Gmail, your Google Drive, et cetera. So you can look at the data from multiple places. And I think that's where we've been one of the biggest things. Because I know like when I use Claude, like I've got it connected to HubSpot of it, it can't just all my, all of my Google Drive, et cetera. And so when I talk to it, I kind of like say, hey, tell me about this because we keep documents about stuff that we do with customers. And so all of that together is a really good place. It kind of brings all that data in together into the one place and makes it able to use. So what I have seen Craig on lots of things is that you'll see it says at the bottom of all of these, Windows, he says AI, Generate Content, maybe inaccurate. And I notice that's appearing everywhere so that people are aware that there are inaccuracies. And I think this comes back to what we discussed earlier. Is that you still have to know what's right and wrong. Do not assume that the AI knows everything and what you're getting is actually correct. I think this is the biggest thing that I have discovered with people. And this is what drives AI software. When you know better, you will just take everything like it's gold, so be aware of that. - It does rise another point. And we probably should have addressed this earlier in shot one where it was a bit more kind of theory and opinion, but I saw an update recently, I blocked about it. But this idea, what's an AI first company or whatever the buzzword is, right? - Yes, yes. - And it's really an AI first company has AI first people in it. Okay, so what's an AI first person? An AI first person is someone that has the expertise and experience to know if the output they get from AI is wrong. So the ability to fact check or error check what AI gives you. And it sounds obvious and yet think about all the people around you that just accept whatever AI says that face on you, oh yeah, that's good. Plunk that out there. That's actually becoming the norm. And it's rare that people actually sit and go, hang on, is this correct? And they error check it. That's a skill. So cultivate that skill and error check the outputs you get. And that's really the secret to successfully using AI. - You know what I saw today, Craig? When I was training a team, you know, in the hubs, what AI, when you can ask it to rewrite, make things shorter, et cetera, there was an option today for proofreading. I was like, oh, that's interesting. So that's just appeared there now. Like what do you want to do? - You know how you can highlight some text. And then the AI thing comes up. I was like, oh, that's new. Anyway, so what does it actually do though? What does proofreading do? - Yeah, so I think it just gives you suggestions if it's grammatically correct. - Oh, okay, okay. - Yeah. You know, I've kind of said before and I'm going off and around here. I've said before that AI is kind of like the spell check of this decade, it's kind of like, of course it's got AI. Of course it does spell check. - Yeah. - The thing is the difference as someone pointed out is that spell check checks your work. So you create the answer and then it checks whether it's actually spelled correctly. But with AI, it's the other way around. AI produces the output and it's up to us to actually check the AI output, whether it's correct. So this whole AI generated content may be inaccurate warnings that we're getting everywhere. It's kind of like, of course, but apparently that's, we decided that that's acceptable. And also shows you the pressure that companies are under. Like HubSpot could have waited a year to roll out some of these tools when they were finally mature and providing value, but no, they've the pressure of the market and just the pace of change means they need to be rolling it out. And that's why there's so many beaters. And unfortunately that's the way the world at the moment. It seemed to accept it, don't fight it. All right, let's move on. HubSpot marketing feature, we're going to talk about the buyer intent tool, but here's the question, is this actually AI? Or is it just that it's one of our favorite tools? (laughs) - Well, there are AI parts in a Craig. And so that's why we're talking about it. And so we talk about marketing, I think this is one of the big things. We're taking away. We're telling why we love it so much. - Yeah, well we do love it. And there is AI to be fair. And there's a new signal set. But what is it, first of all, for those who haven't used it and lots of people haven't used it? Really get onto this one, I think, especially marketers with sales reps, actually love it as well. What does it do? It just tells you which companies are visiting your website. Not the contacts, it's at the company level and you'll get this list. Are these companies visiting your website? And they had three visits and this many page views. You're like, okay, wow. That's actually really useful, but it gets better because you can filter it based on things like intent. Just show me the companies that had certain characteristics that are located here or this industry, et cetera. They use this technology, that's a good one. What about topics they might be researching or while the pages they visited, did they visit the pricing page instead of say the careers page. That'll very quickly show you intent. And you can look in time frames