AI Email That Converts—Without Sounding Like a Robot | INBOUND 2025

Hello, everyone, and welcome to this session. AI email that converts without sounding like a robot. I want to get started today with you getting to know a little more about me and me getting to know a little bit more about you all as well. I'm Ari. I lead global prospect email at HubSpot. And over the past year, my team has sent, well, way too many emails to count. Truly a lot of email at HubSpot across five languages with one goal driving demand without losing the human touch. And I am also quite a fan of art history. I am that nerd where every new city that I visit, I want to check out the museum. And I don't just check out the museum. I tend to spend five, six hours there. So some people like traveling with me and some people don't. But that brings me to how I get to know a little more about you all today. So I promise this isn't just about my fascination with art history. We'll bring it back to marketing email, but in the meantime, welcome to audience participation part one today. There they are. Sorry, just noticed that. All right, don't worry. This is going to be easy. You have three paintings up on the screen in front of you. I want you to pick your favorite. There is no right answer, I promise. But out of these three, we've got Sunrise by Monet, a Portrait of Abraham Lincoln, and Salvador Dali's the Persistence of Memory. I want you to raise your hand for your favorite. Who is picking Monet? Oh, I saw a quick hand here. Wow. Lots Monet fans love it. All right, what about the Portrait of Abraham Lincoln? Yeah, cool. Awesome. Love to see it. And finally, Salvador Dali's the Persistence of Memory. Cool, cool, cool. Now, some of you might be saying, but, Ari, where's the Jackson Pollock? That's what I want to vote for. Well, I picked these three paintings in particular because they all emerged, give or take, a few years after photography was invented. They were a reaction to photography. And in the early 19th century, this invention shook the world. I mean, imagine that for a moment, never seeing a photograph in your life. Now you have thousands, millions on your phone. And artists at the time, they had a strong reaction to this. They worried it would put them out of business. After all, the job of the artists was to depict the world around them. They painted people, important events, historical events, current events, et cetera. And now photography came around, and it could capture that in a matter of seconds or hours. At the time, so many people believed that this would make painters obsolete. Did it? Well, fast forward a few centuries and we are talking about the same thing with AI and marketing. We're asking the same question. After all, our job as marketers has been writing copy to convert audiences at scale. And now AI can do this in a matter of seconds. And it's getting better every single day. So what do we need marketers for now? AI can do a lot of those core skills and it continues to get better. Well, let's look at how the artists responded. You've already seen it. They adapted. Instead of trying to paint everything with detailed perfection for what the human eye could see, they changed their style and instead focused on capturing light or fleeting moment. That's how impressionism came to be. They also figured out they could make a more efficient process by using photographs as models. That's the example of Abraham Lincoln here. Instead of Abraham Lincoln needing to pose for hours to get his portrait, I mean, he's a busy man. He got his photograph taken and then a portrait was made from that. And finally they turned to styles that you couldn't photograph, like dreams. Hence surrealism. The point is, their craft could be deployed in many ways. Marketing is facing a similar moment to photography's impact on painters in the early 19th century. There is a new player in the game. It's AI. And we need to adapt our strategies and grow with this to new technology. Impressionism is one of the most beloved movements in art history. It's time to make it this moment for marketing and this new marketing era. This session is about how we evolve and actually a lot of inbound is about that. We heard about it this morning in the Spotlight. You're probably attending a lot of other sessions today about AI, search, et cetera. So today we are discussing how we can use it to create better conversations and better emails at scale. AI is fundamentally changing the email automation playbook. To ignore it is to be left behind, but to harness it will become your competitive advantage. And yes, I'm going to share what's worked, what's failed, sometimes embarrassingly, and how we're consistently seeing 2x conversion gains using a new approach to email marketing. Here's our agenda for today. Let's get started. First, we want to be on the same page for what we mean about AI powered email. What I mean by that is we are talking about using AI personalization to send unique onetoone emails at scale to our audiences. I am not talking about using AI to write a first draft, but then editing it, changing it, and then hitting publish. I am not talking about using AI to brainstorm campaigns or using it for user Research like analyzing chat transcripts, call transcripts, et cetera. Those are all great use cases for AI, as well as getting new recipes or a workout routine, et cetera. We know there are a lot of use cases, but today is going to be focused on driving those personalized experiences at scale without needing to make changes before you hit publish. So let's look at an example together. Here's an example of an email from yesteryear or like last year, it's got a generic opening referencing the person's most recent interaction with HubSpot in this case, and then tying into a vague value proposition with a proposal to book a meeting or a call with a sales rep. Cool. But you know, we know this playbook. We've been doing it for a while. This is what the future of email looks like. We don't just reference the most recent interaction, but we can look at their whole recent history of interactions with us to better understand what they're interested in and how they're engaging with us. We scrape their company website in order to better understand what their company does and then use that to infer