The Secret to Unlocking HubSpot's FULL Potential with Claude's Integration

About the HubSpot connector for Claude, we're so glad you're joining us today. My name is Melissa Greco and I lead product marketing for ecosystem products here at HubSpot. We have an incredible hour planned for you and this is going to be more of an interview style, really conversational, interactive, hopefully pretty fun as well. Before we dive in, let me introduce our amazing speakers or panel that's joining me today. First up we have Daniel Jacobsen from our product team. Daniel has been instrumental in bringing this connector to life and knows every detail inside and out. Daniel, want to say hi?

Sure.

Hi everyone. My name is Daniel. I lead the product team for our for HubSpot's developer product group. We're thrilled that you're here to join us today. Very excited to share what we've been working on, but also eager to hear some of the questions you have and hear your reception and response as well.

Exactly. And we also have Chris Riley joining us. Chris leads our developer advocacy and technical content groups here at HubSpot and he's been using this connector in some really creative ways that I think you'll love hearing about. Chris.

Yep. Hello everyone and thanks for joining. Thanks, Melissa. I lead our eight amazing developer advocates here at HubSpot. If you've ever visited the developer YouTube or the developer website or our brand new documentation, then you've intersected with some of the work that we've done. And yeah, I'm really excited to talk about because I use AI every day for coding and then on the other side I use it kind of every day for just like purely kind of content creation. And I'm excited to talk about kind of the intersection of those two, how we use the information to help curate even better technical content for our developer audience.

Awesome. I'm really excited to pick both of your brains, but first here's how today is going to flow. I'm going to start by giving you some context on HubSpot's overall AI strategy, why we've invested in building multiple connectors, not just the cloud one, including why Anthropic was such an important collaboration for us. Then I'll cover some of the essential legal and privacy considerations up front because I know those might be some burning questions for some of you. And after that we're ditching the slides, going full conversation mode. Daniel and Chris are going to walk us through through how to set up the connector, share their favorite prompts, including some new ones from our brand new prompt library for Claude. We'll talk about what's coming on the roadmap for the connector and tackle the questions that we hear most often. And speaking of questions, please use the Q and A feature throughout Embedded in Zoom. We have my teammates Ricky and Farheen managing incoming Q and A, and we'll save some to answer live at the end as well. Hopefully we'll have plenty of time to do that, so ask away and we'll do our best to get to as many as possible. And then also on that note, some quick housekeeping reminders. Your audio, video and chat features will be disabled throughout, but like I said, you can submit your questions through the Q and A function in Zoom. And I know everyone always asks. So yes, this is being recorded and we'll send the recording along with these slides and any other resources via email after the webinar concludes. Let me repeat, this is being recorded and you will receive the recording within 24 hours of the webinar ending. All right, let's get started. So let me set the stage with where HubSpot is heading with AI, because it's important context for understanding why this connector exists in the first place. Here's something that might surprise you. It surprised me when our research team shared it with us, but Nearly half of 200 of our 250,000 HubSpot customers are using multiple large language models or LLMs, as you might have heard these AI tools referred to. So not just one, but multiple in their daily work. Some folks love ChatGPT for certain tasks. Others swear by Claude. Some are experimenting with other tools. I'm actually going to take a second here to get your opinion. You'll see a poll on your screen and I'd love to know which AI tools you're currently using. I know I personally use ChatGPT for deeper research and then take those insights over to Claude to help me refine my marketing, messaging and positioning. So how do you use it? Okay, thank you. Our philosophy is pretty simple. We want to meet you where you already are. Sounds obvious, but HubSpot should be the platform that brings your customer context. All that rich CRM data you've been building to whatever AI tools you're already using in your daily workflows. Sometimes that might be work you're doing in HubSpot, but we also know you're not always doing all your work in HubSpot. You might use ChatGPT's deep research feature when you need comprehensive analysis and you've got 10 minutes to wait for thorough results. But other times maybe you need quick insights right right now in the moment, and that's really where Claude shines. Different tools, different moments, different needs. Our job is to make sure your HubSpot data is accessible across all of those tools. This vision is about putting small and medium sized businesses at the center of the AI ecosystem because you've invested time and resources into building your customer database in HubSpot. And that data should work for you everywhere, not just locked inside of our platform. So that's why you're seeing multiple connectors from us. We launched the ChatGPT connector back in June of this year, and then in July we launched the Claude connector. And HubSpot was actually the first CRM platform to integrate with both of these AI tools. We also have the HubSpot connector for Gemini, currently enclosed Alpha, and each of these serves distinct use cases. But what makes this approach powerful is that HubSpot becomes your unified customer platform, powering AI tools across your entire business. Your sales team might prefer one AI tool marketing team, another support team, something else entirely. And that's totally fine. The customer data remains consistent because it's all coming from HubSpot. Now, I know some of you might be thinking, wait, doesn't HubSpot have its own AI features? Absolutely. We have Breeze Assistant and a whole suite of native AI capabilities and Breeze agents. But here's the thing. We are not precious about where you do your AI work. If you're already comfortable in Claude, if it's already part of your workflow, we want to make that experience amazing rather than force you to rip and replace or switch tools. This is all about choice, flexibility, and meeting the reality of how modern businesses really work. You're using multiple tools, you're experimenting with different approaches to like this emerging AI technology. So we want to support that curiosity, not constrain it. All right, with that context in mind, let me add the legal and privacy considerations that I know are probably on top of most of your minds. I'm going to be really upfront about security and privacy here because these questions always come up in Q and A and I'd rather address them head on right now so we can have a clear conversation and focus on some of the outcomes that you probably want to accomplish. So let's start with the most important thing. What data can Claude actually access through this connector? Well, the connector respects your existing HubSpot permissions. So that means if you're a sales rep who only has access to view your own pipeline in HubSpot, or your own deals, or your own contacts and customers, that's all Claude will see. When you connect, you don't suddenly get access to data you couldn't see before. The connector inherits your user level permissions automatically from HubSpot. Now, specifically, the connector can access these object types that are listed here on this slide, like contacts, deals, tickets, subscriptions, object lists, and object related associations. Here's what CLAUDE cannot access. More importantly, custom sensitive data properties, highly sensitive data fields, or and this is really important because we always get asked this question, custom objects. We're going to talk more about custom objects in the roadmap section, but for now just know that they're not accessible yet. Another question we often get is what happens to your data once it's shared with claude? This is critical, so let me be crystal clear. Anthropic, the company that makes Claude, has a very strong privacy stance. Anthropic does not use the data you share through HubSpot to train its models, except in certain instances that you have control over. So like when you choose to provide feedback through their thumbs up, thumbs down feature, or if you choose to opt into training, you should definitely review Anthropic's terms for your specific plan on their side and setting types just to be sure. But from a HubSpot perspective, data flows through our MCP server or model context protocol server and we respect your data center location. So for example, if your HubSpot data is in Europe, it stays in the EU data centers for processing. Another question we get is around admin controls and honestly this is where you have real power. So HubSpot super admins control how and what in your organization can be connected to claude. As of an update that we actually made to the connector in late August, you can also have you also now have granular scope controls. So that means when you set up the connector, you can choose exactly which object types to share. You don't have to give access to everything. And we'll go through this more. We'll show a demo of actually how to set it up and everything. But real quickly, the setup flow is as a Super admin In HubSpot, you install the connector on the CLAUDE side. It's embedded right there. You can configure which objects are accessible and then explicitly turn access on or off for all users. You can see connection and reconnection activity in your audit log, so there's visibility into who is connecting and when. And actually one important security note about disconnecting, so user level access tokens remain active even after you uninstall the connector. So if for whatever reason you're removing the connector, you'll want to revoke user access first, then uninstall. So this is something to keep in mind for your security protocols. And then last frequently asked question on this topic is around compliance considerations. I am not going to pretend to give you legal advice here. I'm just a pmm. Legal is not my role. What I will say is that if you're subject to gdpr, ccpa, HIPAA or other regulatory frameworks like that, you really need to evaluate how this connector fits within your specific compliance requirements. It's not like a one size fits all statement that I can make. We have a comprehensive knowledge base article that goes much, much deeper into data governance security compliance implications and I also highly recommend reviewing that with your legal and security teams before rolling this out broadly if you operate in that space or have concerns in that general area. The link is in the chat now and we'll share it again at the end and in any follow up resources to that article that I referenced. The bottom line, this connector is built with security in mind. You can control access, permissions are inherited from HubSpot, and Anthropic has a strong privacy commitment. But you really need to do your own due diligence for your specific compliance needs if that's an area of concern for you. All right, any questions so far? Don't forget to submit them and our team will do our best to get you answers through the Q and A function in Zoom. Now let's get to the fun part. Actually hearing from our other speakers and using the connector and seeing it live in action. I'm going to hand it over to Daniel to show you how easy the setup actually is. And I'm going to ditch the slides. All right, so Daniel, I'm going to put you on the spot. One of the things we heard loud and clear from our research and from support tickets and customer feedback is that authentication can really feel intimidating. I just walked through some of the admin data security privacy setup pieces, but yeah, walk us through what the setup actually looks like now, especially with the improvements that your team's made since the July launch. Sure.