and whether they're already in your poll, et cetera. So this, I think, is one of the best tools that HubSpot released this year. I'm telling all our clients to use it and when they see it, they love it, especially when they can say, ah, the intent pricing page, show me anyone that's visited the pricing page. Now, there is a research bit as well. You can kind of research and filter down from all HubSpots database or data set of companies as well. But the real, I think, AI part of it is around signals. And this is a new tab that's appeared in the buyer intent tool, which basically provides signals about a company. What is amazing is that there are constantly new things really out for intent signals. And so you'll see, they've nicely labelled it in here Craig, so you can see what signals there are. And so let me just talk about some of them. There's signals around executive hiring, funding, geographic expansion, growth metrics. Layoffs is a good one. Leadership content, merges and acquisitions, product launches, strategic partnerships, technology, investment. And then you've obviously got research intent and so visiting intent, job intent. And then there's stuff to do with email balances. But there are a lot of intense signals and this keeps getting bigger by the month, if I feel, as there's more intense signals. And I think this is the thing that we have to be aware that there is so much data out there that they're able to do these kinds of things and utilize it to do things better. So just be aware of that. So I would be checking your intent signals on a weekly or fortnightly basis to see, is there something new? And as I have discovered with some of the things, as you start to track some of these things using AI and consuming your credits, there are some intent signals you might have turned on at the start, you might want to switch those off as you continue on your path to understand exactly what is important to you. Now a lot of us with B2B, we're not probably looking for people with funding depending on the sectors we're in. So we might turn that off because we're not into that. So you might need to figure out what's working for you and what's important before you track them. - Yeah, so with this signals tab, you basically select which companies you want to be alerted to some of these signal changes. And yeah, I think often people want, "Ah, tell me if there's job changes happening." Now I don't know where they get it and maybe they're scrolling LinkedIn or other tools. But I can see why this has a lot of potential. I will say though, I think it's your mileage may vary. I think it's a lot stronger for US companies than in Australia. And also I've found that even things like just company news or financial general meetings and announcements aren't actually picked up often by this signals tool. So I can see the potentials there but it has been hit and miss. And I know that also because some of the very companies that I'm tracking with signals and it's giving me, oh no, signals detected. Some of those that when I do run them through company research agent, it will pull back details. Oh, we found this about the company or it didn't release this annual report or there has been this. So there must be different AI tools or different data sets or things that it's sources that it's looking at. So just be mindful of that. Your mileage may vary. And but I can see where it's headed. Get this. I've just, while we're recording, I have just in the assistant I've said where it is, helps well get this data from right? Yeah. With intent and intent signals. Oh, what it is. It says, helps with a global intent ecosystem research intent, helps with a proprietary network of 200,000 plus websites that have the helps for tracking code installed and have opted in to share intent data. Wow. There you go. We don't know anonymous IP visits any site of the ecosystem. Pubs for identifying the visiting IP as a company using reverse IP. Matches the visit with one or more research topics on the website. And that's what powers the research intent. Brian, okay, boss. All right. What's next crank? Hubswart AI marketplace. AI agents and assistants. So let's understand, or we talk about AI, there's two kinds of things inside Hubswart. There's assistants and there are agents. Agents generally are doing a particular, what I call it, a particular task or series of tasks and getting your result. Assistant is there to help you through the process like think of it like chat GPT where you're asking questions and you're getting feedback. Or it's being able to answer questions specifically to your business because there's a, I think I saw an FAQ and a onboarding assistant possibly where that you can utilize. So like you've said, assistants you manually interact, agents can be called, most exciting part, it can be called from automation workflow to autonomously do things. So that's the exciting part about the agent. Yeah, I just think it's worth highlighting that the word agent means many things to many people. Even on the Hubswart marketplace, it's got a drop down that says agent type. And you know one of the entries under the agent type is agents and the other type is assistants. So it's like, well hang on. One of the agent types is assistant like so I can understand why people are confused, right? But yeah, I think you've