00:11:41 · Guillaume Moubeche
[00:11:41] S1: Yeah. Maybe, like, the... For the express phase, like, is that something that you implemented at an agent or context level within HubSpot so your team can also leverage it? Because what I'm seeing is that whenever you onboard, like, more junior marketers or junior people in your team, they don't actually always know what good looks like. I think it's it's quite tough, you know, like, to define it for people. And some people, like... And we see it also, you know, like, some people, like, don't particularly like ChatGPT or Claude, they say the output is always like bad or not as good, and the reality is like the more time you spend on it, the better like the output will get. So how exactly do you manage for your team to also leverage your capabilities? And the second question is how do you get that input? Is that like just you doing it, or do you also bringing like tons of conversation with customer, like customer support, like all these kind of things?
00:12:42 · Kieran Flanagan
[00:12:42] S0: Yeah, exactly. So there's a three step process to build a first version of it, which is a lot of that is actually getting your unstructured data. So through sales calls, through support tickets, a lot of it is through your analytics. So you pull in all of the content that is performing well, all of the social posts that are performing well. And what AI is really great at is extracting patterns. And then the third thing is, like, you yourself have to be able to... I think the thing that people are start... Just have forgotten is there's some work for you to do. And so you have to, like, articulate AI why things are good. And so I actually think that a junior marketer, and maybe this is where a lot of people go wrong with AI, would actually not... Does not do a great job of working with AI because they themselves have not worked enough to understand how to teach AI what good is. AI is much better for people who have real domain expertise and can teach the AI how to do things. And so in that taste profile, the first version would be built by people who can understand what the taste and judgment is. And so they would instrument that into AI. Now you would set a... You would... In an... Into an artifact, and you would set it up in a way where every month, it... There is information being pulled in to update what is being learned across all of the marketing you're doing. And that is preloaded into a context layer. And every time a marketer does work through tools that they're provided with, whether, you know, whatever the AI assistant is or the agent, there are rules that will force those things to read the context layer before they actually provide output. So you're not just allowing everyone to go do whatever they want, and you get a mismatch of stuff. And so the context layer really matters, and you wanna have that governed and to make sure that people have to pull information from that, so it's on, you know, style, brand, that it's exactly what you want them to produce. But I do think that one of the big issues with AI right now is people are becoming incredibly lazy. So they just ask AI to, like, teach itself these things.
00:14:39 · Kieran Flanagan, Guillaume Moubeche
And people, I think, misunderstand that AI is much more of an accelerant for people who have some domain expertise, and is a little harder for people who are just, if you're more junior, you're gonna be more tempted to ask AI to create its own taste profile. Right? And it teaches selfies things, and that's actually not where the value comes from. [00:14:57] S1: Yeah. I agree. I I think it's usually like, you know, it's... If you serve like a very bad quality and taste to the AI, obviously, like the output is gonna be pretty, like similar to what you feed it in. Once you have that, like, I'm assuming, because HubSpot has also like a different persona that you're gonna try to kind of tailor the message to specific persona from the the context, or how does that work from there?
00:15:26 · Kieran Flanagan
[00:15:26] S0: Yeah. You can have a taste profile, and then you can also have... You could do this in different ways because I have a content OS that I use for this. And so I'll give you the example for my content OS and go back into the taste profile because I think the content is a good example of this. And I will just peak her to this with the best marketers and go to market people who work with AI use AI to do, like, first version ideation and drafts, but they themselves elevate that work. The worst people just copy and paste stuff. So everything I'm saying is never do this and just copy and paste. Like, you, the human, really matter. Like... And if you just outsource everything to AI, you're gonna just be the same as everyone else. But if I go back to the content system... So, you know, the the example I would have is I have a simplistic, what I call, like, a creative context OS, and that has, like, three core MD files for when I'm creating content. And I try to... I had a very complex one, and I simplified it recently recently because the models are just getting so much better. And it has this creator dot MD file, although I... I've swapped from MD file to HTML files. Maybe that's just, like, a bit of a dorky thing that people don't understand, but it's basically forever in the in the context people are using MDs and dot MD files, but actually HTML files are much better because you as a human can better read them, review them, and work with the AI to improve them. But it has a creator dot HTML file, an audience file, and then a taste file. And so the first one, the creator style, is basically will capture all of the content I've created on LinkedIn, YouTube, and Substack, and it'll extrapolate patterns to say these are the kind of patterns that work. And then I will actually finesse those. And then at the bottom of this, coming back to your original question, it will have rules of how we'll adapt those per platform. Because, actually, one of the things that... You know, I've created content for a long time, pretty big... We have a we have a YouTube channel, LinkedIn channels. I have a Substack. And it's kinda surprising how little things work... Like, how differently things work across platforms. So, the one piece of content does not work across platforms anymore. Every platform has its own set of rules.
00:17:25 · Kieran Flanagan, Guillaume Moubeche
X is another one I try to spend a little bit of time on. And so it has little lines of how we will adapt things based upon each platform. And so going back to the taste profile, it's the same thing that you have an overarching taste profile, which is like the, you know, overarching way that you would create things that would resonate for your audience, but it would have a set of rules of how you would adapt that per persona. And so HubSpot would market to marketers, salespeople, chief whatever officers, and so on and so on. But again, it would give you a a version that you could then elevate. And this is the whole thing is what it allows someone to do is spend 80% of their time on the last 20%, Whereas they used to spend 80% of their time probably on trying to get to the last part that really mattered. Now they can spend all of their time on the last part that really matters. [00:18:10] S1: And do you do that like on your own or do you do that with AI?
00:18:15 · Kieran Flanagan
[00:18:15] S0: The way I think AI is great on... And specifically in terms of content is two ways. It's great for ideation. So what I do is these are the things that generally have worked on my platform because it extrapolates patterns. This is my audience. Now I find ideas that will fit that audience that are patterns like this, and it will bring back a bunch of those things. And then I'll say, like, create a first draft on these three. Let me see how they look. And it will map to my writing style. I will apply my writing style and say, these are kinda like the ways that this look. Then I will actually take one of those and totally rework it like myself. But the next way I'll use AI is I won't ask it to, like... The kinda what you were saying is, like, I work with AI in terms of... I think of content in modular blocks. And so a hook is part of the content. The... Open in paragraph is part of the content. Stories are part of, like, great content. Examples are greater part of content. Lessons, quotes, they're all part of great content. So I have a a skill that will say, how would... I'll give it, like, my content that I'm working on, and I'll say, how would you enrich this? What would bring this to life? And it will suggest maybe a story, and my stories will be throughout time. And so it will provide a, you know, a story. Here's a great quote that could actually be added in here. Here's a great example. And I'll look at those and see if I like any of them. So basically, like, I write an assistant, you know, a researcher. Someone actually will go get me things, but I always create the content. I can never have someone create content for me. It just doesn't feel authentically me. Like, I could never have a ghostwriter. I could have never have anyone create content for me. It just isn't the way I'm able to work, but... So I think that's the best way to use AI, as like an incredible assistant.