I actually just want to take one step back for a second because there were a few questions that came in about like what even is Claude? You know, how should I think about what cloud is? So maybe the easiest way to describe this is if you think about like, you know, Android versus iOS and ChatGPT versus Claude. You know, they're similar in that sort of nature. Like they're, they're built by different companies, but they both have a similar ish feature set today. You know, here I am in the cloud interface. It starts with chat. There's a different set of models that are proprietary to Anthropic, which is the company that makes Claude. So they, they generate these large language models that you can choose from, you can equip it with a set of tools that will go into details, et cetera. But if you're wondering what Claude is, if you know what ChatGPT is, you can think of Claude as, you know, another choice for consumers in this market for, you know, the large language model tooling or chat, you know, chat applications that they might want to use. Going back to the setup experience, I actually want to give a little bit of a historical perspective as well. So Anthropic, the company behind Claude, are the ones who actually created the model context protocol. So if you've seen news or headlines about MCP or model context protocol, that open standard was originally developed by Anthropic. And so the very, very first HubSpot connector we released was only supported in Claude, the desktop application, not the web application. And it was a very technical, complicated setup. You basically needed to create a custom HubSpot application. You need a, to either be a developer or in a developer adjacent role. You needed custom client credentials, you needed to run a server, you needed to do all of these things. So let's talk about how we've made things, you know, hugely simplified since then, starting with discovery. So, you know, if you're in your HubSpot account and you go to the marketplace today, you can access the marketplace from the top. We now have new AI collections or AI Essentials collections that you can navigate to to see how you can get more out of your HubSpot data by using AI native applications. So if I go to browse now, obviously you see the first thing is try the HubSpot connector for cloud. If I go here, it gives me step by step instructions on how to set it up. It's pretty simple. It's actually initiated from Claude. Today we are working on bringing this connector to the HubSpot marketplace directly so you can initiate the installation from the marketplace. But for now, you would start in Claude. You go to add connector or browse connectors, add HubSpot Sign into your desired HubSpot account and then you can begin searching. So let's do that. So I don't have HubSpot installed right now. If I select this search and tools button and I go to add connectors, I'll see HubSpot is right there. All I need to do is hit Connect. It's going to navigate me to the oauth screen. I will select the portal that I wanted to connect to. And when Melissa was talking about scope, granularity and the ability to toggle on which objects you want Claude to have access to, this is what she was referring to. So if you don't select one of these checkboxes, Claude will never have access to any objects of that type, no matter what sort of prompting you do. I also want to reiterate, the only people that can initiate this connection the first time are either super admins or users, generally with elevated Marketplace app install permissions. Usually admins, but not always. And then from here you can, you know, select the objects you want to give access to. For the purposes of demoing, I'm actually going to select everything except for view details about your subscriptions once I hit Connect app. One of the other questions that came in was how do I know that the connector is connected? If you open the Search and tools dialog, you can see now HubSpot shows up as a connector. You can also expand to see the list of tools that have been made available to Claude. So Claude can basically access all of these tools. And then one of the other things that we recently introduced that Claude supports today but ChatGPT does not, they have somewhat different capabilities is the ability to have pre populated prompts directly in the connector itself. So for example, we have this check connector status prompt. This is generally a good thing to run on occasion because we're constantly improving the connector and adding new features by running this, even if you haven't reconnected in a long time, it will let you know, have you actually toggled everything on? Are there more object types that you might be able to give access to that could actually get you more value out of this connector? And so here you can see it's actually getting my user details and seeing what all has been connected. In this case it says I've basically given it access to all CRM object types except for subscriptions, which if you remember what we just went back through, I selected everything except subscriptions and there you go. So it tells you if you do want to add subscriptions. For example, if you need to work with subscription data, here's how you do it. So step by step instructions on what you would actually need to do to reconnect to go and add that additional object type. And again, there will be more object types over time. Melissa mentioned custom objects aren't supported yet, app objects aren't supported yet. We're continuing to work on just improving the capabilities of this connector.