got an assistance. You'd basically chat with them, agents, you set them up to be called autonomously, often with a bit of an input injection of like company name, et cetera. But yeah, let's look at the Hubswart marketplace for agents and we're just the normal Hubswart marketplace where you go for all the other integration bits and even where you go for themes and things like that. They've got a section for breeze agents. Yeah, so let's talk about some of them. Craig, so we've already discussed, oh, you do heard us, we talk about customer agent. There's the data agent, there is. Company research agent, yeah. Yep, that's right. There's a video clip research agent now, which I've just discovered there in the screenshot. You've got a blog research agent. So there's lots, right? And they're rolling out. If you see a beta beside it, probably a good time to go try it because it does not consume credits. If it doesn't have that, it is consuming credits. It does take a little bit of time to set up but well worth it to set up and actually test it out. And then it also has assistance. Like I was discussing earlier, there's an internal FAQ assistant. There's a, what else is there? Is a brand assistant? I haven't tried that one. Yeah. There's a sales coach assistant, which is interesting. And there's an ICP assistant. So there's a few of them. You can create your own assistance within HubSpot as well. They're kind of like baby versions of when you'd have your own GPTs and things like that. So you'd chat with them. I've got this idea I want to walk about. And it's almost like just seeding the prompt for a, you could be with GPT and say, "I want you to act like my sales coach." Yeah. And use these things. That's what the assistant just primes it for that. Actually, when you use install their agent and then you go into set it up, you will actually see the prompt that is in there. And it's a good way to learn and see what's working and what's being asked of the agent or the assistant to give you what you need. So well worth looking at that. Well, that's good because that brings us to custom agents. And this is, I recommend this as a start. Pick one of the agents from the marketplace, install it, and that'll end up in your wallet. And they call it breeze studio or whatever it is. And up there, you can then clone it or you can edit it and then save it as a new agent. And that's really the way to get started. We'll often do that. We'll take the cross-sell agent. We've tweaked it, added our own prompts to it and then called it our own. What do we call it? Customer Excel agent or something like that? We call it in our own portal. So yeah, that's agents. And yeah, as you said, assistants end up in the breeze assistant that you take from the top menu. You can just switch between the breeze assistant and any of your assistants that you have configured. And then finally, I think it's worth calling out. There are a few of the agents in there that are kind of deeply integrated throughout the HubSpot platform. Customer agents, probably the best known. And we'll talk about that a little bit, but there's also data agent and prospecting agent. They are agents, but they've been more deeply integrated throughout the portal. Correct. And I'll probably add to that, Craig, the deal, the sage, and I think it's another one. Is that more deeply integrated? It still just looks across deals, doesn't it? It's more just a prompt. No, it's actually more than a prompt. So I've been using it a bit with different people. So we'll add that in anyway. So yeah, there is. And so just be aware that this is there. Excellent. And not be afraid to try it out. Let's talk about AI and workflow automations this week. So AI and automation. So there are two main ways that you can use your agents in an automation. So you can automate a section within. You can use it from the biggest section in an agent, or from a workflow, right? And this is new. While we do this episode, I connected Claude to our portal. So I went through the whole process. It's actually very interesting because I've been using Claude and connecting to HubSpot to query things, right? And now I was connecting Claude from HubSpot to go the other way to run web factions in Claude. So that was an interesting thing. And we'll come up with this. So Craig, do you want to just talk about some of the things you can do in the AI action set? Yeah. In a workflow, there's actions. You trigger a workflow, and then you can fire off actions. And over the last year, HubSpot's been adding to a list of AI-specific actions. So they've had things like infer company value proposition, or an ICP, or research company news has been an AI action. Summarize a record that's been an AI action. And then the one you mentioned, use a custom LLM. You can call, you know, chat GPT, or Claude, or Gemini, or whatever, from an action within a workflow. We'll come back to you calling a custom LLM. But the big unlock, the one that's really just rolled out recently, is run agent, the run agent action. And what it means is from a workflow, you can call any of those agents that you've installed from the HubSpot marketplace, or that you've customized into your own custom agents. This is the big unlock. Because previously, we'd have to go into the agent, as you said earlier, go into the configure tab, then run a