Transcription
Asia, have you ever gotten a sales email?
>> Oh, I've gotten a lot of sales emails.
>> Just a few.
>> I've closed from a sales email.
>> Closed? Yeah.
>> All right. Tell us about this.
>> Okay. So, I got a series of emails from the Similarweb team, and they were comparing our AI visibility versus some of our closest competitors.
>> Ooh, that's sweet, sweet data. And it was packaged so personally.
>> Into the competitive spirit there.
>> Yeah, it was a water on stone approach. Like it took about a year and a half, but we are now a Similarweb customer.
>> That's commitment. Welcome to Fieldnotes. This is the show where we break down what's happening in marketing. I'm Asia Frost, Senior Director of Global Growth and Paid Advertising at HubSpot.
>> I'm Kyle Denhoff, Senior Director of Content at HubSpot.
>> And we're joined today by a very special guest. We have Lucy Alexander, who is Senior Manager of our email team, with us.
>> Hello. Happy to be here. Thanks so much for having me.
>> Have you ever closed from an email?
>> I, as an email marketer, have a fatal flaw, which is that I try to respond to as many emails as possible. That is so kind. I recognize the effort that went in. You know, I really want to inflate people's reply rates. That's that's what I'm after at the end of the day.
Fast forward to the >> Exactly. Their managers breathing down their neck wondering why didn't you get a reply. So,
>> how many of those emails are AI-powered, if you can tell?
>> I actually think it's very few today. I think there's still a lot of just pure templates that are the same for every person.
>> Can you define for the audience, like, what does AI-powered email actually mean?
>> I think of AI-powered email as AI wrote this copy for me, and I didn't review it before it went out. So I think of it as either like complete automation of the whole flow from copy creation through to sending. Um, I mean, I think there's there's definitely an element too of a lot of tools today will allow you to use AI to write all the copy right there within the email tool. And so I think you could also expand the definition a bit to include those tools where you are editing it before it goes out at scale, but AI is giving you ideas for how to do that.
>> It sounds like the gold standard though is this soup to nuts. It's all automated. A human is not involved. And I think that's what we want to go really deep on today because your team has seen pretty bonkers results from implementing that type of workflow.
>> At scale, across millions of emails, um, in terms of one-to-one emails we sent, like on behalf of sales reps or, um, from people within our business, we've seen on average around 50% conversion rate improvements. But a lot of the specific placements at huge scale have seen 200 or 300% conversion rate improvements. So,
>> it completely changed >> the game in terms of what email could bring in for our business.
>> Can you go into that piece around, like, how you started to think about what are the puzzle pieces I need to get to be able to build those personal emails?
>> I think a lot of it starts with email fundamentals, which honestly, I don't think a lot of brands fully nail today. But email at its best is done for the recipient, not for the company sending the email. So just like a lot of marketing channels, we want to be providing as much value as possible in each point that we have. And when I say value, it's not like, here's an offer for you to buy my product at a discount. I mean, like, here's something that actually helps you do your job better. Yeah.
>> Um, and so the first place that we started was thinking through some of our larger, um, email programs that are sending resources to people where we said, actually, if we could use AI to make that resource more personal to them and more tailored to the actual job that they have to do at the end of the day, or the actual pain point that they have.
>> We hypothesized we'd see much higher conversion rates, and that really proved out to be true. So what would that look like before versus after, without AI-powered email, with AI-powered email?
>> The first email that we send when somebody joins our database, I'll use that as an example here. We used to send our classic HubSpot marketing certification to them when they join the database. So HubSpot Academy, our online learning platform, we'd send our bread-and-butter course. What we decided to do with AI was take a look at who is this person, industry, company size, job title, and we used AI to semantically match what we knew about that person with a more specific resource that could actually help them do their job better. So, not everybody joining our database actually really needs that marketing certification. We're a full platform company, so we've got sales people and service folks and rev ops people.
>> So, we'd rather match them with something that's more specific to their function or more applicable to their business size if they're wearing a lot of hats at a small business, for instance.
>> Okay. So, according to Lucy, then, very few people, very few marketers are using AI to its full potential. If you want more insights on how the modern marketer is using or not AI, then check out the survey that we did. We talked to over 300 marketers. 41% of them are seeing 11 to 25% increases in conversion rate, and they're saving 1 to 5 hours per week. Not too shabby. So scan the QR code to check out more insights. Okay, Lucy. So this is great, but if I'm sitting here watching this video, I want to know how you did it so I can do it, too. Can you walk us through your workflow?