Awesome. And so just to make sure we're crystal clear here and everyone understands, you install it once as a super admin or someone with admin level permissions in HubSpot. You choose what data is accessible and then you turn it on and off for all users on your team, Right?

Yes.

Okay, great.

Sorry, I did forget one part of the demo, actually.

No, go ahead.

Yes, so once an admin has installed it. So if I go back to HubSpot and I go to my connected apps, not everyone else in your organization will be able to use it yet. The other thing that you must do, and it'll be grayed out because we've already done it before, since you can see I've been uninstalling and reinstalling many, many times, you must select give users access, otherwise only admins will have access initially. One of the, you know, the very early pieces of feedback we got from users was I want to, as an admin, be able to try it out, see how it works, see what data is made available, do all of these things without unintentionally giving my entire organization access to this thing. This is what we have today. This is an area where we're still investing a lot in though, is giving more granularity, more, you know, more what's right, more flexibility and what sort of controls that you have, continuing to work on that.

Okay, that's helpful context. And what about those ongoing improvements that you mentioned your team is making? Are those automatically accessible to existing users or what should people do to make sure they're using the latest version of the connector?

Yes, so the, a lot of the improvements are made on the server itself. So the latest version would basically be, you know, go in and reconnect. If new object types are made available, if we make improvements to, you know, how the results are, like results performance or results quality for existing object types, that will basically happen for free. You won't have to do anything. So we are continuing to make those types of improvements. Anything that requires new permissions though, you'll need to go through these steps here, which is basically reconnect.

Okay, that's helpful. And then what about troubleshooting? I know from customer feedback since launching this in July that errors and permissions issues have been a bit of a pain point for people.

Yes. So some of the biggest issues that we've seen are related to access. So because the connection is initiated from within CLAUDE and not from within HubSpot, a lot of Times non admin users or users without Marketplace app install permissions try to install it, being like, hey, HubSpot's available. Can I just do this? And they'll usually get another unauthorized error. It might show up as a 401. We have done some work to improve the error response so that it's clear, oh, you need, you know, a higher level of privilege permissions to actually go and install this thing or reach out to your administrator type of thing. So generally, if users are trying it, or if you're trying it and you get an unauthorized access response, it's because you don't have the necessary installation permissions inside of your HubSpot account. And then, you know, different regions have different data protection requirements, things like that. Every once in a while you might hit an issue because of the location that you're trying to connect from, et cetera. If you do hit those types of issues, definitely reach out through HubSpot support. We'll be happy to work through those issues with you.

Perfect. Okay, so for everyone watching, if you're a super admin, this is actually, I think, pretty straightforward. You install the connector, choose your scopes, give access to your users, remember to like disconnect and reconnect regularly if you want the latest and greatest. But after that, you're up and running. If you're not a super admin, you'll need to ask your HubSpot admin to set this up for you. Now, Chris, I want to turn it to you. Thank you for being here. I want to bring you into the conversation because you've been using this in the wild. Talk to us about what you're actually doing with it.