trigger from the agent type. But now you just set it up in a workflow. This is much more the way to go. And this is a good example of something that I think should have been in HubSpot six months ago. This is what everyone was asking. When HubSpot marketplace rolled out, "Oh, he's our marketplace for agents." The obvious question from everyone was, "Oh, how do I call them from a workflow?" The answer was, "You can't." "Oh, that's odd, but now you can." And so it's good that we end on this kind of high for the year with running agents. And so what we've done in the show notes, I've actually walked through a complete scenario. - So this is a real life example, people? - Of calling an agent. But there was actually that data agent one, is one that's often overlooked as well, and you can call a data agent custom prompt. - Yes. - That's been available for a little while. - Correct. - That's one example of that. We actually just call that with a prompt to summarize playbooks. So whenever someone submits a playbook or logs a playbook against the contact company, we just summarize it with the data agent. But let's talk about running an agent as part of a workflow. Pretty easy to set up. We've got screenshots in the show notes or in our course in the HubSpot community. You basically decide which agent you're going to call. Then you can input the tokens that needs, whether let's say in this example, I'm running it on a contact. I'm calling the customer up cell agent. So it needs to know which is the company. So you call the associated company record. You insert the company record ID into the agent action in the workflow. You save it. And then after you've saved that action, you just add another action in the workflow to add a note against the contact and simply pull the results from the agent run in the action before. It's very straightforward once you've set it up. And so here's an example. Again, this is all in the show notes for screenshots. We basically are just starting to create a repeating task against contacts that repeats every three months. And the task has a name of up cell in it. And when that's due, the due date for that task, that triggers a workflow that basically calls the action to run the customer up cell agent, analyzes the customer across all their notes and everything like that, emails, meeting recordings and transcriptions and always notes and then leaves a note against the contact. So then when I get the task, hey, it's time for a three-monthly review of the client, it's propped there and it puts a full summary. I've got a full high res version in the show notes as well. It even tells me the email I should be sending client. We never use that. I'll just say it's that typical chat GPT kind of email. I'd never use it, but it's good to have some suggestions. But the key thing about this that I find is that it often reminds me of previous topics, problems and questions that have been raised and I've totally forgotten. This is the power of these kind of AI tools. They troll back through all the notes, all the meetings, all the call transcripts and they pull out problems that were raised a while ago that everyone's forgotten about. And then you can highlight that in a call or with an update with a client and find out if it's prior to your not. Sometimes these AI notes and that can be pretty long so you can change your prompts to focus on brevity, et cetera, but that's a real kind of use case for how this can be used. You can imagine putting this in place for sales teams every three months, the task, hey, time to talk about upsell and it pre-filled those notes for you. So there you go. You'd be doing similar things I'd imagine, Ian. - Correct, Craig. And before I could do one, I just want to make it tell people that this does consume credits. So just be aware when you are calling. But then I always come back to this, right? Yes, it is consuming credits, but how much time and value is it adding to the individual and the team, right? That you don't have to do that. You don't have to manually do that. And I think that's the big unlock here when we're talking to teams that, hey, you can really focus on the really high value things and we can get the AI to do all of these pre-work items so that you're prepared really well when you're talking to a customer, right? And I think when you look at it from that perspective, like you can go, man, that's really amazing. What you can actually get achieved and get done. I think don't be afraid. The credits are credit, right? Like you can buy more, you can do more, but I think where is that exponential thing that you can do with workflows and AI actions in them? So I'd be thinking about that. And a good place to start would be thinking, even if you did one a month, Craig, can you imagine at the end of a year where you would be inside what you're doing with HubSpot? You'd be in a very different place to where you are today. Yeah, so when it comes to credit, so I think this is worth highlighting. When it comes to credits, an action like this, like is so valuable, I'm not saying 10 credits, I think. It's not like it's showing up, amounts of it. So when it comes to credits, you look at something like this, this is no brainer. I am prepared for the call that I'm going to make here with a whole bunch of upsell pain points, reminders about