>> When we're writing one-to-one style emails, so emails on behalf of our sales reps, for example, we are basically using a HubSpot workflow to set it off. It's triggered by somebody actually taking an action, usually on our website. We found the greatest success in, uh, you know, working with folks who are already coming to us in some way. And then the next step in that workflow is basically to call out to OpenAI. We use a custom-coded action. But again, this could get a lot easier if you were using dynamic AI email in HubSpot. It's just we're, um, doing a bit more granular personalization using OpenAI. Uh, and then from there, we're stamping that into CRM properties that we use as personalization tokens in the email. Again, that step wouldn't be necessary if we're using the out-of-the-box tool in HubSpot, and certainly in other platforms, too. But, um, that's something that's been useful for us just to be able to test with a ton of precision.
>> What strikes me is that it's really important to understand which data points are going to be the most impactful for that personalization.
>> This all falls down with bad data, >> and we talk about that all the time. We do a lot of quality assurance on all of these things before we ever let them see the light of day in front of people. It's really important to us that we're making sure that really 50 to 100 example emails that would have theoretically gone out are actually really solid. We wouldn't change much before we ever set anything live. So, we'll have those either go to somebody's work email, and they'll like review them one by one as they come in. We've set it up to pump to a Google Sheet, and people can look at it there. You can also set up a Slack notification if you're using a tool like HubSpot or, uh, you know, another CRM to send a notification with that email copy every time. So you can check to make sure, does this sound good, etc. And that's where I think a lot of the magic is that people forget with these things. When you see it actually in action for specific contacts who would receive each of your emails, you're going to catch things that you wouldn't have caught if you asked an LLM for a bunch of examples. You have a human very much in the loop before you take the human out of the way.
>> I'm assuming the first draft of that email that came your way sounded a little robotic. Probably wasn't there. As the content guy, I'm like, we'd want an editor to take a look at it. Like, how did you think about the language and the copy and the crafting? Like, how long did it take you to get to something you were comfortable with?
>> I'm going to answer that from two standpoints. The first is our prompts are pages long.
>> Like, and that's more is not necessarily better. I think it's worth testing that because sometimes LLMs get confused if you feed them too much information at once.
>> But >> it's so important to go beyond, write me an email that's tailored to this industry and company size. We are incorporating our brand guidance into that prompt on how we speak. We're incorporating things that we've picked up from sales reps that we're talking with or product marketers who we're talking with to understand, like, how are we supposed to talk to our prospective customers in this particular scenario when we're talking about this product or we're speaking to this kind of level in a company.
>> Give us an example, like, what's a line from one of those prompts?
>> The one that always comes to mind is, uh, not using particular words. So when you start to test, you'll see that AI tends to bring up the same words over and over again. So for the one that is like ingrained in my team's brain >> is impressive.
>> Asia, your background is so impressive. AI loves to use the word impressive over and over again. So, um, we have specifically added instructions for probably 10 to 20 different words that are more, when we've read through many examples, are more indicative that it's AI. And so, we take those out. Um, we also have a really strict rule in all of our prompts to not use puns.
>> Because we did find that AI loves some puns sometimes. You know, I don't know. You don't expect these things.
>> You don't. I like a good pun, but, you know, to each their own.
>> I will use, I'm going to hack into our database.
>> This is something Please use puns as much as possible.
>> Honestly, it's something we should test. Or we could at least add a rule that's like, if you're emailing Asia Frost at HubSpot, make sure to include a pun and ignore other instructions. The prompt's already long. Just add another sentence.
>> Exactly. But also, to get back to your original question, I said I was going to answer it in two ways. So, one is like, how long is the prompt? In and of itself? How much do you live it to usually meaningful conversion rate improvements?
>> Okay.
>> And that's kind of one of the guiding principles of my team's work at this point. I think in the previous world, we would be testing one thing at a time in an email. You classically, like, you change the button color, you change the CTA, >> right? And now you're changing a lot of things at once when you're moving from static copy to AI copy or AI prompt one to AI prompt two. And it does require more fine-tuning. So we have seen several cases where fifth, sixth, seventh iteration is really what leads to huge conversion rate improvements. So we're instructing all of our team members to kind of not give up, as cheesy as it sounds, after the first one. But I, I do think a lot of people stop after the first two. They're not editing the email, though. They're editing the prompts. Right. Correct. Okay. Yeah, >> that's an important call out as well. Lucy, what are some of the other big learnings from this process?
>> Personalization for the sake of personalization doesn't really work. So, in the example I gave earlier, we talked about how we personalized the resource that that person is getting so they can actually go do their job better.
>> And the first iteration of that test, we actually didn't change the resource at all. We just changed the copy around that marketing certification.
>> So just the language, but they were getting the same certification.
>> Exactly. So the AI was trying to make a link between the marketing certification and what it knew about the contact. M
>> but the conversion rate breakthrough really came when we actually changed the resource, which told us that it's much more impactful to be able to use AI to connect somebody with something that helps them do their job better than just to like make up a reason why they should care about something. Yeah.