Yeah, yeah. And it is the wild. But I wouldn't say anything that we do is, is. Is overly too, too. I don't know out there, but just first a little bit. Actually, I did want to take just a second because there I know that as we go through these webinars, sometimes things happen. Really? And how you're going to know if you have the HubSpot connector. Oh, I'm not sharing. Let me share. How you know you're going to have the HubSpot connector is it's going to simply be toggled here once. Once it's on, if that toggle is off, you're not connected. You will click here and it's going to be on a portal by portal basis. I don't. I'm going to just give an explanation of what you're seeing on the screen first before I reshare it. Again, so my team's responsibility is to create content. Now it's not just any content. Like this is not, you know, just kind of marketing content that we throw out there. Like our content has to be best practices driven, it has to be rooted in real world use cases and it has to solve specific problems. Now what's interesting in this sort of nature of content is that the more specific you are, the better. So a lot of times when we think about creating like marketing content, for example, we want to put it out there and like of it inspire everybody. But for us, we're concerned about how do we help people implement solutions and integrate with HubSpot in very specific use cases. It could be programming languages, it could be business use cases. It's solving real world customer problems. And so while I've been using Claude on the content creation side, there's a major step that has to happen first that I think a lot of folks don't consider, which is curation. And so for us, creating content is a lot of times just kind of gut feeling. And yeah, my advocates are amazing. They have a great pulse on what's going on, but it's not always the best gut feeling. Like we maybe could do better at deciding what sort of content we want to create. And so one of the things that we've started doing is making sure that we're bringing these connectors and this technology into our content pipeline at the point of content curation. Now the reason I wanted to give you that whole explanation is the demo data here is not all that interesting. This is actually coming from a portal that if you have visited the developer booth at Inbound, you may have seen app cards or custom cards where you can create a custom card inside of HubSpot for specific use cases. It's something we talk about a lot. And so in this particular demo portal, we have a custom card for assigning out computer hardware. Pretty straightforward. Now the use case I'm talking about is like, how can I look at the data in the CRM to understand how to what I should write about? And the one thing that I want to point to, besides the fact that one of the things I love about our cloud connector is the details you get from how it's pulling data from the CRM. I mean, if you're somebody who's technical, you get full JSON objects, which is so cool. But I'm going to scroll to the very bottom where it actually gave me some recommendations because what I noticed when I did it on this demo portal versus our live portal is that it started going, this doesn't feel like a blog post, I'm not getting enough information. And it kept on going and the one thing that it really jumped out to me is product guide. So I told you that this was a app card that was about assigning hardware to people. And so its suggestion was you should probably write a guide on how to pick hardware. And that is actually really spot on. And we're talking about like a limited subset of data here. So it's a real, real powerful example even with a subset of data. And so that's just one way that we are using it in our day to day flow. And there's, there's many way where we pull this together. I mean for us to get ahead as a team, it's really important for us to be very in tune with our target audience. And that's what the smart CRM is all about. So it's data from the CRM that drives our interaction with the developer community.

I love that. So it's not just data retrieval, it's analysis and prioritization. And like such a good example of using AI as a thought partner. So yeah, okay, these are your personal kind of team based use cases. What about other functions? What are you hearing from customers and partners and kind of seeing out there in the wild?

Yeah, yeah, my team intersects AI, you know, across the board from day to day workflows. I know there was a question about like as an agency, how do I interact with my customers using the cloud connector versus like maybe using our MCP server in your ide, like your development environment, like Claude. So it's, it's all over the place. As you mentioned, prioritization is a big deal. One of the things that I have found surprisingly beneficial is cross functional collaboration. So when we are working with other teams. So for example, we work a lot with your team, Melissa, and we recently had an example where we were taking video transcripts to help better understand quotes and things that people were saying about our development environments. And we needed to share that across teams so that we could have a conversation about specific messaging and quotes and things that we may want to assimilate into a video in the future. And it was very, very helpful because I started to picture how would we do this otherwise we wouldn't really. Actually I don't think we could even be able to do this project. We would just kind of guess or it'd be one person assigned and there wouldn't be as much collaboration cross functionally. So cross functionally. It's been A really big deal. But also in the sample I showed you and I'll and I'll pull it up again quickly is ticket data. So we have something called Feedback Hub inside of our dev site developers.HubSpot.com feedback but also our support team as tickets coming in. And we have, you know, you can see that ticket data is something and again it looked at kind of shipping the devices out to people in this specific example, but ticket data, responding to tickets, curating responses to tickets and analyzing tickets. With our brand new Doc site, we can use it to analyze questions that people are submitting and have a better sense of how do we communicate back to product feedback. So collaboration has been kind of a surprise to me actually.

That's awesome. And I think it really highlights something important that this isn't replacing your work or your team. In HubSpot, you're still taking action, creating tasks, sending emails, updating records, informing, collaborating with teammates. Claude is just giving you the insights to know which is the best action to take or recommendation to make or feedback to prioritize. Kind of like boosting your team's productivity and collaboration with other teams. All right, time for another poll. Hope you're all still out there, but mind telling us which use case relates most to your role and the job you're trying to accomplish here? Like sales, marketing, customer support, service, success, commerce, or if you're still exploring and unsure, that's fine too. Just let us know. Thanks for that. So while you're doing that, we've been talking about prompts and Chris, you've shared some great examples firsthand. But Daniel, I know you and the team have been compiling prompts from users and building something special. So do you want to give this group a tour of our new Claude prompt library?