things, like that makes it a very valuable conversation for your client. Yeah, absolutely. So a lot of out of you, but on the flip side of credits, the problem people have with credits is not things like this, oh, it costs 10 credits, no problem, that's a no brainer. The problem they have is what you alluded to earlier. I can't control it. I am not sure if it's suddenly going to run away from me. I need more guardrails to protect me from myself, to protect me from my lack of understanding on this incredibly complex platform that you've made over the last year, especially by rolling out tons of features that I don't understand, and then me falling into the trap of suddenly running a workflow that consumes all my credits or blows out. That's what people are afraid of. But the value I think people get, so I really wish HubSpotter had a better way to kind of, I don't know, God, you know when you run into a workflow and you kick off or you change the criteria for a workflow and it suddenly pops up and says, hey, you're about to enroll 6,000 contacts. Do you want to just add them all or only from now on? So we need a similar thing that pops up and says, hey, do you realize you're about to consume 20,000 credits? Correct. And you go, whoa, no, thank you for allowing me, but there's none of that in place here. Correct. It's kind of like the pre-checks before you enable something, right? And it's going to be a problem because people will get caught. There's going to be this pushback and anger and awareness and they will miss out on all the value that they could have been getting, just because they've gotten burnt. All right, onto some of the CRM AI features, Craig. HubSpot's goal is to have a CRM that is updated, right? Where you don't have to update and that's really the key things. So smart properties is the correct way to do that. Now, when this got released, no one had any idea how to use that. And I think we have just been playing around and working things out. And so two kind of things that we often use it for is like to have a smart property that pulls our pain points from call transcripts or might be to do some research. So for example, one that I shared earlier was you might want to go and look at a company website or their documents that says when their financial year is and when they're likely to get money or when their new financial year starts. So you can have a conversation with them and see what needs to be done. So you can run a prompt to say, hey, go look at all the documents and figure out when that financial year is starting. So you can let me know and put that in a smart property. So that's one less thing you have to go and find out. Which would have been manual work before, but no, you don't have to do that. So there are two key things that you can utilize. - All right, just when it comes to smart properties, this is confusing because let's say you add a smart property that does that. Look at call transcripts and pull out pain points. - Yep. - So that's set as a property. And so you're thinking, well, why does that just kind of automatically fill? I've got 20,000 contacts. Is that something I'm going to fill for 20,000 contacts? Because then you go, hang on, this consumes credits. What's going on? And this is the important thing to know. Basically, you have to manually say, fill a smart property. - Yes. - And so there's a couple of ways you can do it individually on a record or in bulk. You can select, say from, let's say you'd create a smart property on a company record. You can select the companies that you want and then you're just going to go fill the smart properties and you'll do it and it'll consume credits for each of those. And you can automate it from data agent as well. But just for people to realize, because I even remember, didn't we talk about this on a show? - We did. - Earlier in the year. And I was totally confused. I was like, it's smart property. And it was all in beta and we're like, oh well, at least we're going to get charged. But that was it. And it turns out I didn't understand it properly. You actually have to manually trigger filling up. - Yes. And so I think the thing to note here is the property is where the data is bigger stored. And the data agent is the one that's going and doing the fetching of the data for you. And that's where the credit usage takes place. It's not because you're happy to have a smart property that you're getting charged. It's the data agent running to get the information. - Exactly right. - And it does vary. There are certain things on a company record and a contact record that it will and will not do. So you might go into, you might start to create a property and you say, oh, I want the call transcript. And it needs to be on the, I think it's available in the company or the contact. I've forgotten which one. - That's right, yeah, yeah. - And so then you're like, hang on, that option's not available. So you just need to be aware that depending on the object, it's currently it only works on company and contact objects. And I'm sure smart properties will rely on other things, but there's the only two things that they're working on at the minute. So. - I'll be interested to see, 'cause I don't think anyone I've come across uses this unless we've prompted them about it. - It seems to be the great online smart properties. People