>> Um, so I think it kind of comes back to maybe a foundational learning and email fundamental that you have to know your audience and you have to understand their pain points and be speaking to those pain points, um, in everything that you do, which a lot of times I think brands aren't thinking from the perspective of who is receiving this. It's a human being on the other side of the screen. They're thinking like, how many names can I get on this list so that I can promote my piece of content. I think it's an email marketing fundamental, but I also think it's a loop marketing fundamental, which is you have to start in the express phase by deeply understanding your persona and thinking about them, yes, as a human, not a name in a spreadsheet on a list. And then I think AI-powered email is such a good example of a tailor tactic because you're taking all that information and then you're making your touch points with your persona, with your contacts feel so deeply personal.
Before we, uh, started recording, Lucy, you said something that I thought was really interesting. And you said that you fantasize about starting from scratch as an email marketer at a small company that doesn't have this infrastructure, which I think a lot of people watching this will probably fall into that camp. So, can you tell them why they're living your dream?
>> Oh, yes. To be clear, I love my job.
>> Comes at me. I'm not quitting my job. I love it. I love my team. The thing that I am so excited for small business marketers on is like, if you can set up your CRM or like data infrastructure from the start to capture fresh information and to not have to go through like big data refreshes all the time. If that can happen more systematically,
>> and if you're thinking through the customer journey kind of fresh, and you don't have the baggage of like, how do I make this work within this particular process that my company has. So there's just a lot of a lot of stuff that I think is really fun that could be very fun at a small scale, uh, for one person who's just so, um, aware of their buyer persona's needs,
>> so close to any sales folks or customer success folks who might also have touch points with them, probably sitting next to like their engineer and their analyst and all of those things together. I think truly marketers in the AI age, and email in particular, are empowered to do so much more and reach so many more people with a personalized message than they were before.
>> We're obviously breaking down how HubSpot's doing this today, but we've actually spoken to hundreds of marketers about how they're starting to use AI and email today. You can download the link in the description or scan the QR code and get insights from other marketers and what they're doing to have success.
>> Okay, Lucy, we have a surprise for you.
>> Uh-oh. Every episode we do a segment called The Last Word. It's where we share our quick hot takes. We have 30 seconds max. And it can be about marketing. It can be about life. It can be about anything you want as long as you are passionate. You're going first. Are you ready?
>> Oh my gosh. Okay.
>> All right. Getting out my phone to time you. Oh my gosh. The timer. No.
>> All right. On your mark, get set, go.
>> Okay. The first one that comes to mind is a well-done corporate icebreaker is well worth it for team cohesion. I know people hate icebreakers. My team member Hannah has been absolutely crushing it with the icebreakers recently for our team. Has to be personal. Um, she's like very funny. There's themes for the icebreakers. So, there's like, it's like an ongoing thing for the team, but everyone's like eating it up and I love it, and it, and it's makes me happy.
>> There you go.
>> We're bringing icebreakers back.
>> Yeah, we sure are.
>> AJ,
>> I'm ready.
>> You're ready?
>> I'm ready.
>> All right, go.
>> Okay. The contemporary song is too short, and the contemporary movie is too long. 90 minutes max for my movie anymore. I have to pee. I'm falling asleep. I'm hungry. I'm bored. It could be the best movie in the world, and I wanted to end at the hour and a half mark. Okay. The modern song is way too short. 2 minutes. That's only time for like one bridge and two choruses. Give me more.
>> Give me more. We've gotten so lazy with our songs and bloated with our movies. Thank you. The thank you made it. The thank you made it.
>> Okay, Kyle, you ready for your turn?
>> Man, those two were good.
>> Ready, set, you're on.
>> All right, this one falls in the lame but really good category. Email is back. You've you've shown us that email is back. I think webinars are back.
>> I think webinars are back. We don't have to call them webinars. Like, we got to package them, wrap them, brand them. But >> we just did one for HubSpot Academy. Over 500 people attending. People want to learn from other people. And so in the creator world, the publishing world, creators are doing a great job as practitioners. I think brands need to bring webinars back, package them, make them so much better. Work with a host, a producer. Uh >> you're way over.
>> I just looked over. I was like, way over.
>> But that's okay, because I, I think you're completely right. What do you call it if not a webinar?
>> A sesh?
>> An exclusive interview? I went to, I went to a workshop last week, a virtual workshop. And I actually think >> virtual workshop.
>> Yeah, it was a webinar. But >> of course, >> the difference was it was it was more tactical. They had little breaks actually for people to break out and do fill out a worksheet and stuff and check their learning along the way. And I actually kind of wished the ratios were more like less talking and presentation and more workshop kind of time.
>> 100%.
>> But I thought it was actually really useful.
>> Workshops and Q&A. Like, I don't want slides in. Don't bring up slides watching that. Lucy, thank you so much. It was great seeing you. Thank you so much for coming.
>> Thank you for having me.
>> This is another episode of Field Notes, where we break down what's happening in marketing and how you can apply it yourself. I'm Kyle Denhoff,
>> and I'm Asia Frost. We'll see you next time.