Sure, let me go ahead and share my screen again. Great. So, you know, while we do have a small subset of prompts built into the connector itself, we have been learning a ton from customers as they give us feedback, as they ask for more features, as they tell us how they're using the connector and what we've done. And what will be made available to all of you is a prompt library, a pretty robust prompt library of more advanced use cases to really help get the conversations started with Claude so you can get a lot more insights and value out of it. So for example, you can filter by content type across commerce, customer success, marketing, sales service. You can dive deeper into very specific use cases on what you're trying to accomplish and go from there. And it's really easy. You can just click to copy the prompt and so I'll go from Find HubSpot companies that have recent high priority tickets within the last 60 days include company details, direct HubSpot links and ticket volumes, priority levels and resolution status. Just paste that into Claude. I can check the connector status. I saw this question come up, but just if for whatever reason you don't see data coming back from HubSpot to check the connector status, select Search and Tools and make sure that it is toggled on. And in this case I'm actually going to say six months because we're using a, a demo portal that doesn't have tons and tons of recent data. So I'll give it a little bit more data to work with and then I will go ahead and click submit. One of the things that's really cool about Claude is a lot of times it'll be a lot more explicit about its thought process. So it'll actually go through step by step on how it is querying Data within your HubSpot account to make sure that it is trying to find the right things. If you're doing things that are more advanced or that might be relying on custom properties, this is where you can usually troubleshoot or find out why you might be getting inaccurate or wrong results. It might have found the wrong custom property, it might not know the names of your custom properties, things like that, but you can actually usually instruct it to actually find the right custom property. So you can be like, hey, actually this is the property you should be looking for and it'll kind of redo the thing and usually actually figure out from that how to proceed with a more accurate and precise result. One of the things we are working on is, you know, instructing it more and trying to teach it more about custom objects and properties so that you don't have to, you know, coach Claude as much. But for now, if you do see issues in the response or unexpected response results, look at how it thought through the problem. Think about how you would, you know, work through the same problem and see if you can instruct it to be a little bit more intelligent. So you can see it's, it's found tickets with associated companies, it's going through and compiling all that. And we'll give it a couple more seconds to finish this. We need a elevated music while it goes. We're trying to do a live show that it's real. We do have backups in case it takes too long.

Yeah, and I wanted to also point out that you customize the prompt so just wanted to point out to people that, you know, these are, these, this prompt library is full of just like thought starters and you can easily customize and adapt to your own specific situation. So meant to spark ideas.

Yes. All right, should be getting close, Getting tickets at this point. So what I'll do, I'll go to my prepared one and then we'll come back to this one at the end because this is going to just keep working in the background. So I ran the same exact prompt previously and I'll show you what it returned. So what it gave me, you can see same prompt except actually did 60 days in this example. It gave me a full list of total high priority tickets, companies, number of companies with high priority tickets, urgent priority tickets, and then high priority tickets. And then it goes through each of the different companies including annual revenue, contact metadata as well as links to tickets. But then one of the like, again, these are really just conversation starters. So this is kind of hard to read. It's just a long list of companies, tickets, revenue, et cetera. So I asked a subsequent question. Create a table and organize them by deal value for each company. And then what it will then do is get a little bit more additional data and it actually prepares this interactive artifact that I can then work with for basically viewing the same data, but in a way that's easier for me to find what I'm looking for. So for example, what I might care most about as a customer success leader or somebody in a service role is, you know, urgent tickets that have the most open pipeline impact. So I can, you know, sort by total deal value or open pipeline or whatever, I can show details and see the high priority tickets and then I can go ahead and actually select this link. So this is a deep link to HubSpot. I'll go from here and then I can find, okay, here's the company and then I can scroll down and see the tickets and actually take action on it. So really the value here is accelerating action back in HubSpot because you can get more useful insights to drive that action. So in this case, my job to be done is, you know, actually go and solve this open pipeline risk because there's a payment processing failure ticket which is really bad. Make sure that that gets acted on as fast as possible and bring those insights right to my, to my face. I'll go back to the one that was running before, make sure it finished. It may have hit a stall. Let's refresh and see if got somewhere and if not we'll just go back to the demo. No, that's all right. That's the demo.

Thank you.

Yeah.

And I also wanted to point out that sometimes certain prompts don't automatically give you the visualization and you have to prompt it to give you like a bar chart or a table or something like that. So I also wanted to point out that you can customize these and you can be more explicit with, with Claude to give you. Because I think that visualization element, instead of just like a laundry list of data that you have to scroll through, is something that's a nice differentiator with Claude. So I just want to point that out. And Chris, have you had a chance to play around with the prompt library or have any thoughts briefly there?