their pain points and speak directly to their use cases in our proposal for a call. This is rich, colorful and personalized, and I promise it's the last art history reference for a while. Okay. Our results from implementing this email strategy at HubSpot are that by adding AI personalization to our emails, my team is driving over 10,000 meetings quarterly for sales. And that's due to a 45 improvement in conversion rate. Now that's the average overall improvement for this program. We are going to, throughout this session look at individual experiments where you're going to see more improvements in conversion rate gains. But this is kind of like program wide. What we've seen after about 18 months of testing and this work isn't over. We are consistently building on it with new strategies and testing. And guess what? We're building it all on HubSpot. Yes. So we use HubSpot workflows and webhooks or custom coded actions so that we can make API calls to OpenAI in order to bring back that personalized email copy. We've also built a basic website scraper, again to get more of that context on the companies that we're reaching out to. And we're sending it all through marketing email. Let's take a closer look at what happens inside the workflow. So we enroll a contact. This could be based on them being on a list. It could be that they downloaded a piece of content, they visited a certain page on our website, et cetera, but we recognize we need to follow up with them via email. Then we scrape their website gives us more information on the company and we ingest additional context from the CRM to build a contact profile so we can have a little summary of this is the person we're about to reach out to in the email and who they work for, what their possible pain points are. We use that profile to send a request to an LLM I mentioned on the previous slide, OpenAI, but it could be a lot of different LLMs. We send the request to LLM and we ask it to bring back relevant email copy, which we then put into a personalization token that we can use in the email. It's pretty cool, don't you think? Well, what's even better is that as we've been testing this, we've been sharing all of the learnings from our HubSpot email team with our product team. Because at HubSpot we're fortunate to work with a really awesome marketing technology team. Big shout out to the technology and ops folks here who have helped supported all this work and they can help us build custom coded actions and make those API calls, et cetera. But we wanna make sure it becomes easier to build this type of stuff in HubSpot so that any business can benefit from it. So we've been sharing those learnings with our product team and you heard this Morning at the HubSpot Spotlight just announced at Inbound is Dynamic AI Email Personalization or AI Powered Email. Our customers can now deliver these one to one personalized emails at scale using context from the HubSpot CRM to personalize it. So what does this mean? Email automation programs used to constantly negotiate between personalization and scale. I mean we've always wanted to target and segment to different customers, but we had limits to how much we could do because if we were running an A B test, we couldn't let it run for four or five months, that's just too long. But now AI gives us that ability to personalize at scale we only ever dreamed possible. And it can be powered by new data sources. Again, the world of contact, first name and company, company name. I mean we've been using that personalization for years, but now we can unlock new types of data and in particular unstructured data. It's messier, but I can make sense of it in order to better personalize conversations. So you've learned a little bit about AI email and learned you have a new HubSpot feature to help you accomplishment Accomplish it. What else do you need me for this session? Should we all leave? No. I want to make sure you are all set up for success in this new world. There's a lot of testing that's gone into the conversion rate improvements that you you see here today. And I want to give you more of that inside scoop into what it takes. So I want to talk about the key skills and qualities that I think you need to be successful and be successful with your teams in adopting this strategy. Being successful with AI requires being a good coach. And a good coach sets a high standard. They give feedback and they give good feedback. They know how to give good feedback. They tailor the message to the moment and they do the drills and they get their team to do the drills and the repetition and practice in order to get things right. And who is your team? Well, AI is your team. It was also another theme this morning. AI has massive potential to deliver conversion success. And AI powered email will be that success for you if you can control the inputs, guide the message, and test relentlessly. So next up, I want to talk you through three failures that turned into three success stories. And by doing so, we're going to look at what goes wrong with trying to use AI in your email programs, but also how to harness it to increase engagements and conversions. You should know that that we have failed at every single new AI use case on the first try. We do not immediately get it right. It has taken a team of many years of testing in order to see these results that I'm sharing with you today. But I wanna give you that inside scoop so you can get ahead. Let's start with where we started. How many people in this room have asked AI to write an email for them? Raise your hand. Yes, I would guess it's most of us. Makes a lot of sense. Now, my next question is, what was your reaction to that email that you got back? Shout out couple of words. What? Sorry, What? I could. Yes, great one. I could have written that better. Sorry. Give me another one. Ooh, 60% there. Cool. It's cheating. Oh, fine tuning. Yes. Okay. Common theme here. It's almost there, but it's not quite. I often find that it's like really wordy on the first draft or Yamini mentioned this morning there are too many EM dashes. Totally get that. So it needs editing. We're not always happy with a copy, but we know it can write copy and it seems to have potential. What do great coaches do with potential? They give feedback. Like any great athlete, AI