Thank you for being here. Yeah. All right, fantastic. We will go ahead and kick off the afternoon, day two, Goldman Sachs Communacopia + Technology Conference. Delighted to have with me on stage, Yamini Rangan, CEO of HubSpot. Thank you for being here again.
Thank you, Gabriela. It's a great conference, and thanks for having me again.
It's absolutely our pleasure. Yamini, I know how customer-obsessed you are. I think that's the best word to describe some of the leadership philosophy that you have at HubSpot. Help us bridge the gap a little bit between what we see on X and some of the feedback we see on LinkedIn about AI and what you're actually hearing in the field from the mid-market customers that are core to the HubSpot base on what the vision is for where AI goes over the next two years.
Okay. Yeah. There's a big wide gap between what is happening on X and what is happening on social posts, as well as what is happening from an AI perspective. I think the fundamental shift that's happening within software is that the prior generation of software, everything that we did was about delivering features that companies then implemented, and then they got their users to adopt it after the training period, and then they got value from software. With AI, you fundamentally shifted that to delivering work for those users. The work could be either a support ticket that you're doing for them, or you are driving pipeline for them by setting up meetings, and that's the quality of the work. So there's a big shift that is happening.
In order to do that, fundamentally, the platform and the product that HubSpot is delivering has transformed, right? It used to be that as a CRM, we had a structured database, and that structured database had companies, contacts, tickets, deals, all of that that we knew before. But now with AI, you can also process all of the unstructured information. This conversation that we're having, emails, texts, Slack, all of that can be in that unstructured conversation. The fundamental thing that we're doing is to build an intelligence layer, and we call that as context. Of course, context is a word that is being used across the industry, but context actually means what is the business information about the business, about the customers within your business, about your pipeline, about your teams? What is that information? That is what is context.
Why is that important with AI? Well, it's important because AI models are trained on everything that is available publicly. Loads and loads of vast amounts of information that's available publicly. The only thing it's not trained on is what's happening within a business, and that is the context of the business. So for us, we're building the context layer and the context graph. What that enables from an AI perspective is to drive the right kinds of outcomes. Of course, HubSpot is focused on go-to-market, marketing, sales, and service outcomes. So we take that context, and we've built a set of domain-specific agents. Our Customer Agent resolves support tickets, Prospecting Agent qualifies the pipeline, our Data Agent enriches contacts and gets an audience ready.
So we take the context that we have built through that intelligence layer, and we enable it through HubSpot agents, as well as custom agents that you can build based on our platform. That is the action layer. So that's the second part of what we deliver. Then the third is really the interface to software, which is, in our case, Breeze Assistant. It is an interaction layer. The pattern here is that you don't have to click on software. You can just have a conversation and have the software. So I think fundamentally, what is happening within software is that you're not delivering features, you're delivering outcomes. In order to do that, we have an agentic customer platform that does that.
One of the debates that we started having yesterday is on how the nature of a marketing stack, and I think this includes B2B marketing, when you include agents, has to be somewhat infinite from a personalization standpoint.
Yep.
Before, if you think about marketing, we were making broad cohort analysis across different types of customers. Now with AI, you can actually be much more granular in that approach. Talk to us a little bit about how you level up, and I think it has something to do with the structured data that you're talking about. How do you level up to be able to give customers a much more granular way to go to market?
Yeah. I think there is a broader change that's happening within marketing. If you looked five years ago, there was a very clear playbook, especially for B2B marketing. It started with your content being searchable on the net. When your content was searchable, you clicked on someone's blue links, and then that brought you to a website. When it got to the website, you captured the email, and then you nurtured it through email campaigns. Those emails were not necessarily personalized. They were much more generic emails that you would sequence, and that's B2B marketing of the last decade. What is specifically changing is that, of course, nobody is clicking bluelinks because AI overviews, AI mode, LLMs have actually replaced all of that by giving answers. First of all, you've lost a pretty big top-of-funnel source, which is just content marketing.