are confused, I don't know how to use them. And I can see why I confuse me as well. By the way, how many smart properties can you have in, 'cause I got confused about this as well? There is a limit. - There is, you like correct? - Yeah. There's a limit on things like calculated properties based on your hubs subscription, but smart properties, no limit because they consume credits. That's right. - That's sort of heavily go through. - We'll happily let you create smart properties. - Yeah. - All right. Let's talk about our service hub AI item. And this is the customer agent. We discussed this is probably, it started out in the service area, and it's still under the service menu as we record this, but the customer agent is probably one of our most favorite agents within HubSpot. And so if you're thinking about utilizing it, you should be because it is kind of like the AI chat flow within HubSpot that uses the LLM to, with all your data to answer questions. So I think it started out as just being a support resource. Now it can even do sales and do anything through the customer lifecycle. And what is actually very interesting in real life use here is that customers have picked up where the agent gets to a particular point, maybe can't answer a particular question. And if the handoff is done correctly, a couple of customers have made sales from it, which has been really cool. But the big thing that I want to highlight to people is that the knowledge gaps that it highlights. And why is this important? I'll tell you why, because you very quickly figure out how people interacting with it, what are they asking it? What do we not answer? So this has led to many of our customers creating short answers, fixing our data content on their site, answering questions, that people are asking and making a better experience for people. So I think I told you this last time, Craig was like, instead of waiting, a lot of people have this fear of, oh, I don't have the information, I'm going to wait, I'm going to see what's going to happen. Most people have enough information. If they actually feed at the right information, it doesn't have to be website pages or have to be a knowledge. Like a lot of customers don't have knowledge bases that we help, but they have lots of information on their website. They have lots of information in documents. So we can feed all of that information to get what we need, right? And so I think don't wait. I would be one of those people that says, hey, use this, figure out where the gaps are and keep iterating and building, because that's the power of this customer agent that we're running. Totally great. One of our favourites for the year, customer agent, keeps getting better. And new things keep rolling out all the time with the way things get handed off. The information can look and update. but that's the other thing So it can actually verify with the contact it's talking to via their email. And you can tell it can it updates and properties. So if, like for example, we don't have the company name and we want to figure out what company they were for, we can actually get that information and we give the agent authorities to actually update that property. So those are the little things that are rolling out. And then there are other things which we haven't done where you could like, let's, this is the most common scenario. I say, I could be talking to our agent and say, how I'm looking for my order. When is it going to be delivered? So if the system can connect via the API and collect the order information with an order number, you can say, oh, yeah, it's going to be delivered on this particular day. So that's the other thing that you can do in the customer agent, which is quite exciting. That's right. And logistics companies come to mind. I think they've integrated their logistics backend systems into the customer agent and first providing that beautiful experience. So test the customer agent. That's by, that's why I am advice. Yeah. And by the way, it also, you can connect as a channel, for example, chatting with it or messaging or WhatsApp or even email. But I've, they've added voice now as well. Is that better? Or I don't know if that's live. It is a beta craze. Yeah, it's in beta. It sounds now for us Aussies here. It sounds very American. But surprisingly, like I was testing with the customer vows and I was surprised at the way it worked. So yeah, you know what, it's only going to get better. Like this is very exciting. What's going to happen? So test, it's like what we should always say, right? Test and measure and see what's happening. And I think if that's more true even today than ever before. All right. We're in the home stretch now and we're just talks about AI integrations. We've mainly talked about AI within HubSpot. And we did a load to this earlier, you can call your own tools from HubSpot. For example, chat GPT or Claude Gemma, you can call that from HubSpot. And you can also do the reverse. You can call HubSpot from your own AI tools. So first of all, if you're calling AI tools from within HubSpot, you're basically going to do that from workflow actions. It's been loaded to earlier and there's a bunch of LLM options in the HubSpot marketplace, Gemma and I open AI and Thropic, of course, even XI. I don't know. Well, GROC's pretty good actually. It is. See