needs a great coach. Coaches observe their team, they make assessments, they make plans for improvement and they give great feedback. So let's take a stab at this together. This email is a little bit shorter, but read through this email and think through what is some feedback you would give for this email. What are some things that we could tell it to do better? I heard get rid of the opening line. Sorry, it's a little hard to hear in here. Shorter sentences. Cool. Needs to be more concise headings. Okay. Start with the customer. There's a lot of I references in it. You're so right. Three pair. Ooh. Okay, so we got some structure improvement. I like it. This is great. This is great. Thank you so much. All right, so we all as a group have good feedback and I'm going to bet that that's going to continue to become a more important skill for marketers to have over the next five years. Giving good feedback. And it's important not just to state what needs to improve, but why it's important. Like start with the customer because that's who you're speaking to or who you're trying to benefit. This isn't about you. You shouldn't be selfish in your email. So we need to make sure we are setting context for a goal and then we provide examples that are great and why they are great. And this is something we continue to do with AI powered email and the team. So it's incredibly important to set a high standard and give feedback. Here's an example of one of our prompts that we use in AI powered email at HubSpot. It's not the full prompt. I've cut out pieces of it and we're not going to get into prompting best practices today though. That is a great thing for AI powered email. But you'll see some of the pieces that we've really thought about over the past year in making sure this is successful. We start out strong on the inspiration front. I love this that my team writes, don't waste time on the milquetoast emails of yesterday. You AI personally write kick ass marketing copy individually to each person you meet on behalf of HubSpot. That's fun. I feel inspired to help my marketer do this job better. And then we provide really clear instructions for how to structure that email. We set a word limit. We say you should structure it in terms of pain, solution and result. And then you can see where the hours of output review came into this as we updated the prompt over time. For example, it was using large, obscure words like ethos and prowess. We had to get rid of that. Or apparently AI really liked puns for a while because we had to be very clear on not using puns in the email. So your process becomes important in making sure to review example outputs. But the learnings from that, it's not about. We can't just make the changes ourselves because we wanted to be able to make these changes at scale. And so we had to update our prompt in order to fix how the outputs looked on the other end. The other thing that we did was we took away some of the control that AI had. We set clear boundaries. So, number one, not all of our emails are 100% AI written. For example, we'll mix marketer copy with AI copy in order to provide more structure to the email. We will also break up the email into different sections. Somebody earlier wanted three paragraphs, I think. Well, here we have opening a pain point and a solution closing, and we actually prompt these separately in order to get better results. So we started this section with a poorly written AI email. Let's look again at another example. One of these emails is 100% written by a marketer and the other was a variant used in an A B test where parts of the email are written by AI. And my question to you now is which one was written with AI included? I'll give you a moment to read and then we'll do a vote of hands. Who thinks it's email A? We got a couple of hands. And who thinks it's email B? Few more hands, but it's somewhat divided. The answer is email B. And here you can actually see the sections that are personalized by AI and the data sources we are pulling from to do so. We'll talk a little bit more about data sources later on, but this illustrates hopefully how we're not sounding as robotic, but still adding in some of those personalization moments. Okay, what's next? Well, another fail we had was on personalization actually. And it turns out personalization for the sake of personalization doesn't necessarily make a difference. Sometimes marketers like to get really fancy and we're like, hey, I'm going to mention your industry and you're going to like me better because of it. But we actually need to make sure we're solving for the prospect or customer. At the end of the day, we have to infer their pain points and help them do their jobs better and that's how we'll see better conversion success. Otherwise, it's just like name dropping. And that's what we saw in this example. This is an example of a marketing email where we first tested AI in just personalizing the copy. And so we were able to take some information from the CRM, personalize that copy, but we had everyone get the same recommendation for a course to take. The inbound certification and conversion rate was actually flat. So what we actually needed to do was better tailor our message to our audience and provide more value in our recommendations, not only the personalization of the email copy. And that's what we did in the second version. So we had personalized copy, but we also had personalized content and recommendation. We actually changed the course we recommended based on who the person was, and that was based off of inferring what their pain points were and actually asking AI to make a recommendation on which course we should recommend to each person. This was responsible for a 2x conversion rate increase. Okay, I have one more fail for you. And an important truth. AI personalization is only as good as the data that's behind it. Let's take a look at this example. We reached out to Brian here and referenced that Brian was the COO at his company. But we got a response back saying, hey, actually I'm an intern here, but all the best. That was embarrassing. AI is only going to be as good as the data. That powers was another mention in the spotlight this morning. And if your data is incorrect, you can get bad results. But data is also an opportunity. It's