Now, that is replaced by omnichannel marketing. The big shift that's happening in B2B market is that it's not just about the content that you generate. You got to show up in social. YouTube is one of the fastest-growing sources for B2B marketing. Same with TikTok, and same with a lot of these social channels. You have to be now present across social channels. In addition, there is a new source, which is AEO, which is how do you show up in the answers that LLMs are actually generating, and where do you have the share of voice there, and then how do you expand on that? The big playbook change that's happening right now is you've got to be able to enable omnichannel marketing. Within that, there is personalization that you asked.
The question that you were specific about is how can you use personalization? Look, what is happening is that if you use an LLM to generate a generic email and send it to the set of contacts you have, the click rates and the open rates are just going down. There is no room for just AI that is generating emails that is not personalized. It becomes really important to get the intent signals of your buyer, to know what they are interested in, what they are researching, what would actually drive conversion, and use the context that I talked about earlier to generate an email that is very specific to what they are looking to do. That is what drives much better marketing campaigns.
Broadly, to step back, playbooks are changing pretty significantly, and you really have to be omni-channel marketing, and that is the vision that HubSpot has. We want to drive and be the platform for omni-channel marketing for B2B.
Well, let me pick up on the change in strategy that you are talking about, going from SEO to AEO. I want to better understand, going through a transition like that can sometimes be disruptive.
Yeah.
Not just for your customer base, but also for HubSpot's—
Sure.
—pipeline, if that makes sense, customer pipeline. So give us both sides of that coin. How do you help smooth out the disruption for your customers' customers, and then how do you do it for your own business?
Right. I'd say one thing, which is that it is not a simple replacement of SEO with AEO. I think SEO was a broader playbook, and if anything, it is getting replaced by omni-channel marketing as a playbook. You just cannot replace it with AEO, because even though AEO is nascent, the volumes are not there to completely make up for the volume that you lose in SEO. So that's a point I'll make. But having said that, maybe I'll start with HubSpot. In 2022, a pretty significant portion of our top of funnel was generated through content leads. Since then, obviously, it has declined. We've really embraced having a multi-channel outlook in terms of the top of funnel. Back in 2003, sorry, 2023, we acquired a podcast network, and we are the number one podcast network for business in go-to-market.
That has been a source for us. We've expanded pretty significantly in terms of social channel. YouTube, we have 10 different channels within YouTube. They all generate leads significantly, and that has consistently grown since 2023, and the same thing with a number of social channels. In addition, I'll go back to what you said before, which is we've used email personalization in order to drive much more conversion within the top of channel. So from 2022 to now in 2026, the composition of our top of funnel has changed pretty dramatically, and it is now a multi-channel playbook that we leverage in terms of our marketing. Now, that is exactly the playbook as well as the set of products that we are leveraging for our customers, and we're helping them become multi-channel B2B marketeers. Obviously, part of that is providing visibility into social channels. We added TikTok.
That's one of the fastest-growing channels for our customers. We've added Reddit, Instagram, YouTube. All of the visibility across those social channels are important for our customers. At the same time, we also are enabling AEO. Many of you know that in April this year, we launched our AEO product, and the product does three things. One is that it gives you the share of voice that you have across LLMs compared to your competitors. Then it makes recommendations on what you need to do to get your content strategy up. Then it takes those actions. We're one of the very few marketing solutions that can actually do that full cycle between visibility to recommendations to content actions. We're seeing traction within our customer base. Nearly 32% of Marketing Hub Professional plus customers are using our AEO.
We're on this mission to really help expand marketing into an omni-channel marketing strategy, and that's really what we're executing for our customers.
I really appreciate the context of how, over time, the strategy evolves and layers.
Yep.
What I want to understand better is we can get a little bit into some of the near-term headwinds, specifically on your new customer adds number.
Yep.