how that's going. Yeah. And then I said, I don't know what coheres, but there you go. And I'm sure there'll be others. I was surprised. There wasn't Mistral and a few others in there as well, but I guess they're probably coming. And you've got a screenshot in here of, let's say you're calling Claude, you've got to choose the model. And I think the important thing to highlight here is you set this up in a workflow to call out to whatever the external tool is. Yes. So you're setting the Claude in this instance. Yeah. You're actually setting the model that it's going to use. And this creates a bit of a maintenance task because let's say they sun set a model. Remember Chatchee PT? Just sun set a whole bunch of their earlier models. If that was in your workflow before, then it was going to happen. I just just stopped working. So there's actually a maintenance overhead here because if you start using this in a bunch of workflows, then you want to update the model to something new. I think there's no order way to do that at the moment. I don't think in any of these models, there's an option that says just use the most recent or the most advanced. So there's a maintenance issue there. But what's been your experience with these? And you found it a useful option. I haven't fully tested it as yet. I mean, the middle of testing Craig, but I just discovered this and I was like, I'll hang on a second. He's going to be some maintenance. But you know what? Just as everything gets better, I'm pretty sure that you know, use the latest model light turn up inside. Now, I guess even with that, and I'll, this is the caveat, right? Like I was doing something similar to this, but on the other side. And I happened to use a new model that rolled out in Claude. And I ran something, I think I was putting together a, what do you call it, a case study. And it chewed up all my credits for the day. And then I was like, at the end of the case study, after it done all the work, it's when I tried to refine it for using another prompt. It said, oh, you run out of your credits for the day. Sorry, credits. I'm so credit so Claude. No, Claude. Claude, yeah. Yeah. And so then I then I realized the model I ran was the latest, which happened to be the most expensive at the time. Oh, of course. And I didn't even research. I'm like, oh, I just chewed all my credits and like a couple of problems. I'm like, oh, okay. So then made me aware of the different models, but it's good at, and you know, et cetera. So I think like all of this, like you learn as you travel the road, right? You're like, you get to places and you're like, oh, I shouldn't have done that. Oh, this is what I need to be aware of. And I think at no point would you ever figure that out by reading documentation, it's by playing with the tool that you actually figure that out. So definitely, I think what we're going to see is there's going to be lots of education that's going to need to be done and awareness around what's running. And I think regular reviews is the other thing that's going to have to take place with this. Well, I'm just glad that when it ran out of Claude credits, it didn't immediately bump you into another Claude tier and charge you more than. That's right. All right. Well, the flip side of calling your tools from within HubSpot is calling HubSpot from your AR tool. So there's been the chat GPT connector, the core Gemini, there's a bunch of others that connect into HubSpot. Chat GPT I think was maybe six months ago, but I think Gemini one was only recent. Yes, it was. So I asked you pretty new. And this just means you can connect in, let's say chat GPT, you connect it so it's talking to HubSpot. You can from chat GPT or I know you do this from Claude and you're calling into HubSpot for a to pull data out of HubSpot and for you to use. Now, I think the really the only comment we need to make here is where we started, almost at the top of the show around privacy and governance. Yes. Because when you connect in you only get access to what you can see, but quite often the people playing with these tools are super admins, connecting them in. And if you share that chat GPT instance, so with other team members and that's been connected with your own security permissions, then they may get access to stuff they shouldn't be saying simply because you have access to it. But on top of that, there's well, if the tools pull out that data, what do they do with it? And you've got some points here about how they retain it. Yeah. And so I put this in here Craig because we had a conversation with the enterprise customer and they were talking about what's the retention policy. Does it use my data to train the model? And so why I put this in the notes is just to be aware that a lot of the, they will all have policies. So in this case, we're just going to talk about OpenAI. And so they were saying, you know, if you're on enterprise teams education plan, the data is not used for training, consistent with their terms. And it does say if you want a pro pro plus plan, which is a customer consumer tier, you have the option to opt out, which means that you're automatically opted in for model training on your chat GPT settings. So it says it'll use it to improve connector performance, etc. And retrieval accuracy. You might not want to do that. You might want to turn that