important to think through the data sources that you can use to provide the personalized email and what you can share with the LLM to get the writing right. For example, we started our AI personalization tests with our Prospect database. So people we were looking to convert into customers. And then we started seeing good results there and we started to say, hey, we should also use this in our customer based programs. But we didn't see the same success. You can see it. We actually saw some declines in conversion rate. But it was at that point that we said, hey, we should probably reevaluate the profile we're putting together for our customers and whether there's additional data we could provide to better personalize the outputs. And that's when we started adding in additional data points specific to customers and even how they're interacting with our product. When we added this new context, we saw a 76% increase in conversion rate. And you can also start to see in this visualization those iterations that we're putting into our testing. So when we talk about data sources, I like to think about it in three main ways. You have your Contact data sources. So that's information on the actual person you're sending the email to, their job role, what they like to eat for breakfast in the morning. Okay, you probably don't know that, but context level data, then you have company level data. So what is the business trying to do? What industry are they in? That's where we scrape the website. We might look for how big is the business, what are some of their main pain points, they might be struggling with, etc. And then three is the context you have, that is how somebody or the business is engaging with your business. So their recent activity, what forms are they submitting, what websites are they're visiting, how are they attending events like Inbound, but how they're engaging with your brand. So contact company and context and you can now unlock unstructured data. This is one of my favorite experiments. This email is sent after someone has a chat conversation on our website. And as we implemented AI, we said, hey, it would be incredibly valuable to add the context of the chat transcript into the AI personalization. And we saw a 200% increase in conversion rates. It was definitely successful in being able to make a direct reference to that conversation they had. It shows that we're understanding and we're carrying that context into the additional conversations they're going to have with our company. Getting the right data points is a great example of doing the drills, the reps and the practice. Because we've seen it takes three plus iterations on average to see results. And the magic happens after a lot of teams give up. This is a totally separate experiment where you also see the number of iterations, but afterwards you see massive conversion rate increases. There are a couple of important process implementations we've made in order to help us along with this. Number one is, is we do a lot of review before we ever set an experiment live. So we'll select our data sources, our email that we're gonna test on, we'll write a prompt and then we'll actually test it on a bunch of contacts and see what the results look like and make updates to the prompt in order to improve the outputs before anything goes live to customers. That's important, especially because we all agreed that usually that first email that we're seeing is not quite there, it just has potential. So reviewing outputs ahead of time, we often do a few iterations of that before setting the experiment live. Once the experiment is live, we continue to review those outputs and we look for emails that are successful in converting and we give that back to the AI as successful Examples so that it can continue to improve its output. And that has helped a lot with improvements in conversion rates as well. So this is bringing us towards the end of our time together. But I don't want to wrap up without coming back to where we started. And of course, that brings us back to art history. Not long after photography was invented, artists and inventors learn to stitch photographs together and create the moving image or film. The takeaway here is technology will continue to advance and we need to continue to evolve with it constantly adapting. Our job is learning the tools and using those to our competitive advantage. So what are more of those new tools going to look like in the next five years? In the next year, even well before AI, we used to consistently think about targeting, list segmentation, design, copy, cta. These were the things we were always testing in email, AB testing consistently, and today that's evolving. We are now focused on prompting. What does a good prompt look like and how do I continuously update it to get better outputs? Am I selecting the right data sources to build that contact profile? How do different models perform? I mentioned that we use OpenAI, but we're also testing other models and in particular in non English, we're finding that Claude works a lot better for writing in German, speaking Spanish, French, et cetera. I see some head nods. We're testing new automation use cases because AI makes it possible and thinking about how we actually stitch these tasks together and build agents to send email. In conclusion, photography didn't kill painting. It expanded what artists could do. It gave birth to new styles because artists started asking better questions. Not what is this going to replace, but what can we now create? As marketers, we're at the same crossroads. AI can write faster, but it's up to us to make the message resonate. It can personalize, but only if we select the right data sources and prompt. Effectively, the most powerful email programs in the world are not going to be powered by AI alone. They'll be powered by marketers who know how to make AI work for them and work relentlessly on testing what's effective. After today, that includes all of you. Thank you very much for attending this session. We will do Q and A now and we have a little bit of time to do so. There are microphones set up here and feel free to hit me with those questions. Thanks so much for coming and also would love your feedback on the session as well. Thank you. Cool. Let's start here for Q and A. Hi.