Do you think that HubSpot as a company is now in a position where the headwind from an evolving SEO to AEO type market, the worst is behind you? Or is this a period of time where it's going to be uneven? Maybe we isolate just that piece, and then we can talk more.
Yeah. I think, specifically in terms of the net adds that we talked about in Q2, I think that there are a number of dynamics. I don't think it would be just fair to say that it is the SEO to AEO—
That's fair.
—transition. I don't think that is the case. As the composition of top of funnel is changing, you might hit air pockets in a quarter where one source is declining and another source is not picking up. But I think overall, we're really comfortable with the broad strategy that we have adopted since 2022, and I think we'll continue to do that from a marketing perspective. I'd say, specifically on net adds, we took some deliberate actions in Q2. When we came into this year, the first quarter was solid, and we looked at our AI adoption within SMB. It's still early days in terms of AI adoption there. We looked at what are the blockers, or what is creating friction in terms of AI adoption? A couple of things. One is that, as we've just been talking about, they're not buying features and evaluating features anymore.
They're actually buying work. A lot of times, they want to make sure that that work works within their environment with their data. What does that actually mean? If we go and tell them that, "Hey, we're going to qualify leads for you," or, "We're going to have support tickets resolved for you," they want to see that with their knowledge base, with their history of support tickets, it's actually going to work in their environment. So one of the things that we did was to increase the trial period and give software during the trial period so that they can make an agentic use case work in their environment. That obviously slows down in top of funnel. It has headwinds in a couple of areas, but we think it's the right motion.
Again, early in a platform shift, the more confidence that you build in terms of your capabilities, the more you use your distribution and reach to really seed the right use cases, you can then compound the adoption later on, and that is exactly what we did, and that obviously has an impact in terms of near-term headwinds.
Remind us, when did you switch to the longer trial?
In April.
Okay, perfect. The longer trials are how long?
In terms of agentic use cases, it's between 14 days to 30 days, depending on—
Okay, sure.
Yeah.
You already have data from those initial—
Yeah.
—cohorts from April and May. Share with us a little bit on how that strategy is working. When you look at the data for the folks that have had longer experimentations,—
Right.
—what does conversion then look like?
Right. I think it is still early days, because I do not think that in one cycle you can look at all the conversions and say that is what you are going to get. I would say that in terms of the number of trials for agentic use cases, that has increased. Conversion really depends on whether the customer has their data already in one place. The level of data and the consolidation of data has to do with the conversion of the trial. If not, our partner ecosystem as well as our teams are helping them get the data in the same place and then driving the conversion of the trial. I do not think that this is an inflection that we see in one quarter.
It is the foundation that we can put in place to make sure that customers get comfort and confidence to be able to do that.
Now, I think the broader question is what are we seeing in terms of agentic use case adoption? We shared a lot of the trends that we are seeing there. I would start with the way we look at it is what percentage of our customer base is using agentic use case at any given point, and we call that the reach. What is the reach of agentic use case? I would say the beginning of the year was about 37%, and now across all of the use cases, it is well over 55%. Specifically within that, if I look at the agent use case, we have a number of HubSpot-driven first-party agents, and that has grown from 9% to almost double that over the past six to seven months. We are seeing the reach of agentic use case increase. Then the question becomes what happens with the depth?
Are they using it every day, frequently, or not at all? That is where a lot of the focus is in terms of our ability to drive adoption. We have seen the depth increase threefold, which means the number of agentic actions our customers have taken has grown over 300% in the same period of time, which is a good leading indicator. Both of these are early leading indicators, then this is what drives the emerging lever for us in terms of credit monetization in the future.
Let us talk about credit monetization.
Sure.
How do you pick the value of a credit?
It really is simple and easy for point solutions. It is harder when you are doing it across a platform. Let me give you a couple of examples of how we think about actual credit utilization. Our Customer Agent is a pretty easy example because it resolves the support tickets. There are a number of companies that are providing Customer Agent, so you can actually look at the market, you can look at the metrics. There are broadly two ways in which we look at this. Many of the companies out there are charging per conversation, some of them are charging per resolution. One of the first decisions we made early this year is to tie it very closely to the outcome that agent delivers, which happens to be the ticket resolution. So that is how we picked it, because it is much more customer friendly.