Yeah, I do. And this might be more existential. You did say that we wanted to keep. Make this fun. So we'll. Yeah, we'll get philosophical and I've noticed some of the questions, so I'll try to weave these in. You know, no tool, absolutely no tool is a replacement for good, high quality data. And I think if you went to Inbound this year and you notice like Karen's emphasis on how important data is to us and how important it is for us to enable customers to get high quality data and manage that data, but it's still up to the organization to have high quality data. And so nothing is going to replace that because that will influence the quality of responses. I mean, you think even a person trying to sift through, you know, reams of paper to get an insight like the nature of the data itself will influence the response. Nothing replaces that. Now, the amazing thing about the prompt libraries, and I love this trend of prompt libraries and MCP servers, is that it's a starting point, like you said, like these are de facto, like you have to use them, their guidance, their best practices. There's things that have been tested. What I've really liked, I've seen organizations do, is adopt their own internal prompt library, which is an adaptation potentially of ours, creating their own crowdsourcing. Wow, I haven't used that term in a long time. Crowdsourcing amongst the organization. And that's going to be really important because there is a learning curve here. Some of the questions that have come to end have been the nature of what you ask the system. And what I've noticed is that if you don't know what you want to do with the data, you don't know the answer. That's why you're asking, obviously. But if you don't know what you're going to do with it, you should probably figure that out first. And that's where the prompt library helps in adopting your own. And I've been really passionate about this topic of context engineering because that's where I think all of this is kind of converging is the primary skill set we all need to develop. There were two questions that came across that were about more deterministic things. So like Daniel showed a chart and there was analysis. What AI can do with the cloud connector can do for you is help with that initial analysis and like give you the shape of the response that you're looking for. But if you want repeatable, deterministic things, you should consider workflows. You should consider potentially a custom integration to have what essentially educated. Here's the type of analysis you should do, because here's what I found and take that and make it in a linear, more deterministic way in a workflow is something that you should consider. And I saw two questions related to that. And again, this comes down to knowing what you're going to do with the data. Is this something that I'm going to show up to a meeting and make my team look like an all star one time, or is this something that I need to report on now on a regular basis? Because we just found a needle in the haystack and it was Claude that identified that for us.

Perfect. Okay. Yeah. So everyone feel free to visit the prompt library. Test out the prompts for yourself. The link is in chat. Adapt them to your specific use case, and if you create prompts that work really well for you, please let us know. Share them. We'd love to add community contributed prompts to future versions of our prompt library. I am going to. I think this next section warrants a visual aid, so I'm going to go ahead and share my slides again. But I want to shift gears and talk about what's coming next. I know from talking to customers that there are some limitations people are hitting. I want to be transparent about what's on the roadmap. So I'm going to bring this back. First, let's celebrate what we've recently shipped. So, Daniel, I know this launched in July and your team launched some recent improvements, I think closer to the end of August. So why don't you share what some of those were?

Sure. So some of the recent improvements that you might have seen, we added more object types, especially focused on commerce use cases like subscriptions, carts, orders, et cetera. We added the ability for admins and users with marketplace install permissions. To toggle on what object types are available to the connector. So there's a lot more fine tuned and granular controls and we also support multilingual use cases now as well. So if you're using Claude with a different language, it will work.

Awesome. Those are great improvements. What about custom objects? The big question, I know a lot of customers have brought this one up.

So custom objects is something that we're working on. It's a lot harder of a problem to solve. So when you think of standard object types and how Claude provides useful responses based on standard object types, we can basically provide static instructions to Claude. So we can just say like, hey, here's how you use the contact object and here's how you search for it and here's how you can basically interact with it. With custom objects, we don't have that luxury. And so we need to figure out how do we give Claude enough information about what that contact object, or, sorry, what that custom object represents, how it's used, how it's queried, all of those things which we don't entirely control. And so that's why it's a much trickier problem for us to solve. But it is something that we're working on and so you can expect to see something in 2026 where we'll have more support for those types of use cases.

Awesome. I was going to say. Yeah, more variables and the timing on that I'm sure is everyone's next burning question. So thanks for sharing 2026 other things that are top of mind for your team. It looks like there's a couple more there.

Yes, one of these came in as a question, but soon you will also be able to use Claude to write data back into HubSpot so you can have it update a deal or update a contact or update a company directly from within Claude. Again though, this will always be opt in. So as an admin you may not want to give Claude the ability to mutate or update your data. You have to choose to give it that access, otherwise it won't be able to. So again, that'll be choice based. Also, we will support auditing so that you can basically see what updates the connector has made on your CRM data. So if something does go wrong, you can kind of troubleshoot and figure out how that happened. And similarly, we won't support to delete actions, so you'll be able to do updates and creations, but Claude won't be able to go and just delete a bunch of your CRM data. So we'll we're doing this as careful as we can, but we do know there's a lot of power with it and a lot of people have been asking for it.

Yeah. And I guess on that note as well, can we talk a little bit about current limitations, things we're like not planning to change, especially from like a security perspective?

Yes. So right now you can only have one HubSpot account connected to Claude at a time. If you do want to connect to multiple accounts, the process is to disconnect and then, you know, install the connection again. But in the other account, we know this is a little challenging for agencies that may work with many different HubSpot accounts, but it's a current limitation that we don't have plans to solve anytime soon. And then we don't actually plan to give access to any sort of sensitive data properties to any of these types of connectors just because of data sensitivity, privacy, et cetera.

Makes sense. And Daniel, I'm actually going to keep you, for the sake of time, can you maybe give us, for those more technical folks on the line, like a minute, 30 second technical, except explanation of how the connector kind of works under the hood? I'm sure some folks might want to know those details.

Sure.

So if you're more of a software engineer or in an IT role, I'll give a quick TLDR of how this all works. So behind the scenes, whenever a user installs the connector, we're creating a basically a special account that represents the access that the user has and the app has. So when the admin installs the app, that is basically dictating the access that the app has in the portal. And if a user has a subset of that access, they won't have the full app access. If they have a superset of the access of the app, they still won't have a superset. They'll just have basically what the app has access to. And so that is basically what restricts access to data from HubSpot so that there is no way for a user to be able to get data they personally don't have access to on HubSpot. Similarly, it's using Model Context protocol, which we mentioned earlier, which is an industry standard protocol that's an open standard that basically allows AI, generative AI tools to connect to third party data sources. It's growing pretty rapidly in the industry, and so that's what we're embracing here. But that's the very quick summary.