So the idea of personalization at scale, the scariest part of that is Hallucinations. How do you account for or deal with the idea of sending out a ton of emails that could just be wildly wrong or have a key fact in there that's wrong and then just causes the recipient to check out? How do you guard against that? How do you account for it? What do you do there?

Yeah, the best thing that I can say is you need to constantly keep an eye on the outputs. Now again, that doesn't mean. That means we cannot review every single output and every single email, but as long as we're consistently checking on the outputs, it helps. Another thing that we're looking at now is actually building an agent to review outputs and flag examples of hallucination or where we might need to continue making updates to prompts as to make sure that that isn't happening. So also building in systems to review the outputs over time and flag for us where we need to make changes. But it's definitely a risk in this and we do see some examples. But the better that we prompt and give very clear instructions, we see less and less of it. Great question. Let's do one over here. How do you protect your organizational data or company data when you are putting it into an AI system, an OpenAI that can then be used across the population? Yeah, good question. So at HubSpot we're using an enterprise license where that data is not shared publicly. So it's a. Yeah, it's not fed out into the. We're not using public APIs. We have like a specific one for enterprise. But it's a great question in order to protect privacy of your customers. And that's an important thing to keep in mind. Good one. All right. Yes. Okay, thank you so much. Do you think we could have an overview of what the workflow looks like or could we find this type of deep training somewhere? And who do you send the emails from? Are your BDR okay with being sent under their name? Ooh, good question. Okay, example of the workflow. The best I can do for you is that slide that kind of shows the high level because I can't give you access to exactly how it's built in HubSpot, but that's the general idea. And then great question on who we send the emails from. So we had some examples where it's like the HubSpot marketing team, but a lot of the sales emails we are usually sending on behalf of the sales rep. It could be different sales roles. And then I also heard a question in there of how do the sales reps feel about it, which is A great question and one that we have discussed internally for a long time. I would say sales reps are getting more and more happy with the emails we were sending. With AI we used to get far more complaints about the emails we sent out when it was just marketer ring copy. The personalization is actually enabling better alignment with our marketing and sales teams to send emails on their behalf because it's actually personalized. And that used to be what sales reps really added. They could add the personalization before they sent an email. But what we're doing now is marketing sends the email and sales is following up on the phone, on LinkedIn, et cetera, and it's creating more efficiencies in our sales process where they can focus more on closing deals. Thank you, Appreciate it. Yeah, great question. One more over here. Hi, question for you. You mentioned before sending live to clients that you would send to multiple contacts. Maybe I just misheard what you said. Could you elaborate more on like with the AI experiments and you would segment. More segment. Wait, sorry, can you repeat that one more time? Yeah, sure. You mentioned that when you were segmenting the AI emails, you would send them out to multiple contacts before sending them live to clients. Did I hear you wrong?

I just want to elaborate on that.