Then you pick a price point, which is not super competitive. We do not want to be always the top player, but we want to be a premier player in the market, so you pick a number. I think the way we think about agent rate cards depends on the outcomes that we drive, and the number of channels that we drive, and what is happening within the market. That is not the same with a Data Agent, which is enriching records and preparing the audience and the prospect match. That could look very different. But to simplify this back for a customer that is today buying apps. In the past, it was pretty easier. They were able to buy a seat, and now, because software is delivering work, it has to be not tied to the seat, and therefore, there is a credits mechanism that is growing.
The industry now has a lot of conversations about, do you call it tokens? Do you call it usage? Do you call it credits? Do you call it outcomes? We believe that there is just a clear model, which is a hybrid between seats and credits, and credits consumes anything that is usage-based and tied as close to the value, and that is the philosophy that we have, and that is what we are driving in terms of how we position back to our customers.
I want to run a couple of hypotheses by you.
Sure.
On the compare and contrast between larger customers and smaller customers.
Okay.
One school of thought is enterprises have the resource and the talent. If they want to build stuff in-house, they can. If they want to put together complicated marketing stacks on top of Databricks, they can go ahead and do that.
Sure.
SMBs, they really need their trusted front office vendor to guide them a little bit more because they don't have the resources. Therefore, SMBs are more insulated from the DIY threat. Now, the other school of thought is SMBs, much more so than enterprises, are more willing to take risks. They're more willing—
Sure.
—to experiment. They're more willing to try things with new vendors. Do you have a view on what's actually happening in practice?
Yeah. I'll start with your first hypothesis, which is SMBs in general want it to be easy and want it to be fast, and we've seen this consistently throughout our 20-year history. When we got started as an inbound marketing company, it was easy. When we actually moved from that phase one of being a marketing solution to a CRM platform, almost everybody said, "Well, it's going to be really difficult, because why would an SMB now want to buy a platform, and there are other platforms out there and other solutions out there?" I think the approach that we took is to simplify a very complex technology and make it easy for SMBs to adopt. Because a typical company with a 50-person company or a 200-person company, they only have limited number of resources, both people as well as dollars, and they need to get to their growth goals.
We got to be really good at democratizing complex technology and making it available for our customers, and that fueled the last seven years of our growth, where we have gone from a single hub company into a platform company driving multi-hub adoption. We proved that model. The question is: Is that what happens and plays out with an AI? I talk to customers every single day, and the ability for them to assess every new model that happens to come out every week and put together a whole series of agents is quite complex, and that is why we believe that we have a huge opportunity to take complex technology like AI and make it super easy for SMBs to access and get value from, and that is the value proposition that we have, and that resonates within our customer base.
Having said that, I can see why you are asking the flip side of the question, which is DIY. It is pretty easy for people to string together a set of things. I think in the early stages of a platform shift, you will see a lot of fog and noise, and so there will be a level of experimentation. But beyond that phase, we do think that we deliver a value for the SMBs by making everything simple and easy, and that is going to continue to resonate.
Fantastic. Okay. We got a lot of questions on Salesforce and Claudeforce.
Sure.
The question I want to ask you for your business is, what do you think a healthy relationship with a frontier lab company looks like, and is there a scenario where that type of agreement could also make sense for HubSpot?
Well, we've had that connector for over 15 months. We are the first CRM to have a connector with OpenAI that we launched in June 2025. We launched our connector with Anthropic and Claude in July 2025, and we have a connector with Gemini that we launched in September of 2025. We're the first CRM to have connectors across all three frontier models. I can give you data based on the last 15 months of what the interaction patterns look like. Simplistically speaking, you have a conversational interface in any one of these frontier models that is now deeply connected to the go-to-market skills within your applications that enables the end user to just use natural language to ask about trends and also take actions within the application layer. That is exactly what we have seen over the past 15 months in terms of the data.