Yeah, that makes sense. And for folks who want to go deeper on those technical details, we do have a brand new doc site that we launched just last month powered by mintlify. So I think we're going to post in chat here shortly. Just a quick one plug for our MCP docs if you're interested in perusing that technical documentation. And just to be clear from a non technical perspective like myself in the audience, you don't need to understand any of this to use the connector. This is just for people who are curious or need to explain this background context to their technical teams. Okay, so before I open it up for Q and A, let's just do a quick recap. Let me pull up some some final resources here. So here's the link to the prompt library. It was shared in chat. It would also be shared with you after the webinar including the recording. So peruse it, explore it, use it as your starting point. We also have just general information on our landing page here, an overview of features, benefits, getting started information. We mentioned the help article in our knowledge base. So this is your technical setup guide. Troubleshooting resources where you can go deeper on data security and compliance questions that you may have. And if you run into issues, reach out to HubSpot support for questions about Claude specifically Anthropic's Trust center and Help center are really great resources that we recommend to check out as well. And please please share feedback with us. Use the feedback mechanisms in HubSpot, reach out to your account manager or email us directly because your input really does shape our roadmap, as you've heard and as Daniel shared with some of his forward looking statements. Hopefully you're all still with us now for one final poll. How are you planning to put what you've learned today into practice? Will you set up the connector? Explore the prompt library? Take this back to discussion for discussion with your admin or security teams. Or maybe you're already doing all of these things or unsure of your next steps and that's totally okay too. Maybe need a little more research there. Let us know what you're thinking with this quick check in. And while you're doing that, finally one quick final thank you to Chris and Daniel for joining me today and engaging in conversation to our team behind the scenes, helping to support and answer questions. Again you'll receive a recording of the webinar and all the resources. Now let's see what is up in Q A and I can stop sharing.

There's a lot out then Q A. Yeah, I, I every time go ahead do one of these. It's jump in. It's A big measure of the quality of, of a webinar and the interest the amount of questions we get. So the questions are amazing and know that we like we can't answer all of them right now and we'll do our best and I think some of them we've kind of answered in a more broad kind of context. One of the interesting questions that came across from Ben was around and I think that this is something that's interesting for everybody is is context and like the, the size of query you're sending, how long it takes to get a response, the amount of tokens that are being used and this all fits in the area of context engineering. And I think that one thing that's really cool is that as your team gets better with the prompts library and so forth, you can have kind of a layered structure of prom prompts where you have a set of prompts that are kind of like pure discovery. Like I don't even know what I don't know. Tell me that. And then you can have a next layer of specificity. It's the specificity, especially with these MCP driven tools that is going to give you faster responses and fewer tokens essentially because if you're there's something out there called a rag which is just an intelligent way to access a database and the way for that to be the most efficient is telling the RAG directly where to go. So for example, if you wanted to look at a specific property on contacts, then you tell it to do that. I know that. Another question and Daniel started explaining this with behind the scenes was like how is this all built? One of the cool things about the connector and if you are technical, if you watch oh it goes through and makes these assumptions or starts to bring the data in, you'll see the full JSON objects there and you can really kind of reverse engineer in your mind what it is asking HubSpot for to get its answer. And that actually can help you optimize your prompts. You can start to determine like oh, I probably. I know it's always going to be this property in context so I'm just going to tell it to get it specifically.

Yeah, I'm seeing a couple of other interesting questions come in actually Daniel around how about reverting changes? I know right now you can't take action back in HubSpot through Claude, but it's something like write functionality is something that your team's thinking about in the product roadmap. So obviously that brings up questions around like yeah, creating chaos with Koernix writing back to HubSpot.

Yeah, it's a great question. It's something that we've explored in terms of supporting Write Actions. Data reversion probably won't be supported in our very initial release of Write Actions. However, auditing will be. So you will be able to see what updates were made by the connector to revert those changes. It will be manual at first. We do know, you know, being able to revert potentially disruptive changes is something that folks may want. I think the thing that's a little unknown is how. What is the frequency of unintentional changes as a result of using Cloud with write access? You know, these models keep getting better and better over time and it might actually be that, you know, 99.9% of the time it's actually fine. In which case, you know, data reversion tooling is expensive and hard to build if. If it becomes an issue. It's something that we'll definitely invest in, but it's kind of an unknown right now.

There's also another question here, kind of on the product roadmap topic as well, around the ChatGPT connector. What's the current state of that ChatGPT connector compared to a CLAUDE connector? What's the. The plan or improvements to be implemented? I think in general, just working with you and your team directly, Daniel, the goal is to constantly kind of strike that balance of trying to keep parity between the two so people have access to the same types of objects and functionality and whatnot. So I don't know if you want to add to that.

Yep, so today they are a little bit different. It's mostly by virtue of just how fast the technology is moving. And so the version of the connector we built for ChatGPT is a little bit different because of the support that. That ChatGPT had. That said goal, long term, is that the underlying, you know, connector server is the same for all of these things. And we will build additional, you know, ChatGPT or Claude or, you know, Gemini or whatever specific functionality that can make it even better in those products. So, for example, Claude supports prompts. ChatGPT does not today. So we will do those things for Claude where and when it makes sense. But in terms of object access, in terms of read and write capabilities, in terms of, you know, granular scopes and permissions and all of that, the goal is that that's basically the same across all of them in the long term.