We wouldn't send them out to context. We would review example input examples. So for example, in HubSpot today, you can send a testing email, for example, and view it as a particular contact. What we were doing was essentially simulating that with the AI email as well. So we'll take a few hundred examples for one rev where we'll basically pretend like we're going to send it out, but it's really just a test for us to see what the outputs are and we look through those outputs to then make updates to the prompt. Did that make sense? Cool. Yeah. Yes. So on the sales side, when you're thinking about like the follow up touch bases in those emails, did you use the same strategy that you showed here for those follow up emails? Like when it's outside of LinkedIn or you know, connecting with them in a different way, like how did you find success there? Yeah. So like after the first email, what about the second email? Third email? We're actually still building it into our follow up email strategy. So I'll probably have to follow up with you on that. One of the things we are looking at again like based on the data source idea is like what are those additional examples of unstructured data? So making sure we can give the AI references to interactions, they've had with sales reps. So if there was a call that got connected, things like that, we can use it in the follow up emails as well, but that's still in progress. Cool, thank you. Yeah, no problem.

I wanted to follow up with the gentleman that asked about the hallucinations and then also what this young lady that was in front of me had talked about in testing. It seems to me that you could pretty much destroy a company if you sent out a bad email. And it seems like there should be better safeguards in making sure that that doesn't happen. I understand that you're doing the testing on maybe 200 contacts, but that's really not sufficient. I feel like there needs to be a queue. So here's the, you know, here's the email that we're suggesting. But even if you had, like we send out, you know, today we sent out 30,000 emails and if I send the wrong email, it will devastate my company. I mean like put it under. So how do. I'd rather have an editor go through and just say, okay, this is good, this is good, this is good because you can see the hallucination pretty quickly.

Yeah. I would actually, I would recommend in cases where it's going to be really high risk, you add the element of review in and that is a piece of the strategy we're implementing as well today. So in areas where we feel confident in the AI outputs and its lower risk, we do that AI automated email at scale. In places where it's higher risk, especially like really target accounts for et cetera, or we're focusing more on enabling our reps with draft emails that they can review and then send. So I do think that it's good to have that level of, to think through where are the places where we feel comfortable doing this? You do a lot of that review of outputs and you also continue to review them or build in safeguards to like flag areas where it may be a concern.

Are there plans for some type of queue with HubSpot? Yeah.

I believe the prospecting agent takes more of the approach of like drafting the emails so that a rep can review them before sending them. Which would be different than using the example that I use today. Yep. Okay, so I would check that out. Yeah. Yes.

Hi. So we're, we're using something similar to what you kind of mapped out and showed over there for like lead funneling based on industry and vertical towards people on the team that specialize in those areas. Right. So kind of like taking this whole thing and then sending it down the line. And then right now we're working on that hyper personalization of the first handshake message with potential follow ups. My question is, you mentioned over there the scraping element of it, which is brilliant.

Cool.

How are you approaching, how are you like, I can see that working really easily and well on one on one scenarios, like one case scenarios. But how are you doing it at scale with pushing it into that flow? Right, so it's like first handshake email. But that handshake email doesn't wait for someone to do that research. It actually sends that about information into the email.

Okay, so actually the website scraper we do before sending the first email, and I can't speak to exactly how we do the basic website scraper, we'd need to chat with our marketing ops team who built it. But basically we do a basic scrape of the website and we get some context that we build into that contact profile and that's before we ever send the first email. So it is in that first step. And though you do make a good point about like there's often additional research that a rep or somebody will do for certain accounts, I will say we're still figuring out how to do all of that at scale. Like there is a lot of sales reps, you know, do an incredible amount of research about, you know, who's the right person to reach out to get access to different contacts at the company, validate information. And there's a certain level of that we're not quite at, but it does help with like building that initial outreach and doing so over email. And then reps still will do like research on top of that for their call tasks, their LinkedIn outreach, et cetera.

Thank you.

Yeah, no problem.

Yeah, I think my question may play off a little bit of what the other gentleman was asking because I was wondering in particular when, you know, you mentioned using personalization tokens for this unstructured like kind of these potentially like, you know, paragraphs worth of information that's unstructured. How, what, how do you actually do that? Like what is the technical process? Do you just have like the whole email is one giant personalization token or like how do you parse out the four or five that you need from, you know, a paragraph's worth of copy?