We now have over 350,000 users using each of these connectors every week, and that has grown. What we see in the data and the patterns of these users are that, one, they're using it for a number of insights. If I'm a campaign manager and I have 10 million contacts and I want to drive a multi-language, multi-region campaign, I'm going to work in HubSpot all day long. But if I am a CMO and I'm preparing for a meeting and I want to create an artifact based on the campaign trends and the pipeline trends that I want to share with someone, then I'm going to use one of these interfaces to actually grab that information.
What we have found in the 15 months of data that we have collected across these connectors is that users of these cohorts engage more, and they use HubSpot more, and the retention is about 12 points higher than anybody else that is not using these interfaces. So there's a lot of dialogue in the industry of, is this good? Who gets the value if you lose the interface layer? The way we think about it is that way back when browsers came on board, you needed to get your applications to show up on those browsers, and you did. Right now there is a new interface layer, which is any one of these conversational interfaces. So your applications actually show up there, and it actually drives better engagement and better retention.
Of course, as models and these layers get better, your job as a company and our job as a company is to provide additional value on top of that, which is exactly the pattern that we're seeing.
This is a really good point. You have 15 months of data on—
Yeah.
—some of these use cases. If you were to think about the sweet spot of customer, let's say 200 employees to 2,000, where historically you've been able to take share from Salesforce. Based on the functionality you already have and what you can see from your market intelligence, what you're hearing from customers,—
Sure.
—it sounds like you already have essentially the same functionality as what Agentforce would offer. Is that connecting the dots too explicitly?
We've had that. I can't speak about a competitor's product, and I'm not going to do a teardown of a product, but I will say that connectors do something very simple. They take the skills within your application and make it available to a new interface, and that is exactly what we have had with thousands and hundreds of thousands of users that are using it. I know that in every one of the conversations I've had today, I've gotten this question. I'm like, "Let me give you the data over the last 15 months of what we have seen," because our product is generally available and being used every single day.
I would love to pick up on the conversation you and I started having on the earnings call.
Sure.
Look, there's a lot of different things happening in terms of buying cycles right now, experimentation, et cetera. There's a lot that HubSpot has already done to drive outcomes in this new business environment. If you were to put on your prognostication hat for the next six, 12 months, how long do you think we're going to be in this type of environment that feels like it's pulling teeth a little bit versus when the momentum actually starts to compound?
Yeah. That's a very difficult question to answer because I don't think that anybody here has a crystal ball. But what I will say, looking across our customer base, is that we have seen some patterns of adoption. I'd say early adopters within the customer base that are driving adoption from experimentation to full-on scale. They turn on every agent HubSpot delivers, they build custom agents on top of HubSpot's platform, and they're consistently beginning to use HubSpot. There's a whole portion of the customer base that is early adopters. I'd say there are what I would refer to from the technology adoption cycle as early majority and late majority. Now, those customers are a bit more cautious, and they are experimenting. They are turning pilots on, and they will continue to move into the spectrum of large-scale AI adoption, but it will take time.
It's not even across the industries that we serve, and it's not even across the segment of customers that we serve. As we look at it, and this is what happens during every platform shift. Across every platform shift, you don't get uniform distribution day one across your entire customer base. You do get pockets that become early adopters, and then you get pockets that are waiting to see where it goes. Our job is to look at that majority of customers that are cautious adopters and say, "What's the friction point? What are the barriers in terms of large-scale adoption of AI within the segment that we serve?" It comes down to a few things. Number one is data. If they have the data consolidated in one place, if they have really high-quality data, then they are faster adopters of AI.