There's a question here that just like the product marketer in me is excited to answer, can you get insights into campaign data through this connection? So for example, if a contact filled out a form or anything like that. Yeah, so any, any Data that's in HubSpot that's a part of that object list that's available through Claude, you can, you can query. So something that I personally really like doing. Thanks Dan.

I'm going to share my screen too. We did another one where you can basically see like you know, revenue generation and contact creation as a result of First Touch and Last Touch campaigns. There are some prompts in the prompt library that give examples of how those can work.

Yeah, the prompt library is helpful to answer that question as well. Something that I personally like doing is like looking back at the past quarter of tactics that are being tracked in HubSpot and understand like what was our highest ROI generating tactic to help inform when I'm like budget planning for next quarter or campaign planning for more effectiveness in my campaigns. So yeah, you can definitely, you can track those things and query Claude through your HubSpot campaign data.

One question I see. What's a good way to get CLAUDE to count the total number of records in HubSpot for a specific object or to provide the fill rate for a custom property for like data quality audit use cases? I'd encourage you if you tried those a while ago and haven't tried more recently. That is an area that we did struggle with initially. Any sort of analytical query we did not perform well when we first launched. So it's like count the number of objects with this creation range or whatever. Those would struggle. We've made a lot of improvements to our backend to improve the response quality of those and if you still have issues with those, please send feedback with the prompt as an example. This is something that we can and will consistently try to improve. So thanks for that question. Let us know if you have more feedback when you're trying this.

This is a question we always get. Can you talk more about how this conflicts or complements Breeze since that's our native AI assistant and agent technology. I know Chris, you usually have a decent answer for this one if you want to take a stab.

Well all things AI I think my responses to the stuff change you know the well first of all right tool for the job, right? Like when you are day to day inside of HubSpot you're avoiding context switching. You don't want to go off on a tangent in like you know where you're doing pure discovery and it's a part of your workflow. Like Breeze is the place to be like Breeze is built for your day to day HubSpot interactions. Breeze has some additional information about the system that the other connectors will not have. Because keep in mind it's not just the information that's in a record, it's all the metadata around how the system is set up that's important. And it's also tapped into other things, for example our knowledge base, developer docs, etc. Now one of the benefits of the connectors is you can bring in not just HubSpot, you can bring in other tools in your day to day productivity stack and get information from that in augment it. For example, we have inside of Google Drive we have guidelines for creating blog posts. I can bring those as a part of the curation process so that Claude knows that when I curate blog posts based on CRM data, I wanted to also use context of how we want topics to be written about and how we title them and how we create abstracts and so forth. So there is no perfect answer. The thing I think that a lot of times happens with AI is like we think it's going to be a magic bullet for everything. It's not like you still have to decide the right tool for the job and the tool is not going to like change everything for you. You have to understand how you're going to use this is true for project management, it's true for everything. And so I think it's great as a discovery tool, all of these. But please spend some time saying this is how we're going to use it inside of our organization and kind of have anointed practices and use cases because that'll help you to distinguish like day to day. Let's use Breeze like you're in the system. It's right there. Let's use that. Did I answer it?

Yeah, I think that's helpful.

Is that consistent?

Anything to add, Daniel? No, I think that's okay. Something else that's like coming up I'm noticing is kind of like reverse engineering the connector. So like are there best practices for information that we should get better about like from a hygiene perspective in HubSpot to ensure that we're getting the most out of the connector. I don't know if there are thoughts there. I feel like that's a little bit more.

I mean I. All I'll say is I. And not just with Claude like leveraging AI, leveraging cursor, leveraging our MCP server to look at the structure of data inside the CRM to inform things like. So Daniel mentioned getting Access to custom objects is coming. Custom objects is such a critical aspect having HubSpot grow with you as a business. But creating custom objects in deliberately and in a smart way is very, very important. Like for example, please don't create a custom object that has properties that mimic properties that are the same as properties that already exist inside of HubSpot. You can actually use AI to help you decide what sort of custom objects to create based on data you've collected historically. And that's one of the like the coolest cases I've seen. And yeah I think Melissa there, there is the like going the other way. Like tell me how we're currently doing this and what could we do better with our data and to I mentioned about data quality. You can even use it to help inform better data quality.

All right, anything else you guys are seeing that you want to make sure we address before with like the one or two minutes we have left here?

Not, not specifically. I do want to go back to the kind of the deterministic aspect of things and just say that like if you, I won't give you a bunch of my history but in college I was working on a specific type of technology that was basically an over engineering of everything that you could just use statistics for. If there is something that is straightforward, linear, deterministic and you can create a workflow for it, I would do that. Like you don't need to use the AI tools to solve every single problem. Like think about the types of problems you're going to solve. And I think definitely in the use cases of like I don't even know what, I don't know that's where these tools are just unbelievably powerful and make you shine as an all star. If it's something that you can literally do account and stats with, then maybe go and do that instead.

Perfect. And then a question that I'm seeing here, I've seen a few things come up around like Gemini Copilot, other AI tools. Like I said we're in closed alpha right now with a connector for Gemini and we're also working on a co pilot connector as well. So yeah, keep the feedback coming, let us know what you use in your daily work and we're going to try to do our best to ensure that your HubSpot data is accessible wherever you're getting your work done. And with that I think we will close it out and thank you all for your time and we'll be in touch with those follow up resources in this recording. Thanks everyone.