Good question. So yeah, we started with it basically being the full email and then that actually didn't work as well. And it was mostly because the AI couldn't do the full email as well. And that's when we actually inserted it into particular sections. What we use the most is that like body copy 1, body copy 2, body copy 3 or you'll have like truly a section of the email that you. But it's using a personalization token now it's going to be easier with the new HubSpot functionality than what we built. It gets a little bit beyond my technical expertise to explain exactly how the contact property works.

Yeah, I was just trying to visualize how to not only do it but explain it. So I think that helps me though.

Thank you. Cool. Yeah, awesome. Great question. Yeah, hi, great session. Thanks. I was going to ask about when you did the testing, have you included multiple people per company and if you have, how did that result, was it a positive or a negative result? And that's pretty much it. Yeah, we have done a little bit of testing with reaching out to multiple contacts at a company. We have not seen it work super well yet. And the reason is most of our email programs are based off of a contact taking an action. That's what we, you know, the contact takes an action. It's a certain level of intent and it triggers our follow up. But we know that's not always the right person to reach out to and that's why we've tested reaching out to multiple contacts. The challenge though is you don't have as like it's trickier to find that in or that reason for why you're reaching out. And we find reps still do a great job of it. We're getting better. We have like some recent tests I actually still need that are we're building the results from in the next couple of weeks to see if AI can help build that opener to contacts who didn't take the action. So it's something we're still learning about too. Okay, great question. Thank you. Yeah. Yes, hello.

I was just wondering in terms of effectiveness do you find the emails are better just more short and body copy especially with AI integration or have you also tested with more of a newsletter look with images and different columns and whatnot?

Yes, great question. We have done the most amount of testing with like the sales style emails where it tends to have like those very short concise in those three paragraphs. But we have done some with marketing email and are continuing to do it as well. And I would basically think about it as anywhere you have like a text module you could add AI personalization there. So we had one example with the academy content that would be one. You can definitely do it within newsletters as well and we're seeing results there too. For example Product announcements. It's something we've started adding AI personalization to and are seeing good results. Okay, thank you. Yeah, absolutely. Yes. I was actually wondering if you had done any testing with emails that are from a person like just signed like Best Katie versus Best Katie, CEO of a certain company or like how personal do you get and do you see better conversion response rates based on that? Yeah, based on the who you're sending from, if you include like title, et cetera for the. Yep. I don't know that we've a B tested that. It's all structured in the signature. I think it's generally just their name. I don't. We haven't a B tested it. It's a good question. Maybe we'll check it out. Okay, cool. Thanks. Yeah, no problem. Yes, I've already asked, but you can ask again. Thanks. Would you say that it's better for the prospection email like these sales one to one or you're talking about respond to a quote request, a contact request or demo request? Yeah. Actually the first places we implemented this were demo requests. I'm wrong. It was like the third, but it was early on one of the places. Just because those are such important places to see conversion rate gains and we've seen it across the board pretty much every. Every program. We only have like one or two programs where we have yet to start testing and most of them take time. So we have to figure out the right prompting, the right data sources. But we're still seeing consistent conversion rate improvements across different types. Okay, thanks. Yeah, of course. Thank you. I know we're at time, but I feel like we could do like two more. Let's do them quick.

A question about more on the prospecting side.

When you don't have very much information.

What data sources do you like? Aside from scraping the website, what data sources have you found that give you good information?

Like some intense signals when they're kind of sparse? Just curious about that. Yeah, good question. Hmm. Website scraping helps. When we have zero data to personalize, it's often where we. I mean we've used industry or job role. Some of those get filled in via enrichment as well. So even if the person hasn't given us that information again, it can be a place where the data isn't 100% accurate, but we can infer or use the enrichment to get us more information. But the more information you have, the easier it's going to be. As usual. Thank you.

Cool.

Thanks. Okay, last one. Hi.

This is more of a clarifying question, but when it comes to the contacts data, like the website scraping, is that information then living in the contact record or the company record?

Usually in the contact record for the contact information.

Okay, so even if it's coming from the company, the. The person that you're going to reach out to is going to live in their record.

Yeah, sometimes we'll have it in both places. Honestly, we know that Ari works at HubSpot and we also have a company called HubSpot where we can reference both properties and hopefully they're the same information. So there's definitely some overlap. Got it. Thank you. Yeah, no problem. Awesome. Thank you all for hanging around and asking amazing questions. Really appreciate it and have a great rest of Inbound. Sam.