We spend a lot of time with our partner ecosystem to drive that type of data consolidation and cleanliness within the data. The second barrier is really the comfort with the new technology in a platform. "Is this going to work in my environment?" is the number one question we get. This is exactly why we took a number of deliberate actions to drive trials and then to seed those use cases so that the comfort level increases. Once we have been exposed to agentic use cases, the third question is, "Is my cost going to be predictable, and can I get visibility and control over the spend?" There again, we are driving a lot of customer-friendly policies.
Every customer of ours can set thresholds, and they can do it by use case, by the agentic use case, and they can track their credit spend across their different use cases. We have made it pretty friendly to really look at the outcomes that we are delivering to the cost and tying that as close as possible to that. I think every cycle is going to go through this process, and our job is to help navigate the customers, look at the barriers, and remove the friction in terms of AI adoption.
We are big believers in this idea of taking AI and applying it to business-specific contexts.
Yes.
I would love to understand a little bit of the nuts and bolts on how you do that.
Yeah.
What is your internal AI strategy? How do you think about mix of frontier model versus open source model? How do you think about applying SLM to context data in a way that is not a headwind to gross margin long term?
Yeah. That's a great question. Just to peel back the question, how are we thinking about the internal LLM usage, and how does that impact the gross margin of our business? I will say that we've talked about our platform as being an advantage for many years now, and you've asked me many questions about primary colors within platform and how we actually leverage that.
I have them here.
You do. I will bring that back into context, Gabriela, because one of the things about HubSpot is that we have a platform-centric approach to building. What we've done in the last three years is we have created an internal agentic harness for go-to-market context. What does that mean? Lots of words there. Context, everybody understands it is what is specific about a customer's company, business, their end customers, as well as their teams. That is the context graph that we can construct for each of our customers. What we have then been able to do is to actually take the agentic coding available outside and create an internal go-to-market harness that becomes our infrastructure. That means the evaluation loops that we run are much more efficient than off-the-shelf agentic coding platforms that you can build.
That has our libraries, our skills, our connectors, that is optimized for our developer environment. The reason this type of a platform approach is important is that we can be the Switzerland of AI, and we are the Switzerland across LLMs. What that means is for a specific use case, we can optimize the use of an LLM. We might decide that a content use case, there's a certain LLM that performs better, and we are able to switch across LLM. The way we do that is to focus on two things. What is the best cost-optimized approach, and what is the best outcome-optimized approach? So we look at two things across each of the use cases, and we're able to switch. The reason that's important is also why we can optimize gross margins.
Because we have the infrastructure and we have the go-to-market harness that can go across LLMs. We're now able to optimize in terms of gross margins. For example, we've switched to open weight models, and that has brought the cost down across multiple use cases while keeping the quality of the outcomes high. In the past six weeks, there are a number of models that are now at 80% the cost of other inference models, and we've been able to switch that. I think the industry is still very early in the cost optimization phase, across LLMs, but we have now the capabilities and the neutrality across the LLMs to be able to drive that, and that helps us to really change the gross margin trajectory of the business, even though, of course, everybody's going to have some level of pressure because of how the transformation works within AI.
Any clues you can give us into UNBOUND next week that we should be paying attention to?
Well, it is going to be exciting. As many of you know, next week is our conference. We call it UNBOUND because—
Thank you.
—we've gone to the new era from SEO all the way here. It is exciting. It's fully sold out, 15,000 customers, and of course, we have an analyst day. There are probably a few areas, I think, in terms of our product. We talked about the intelligence layer, the action layer, as well as the interface layer, and we are clearly driving that vision of being a system of context as well a system of agentic action, so a number of launches that furthers that vision and accelerates that vision. We talked a little bit about how marketing is changing for B2B, and we think that we have a huge opportunity ahead in terms of being the omni-channel marketing platform for our customers, so we'll share vision in terms of how we are accelerating that.
Of course, things are changing from a core CRM and how users get to interact with CRM, so we'll talk about that. It will be action-packed.
I still remember your Dario interview from last year.
There you go.
Yamini, it's always a pleasure. Please join me in thanking Yamini for her time.
Thank you so much.