off. And then it talks about its data retention, poll, practices. So just be aware that like they've, they've all got information. They've stipulated it pretty clearly. And I think this is really awareness piece around what's happening with your data. Most say if you're enterprise, this is very critical. And I think it's, yeah, it's really important because lots of companies have sensitive data. They don't want it being sucked out into other systems. Correct. So it makes sense. By the way, I'm glad to, I wonder if Facebook will ever allow there, there, whatever their things called to connect to other tools because meta AI is it? Yeah. Yeah. Whatever it is. Because I would never connect anything from Facebook into HubSpot because no matter what they said, oh yeah, I would never trust Facebook. I'm sure you connect it and they would just suck everything they possibly could. Permission. Anyway, moving on. All right. Under some list of feedback, Craig. Actually, let's just do this first. Yeah, let's just finish off. All right. So various incremental tools. We've covered everything in HubSpot, you know. But as we know, HubSpot is touching everything in AI. Sorry, every feature in HubSpot is being touched by AI. From segments to work, flow as reports, even on views, Craig. It's pretty amazing in terms of what you can do, even on a view and so on. I think what you want to be doing is understanding, getting guidance and just using things inside HubSpot and just being aware what you can do because it just gets better. I think you've got to keep playing with it and just expect the results to be underwhelming at first, but they will get better. For example, chatting earlier today, someone they're not really sure, you know, customer reports. It's all very confusing. So they just use the AI prompt, hey, here's the kind of report I want. And it just created absolute rubbish like just let them down a rabbit hole of frustration. And I'm like, yeah, see that's, I can see where they're trying to get to. Just prompt what you want the report to do and we'll work it out. Well, that's good if it does and sometimes it does, but often it doesn't and that's frustrating. So take it, what's the word set your expectations low for these some of these things, but it's getting better. It's getting better. All right, Lister feedback. Actually, there's further reading. There is a McKinsey state of AI report if you're interested in learning or reading more about it. I think you put a couple of articles about everyone in Seattle, hey, I. What's really funny, this is, I know we're going off topic. And by the way, that's the show folks in terms of all the features we've covered all that. We were just finishing off a few comments, but in Seattle, you know, Microsoft and that, they're kind of forced to use these AI tools. And this article about Microsoft employees being forced to use Copa and everything improves that they're using AI tools and that. And if they're not, they'll often lose their jobs. And it's like, well, everyone hates AI because they're forced to use all this stuff that's not fully baked yet. So listen, is if you're getting frustrated with AI, just realize that you have options, whereas many people, it's kind of like their life. Their livelihood or their job depends on it. How frustrating would that be? Embrace where strategically beneficial is the summary. If you do want to learn more, I encourage you to go, we've got a HubSpot AI course, which is bottled off this episode inside the HubShots community. Go there, don't have to do everything. Choose the path that you're interested in. Implement one thing, that's by thing. What is the one thing you can implement? Craig and myself are in the community and so a lot of the team and so you'll be able to get some answers there. There are lots of great resources out there, so but I think with the biggest thing with all of these is, don't be afraid to test and measure what you're doing. And then keep it already and making it better. I think that's really the tip. And talking about tips, if you have not signed up for the daily tip, now would be a good time to do so. And that is delivered every day in the inbox with one tip that will make a big difference to your HubSpot journey. So I encourage you to subscribe to that and then you get it straight from Craig and then you can reply. And if you have any questions, let us know. We will be happy to help you out. And finally, the HubShots framework. If you're feeling overwhelmed with HubSpot, this is a good way to a, know where you are and be get on track. And you can utilize this. Share with your team, print it on a big A3 paper, fill it out and make it a living breathing document. Any last words, Craig? It's been a big episode. We've covered mostly IIO features. There's plenty more that we could dive into in minute detail. We do have our updates that matter. Of course, in the HubShots community as well, where Rosyn and other members of the team dig into very specific areas. I also check that out as well. Well, until next time listeners, catch you later, Craig. Catch you later, Ian. Hey there. Thanks for the senior this episode of HubShots. Together, latest show notes, HubSpot tips and marketing resources sign up at hubshots.com. You can also book time with us to help you grow better with HubSpot. [MUSIC PLAYING]