205: Get It to 80% - What AI Actually Changes for Marketing Teams

Speaker A: Welcome to Alter Everything, a, uh, podcast about AI analytics and the future of work. I'm your host, Joshua Burkow. Marketing has always been about staying ahead of what's next, and right now that next it's AI. But the reality inside most teams isn't transformation.

Speaker B: It's actually a lot of confusion.

Speaker A: There's too many tools, too many experiments, not enough clarity on what actually works. And the question isn't whether AI will change marketing.

Speaker B: You should be clear that it will.

Speaker A: It's where does it exactly and actually help.

Speaker B: Where does it break?

Speaker A: And today's conversation is about what happens when AI meets real marketing teams. You know, there's no shortage of content about AI and marketing, but most of it focuses only on what's possible, not what actually happens when teams try to implement it. My guest today, Martin Broadhurst, founder of Broadhurst Digital, works directly with companies trying to figure out this exact problem in real time. So today we're going to focus on a few things. One, what actually happens when marketing teams adopt AI, how marketing work is changing at uh, a practical level. And lastly, what still requires a human, even in this AI first world.

Speaker B: Let's jump into it.

Speaker A: Martin, welcome.

Speaker B: You've been working in inbound marketing since 2012, it sounds like, and now you're helping companies integrate AI into their marketing workflows and tech stacks. And I am going to love this episode. I already know because first of all I need to know everything there is about marketing, so hook a brother up. But also, marketing is one of those areas in AI that sort of led the pack and jumped out in front from social media and content development and all these other things. Um, and I think it's going to be a great conversation. Welcome. I'd love for you to briefly sort of explain what you think inbound marketing is from your words. Um, for our non marketing folks, I'll give you an easy one to start and then we'll get into the hard stuff. Sound good?

Speaker C: Sounds good, yeah.

Speaker B: Cheers.

Speaker C: Joshua. I appreciate the soft layup. The, um, inbound marketing question. Well, inbound marketing for the non marketing listeners, it was a phrase coined by the founders of HubSpot, Dharmesh Shah, uh, and Brian Halligan in the late noughties kind of became very prominent through the 2000 and tens and it was really about using content marketing. It was originally how you used SEO and social media to build audience, attract people to your, your own media channels, generate comments, replies, all of that kind of engagement. How do you then convert them into leads and how do you then close them into customers and turn them into uh, evangelists. And it's, it was the playbook for before the 2000 and tens for most tech companies. And it's still a playbook that many are trying today. But it's interesting that this year we're seeing a shift, a shift from HubSpot in particular, who coined the phrase as their inbound conference, is now becoming unbound and that is largely driven to the changes that we're seeing in the way that consumers and buyers are, uh, consuming marketing content.

Speaker B: Thank you. Before we get into the weeds, I like to sort of start high level because a lot of times conversations get started in the weeds and there's no context. And I think if you could sort of give your take on marketing and AI and more specifically, it wasn't born yesterday, but it hasn't been around for forever and it's sort of this fast growing thing. Where does it stand today? And then what are some of the challenges when companies try to introduce AI into their marketing teams?

Speaker C: I think that's an interesting question when we think about what we are considering in that bucket of AI. Marketing, you're right, has been kind of at the forefront of uh, using AI for a long time. We've used it for the P's of marketing, the four P's of marketing, product price, promotion, place. These have been used for things like dynamic pricing, which has been algorithmic for a long time. We've had recommendation engines that have been AI driven. Now they're not generative AI forms, but they've been marketing applications of artificial intelligence that have been around for a very long time. Where it stands today through generative AI is obviously a very different landscape. I think the launch of the GPTs and large language models has obviously changed that. It's interesting you've mentioned in the intro that marketers have been using AI. They were early on the bandwagon there and I think they really were. Marketers are always looking to get that next competitive edge. And I think the language models, the early language models really lent themselves to marketing because so much of what marketers do involves writing copy. Uh, for example, we talk about inbound marketing a moment ago. That's lots of blogs for ranking in search. It's writing ad copy, it's writing ad headlines, it's writing email marketing communications, it's writing press releases, et cetera, et cetera. And what we saw with generative AI was pre chatgpt. So when large language models were really still kind of a niche technology, they were being used by marketers. With tools like uh, Jasper and copy AI and lots of AI powered copywriting SaaS platforms basically that did that piece of the marketing journey, obviously that's changed a great deal. We now have large language models like Claudopus 4, 7 that are doing agentic coding, we've got codecs from ChatGPT, we've got models that are capable of more long horizon reasoning tasks and research agents and agentic behaviors. All of these capabilities that have been unlocked far beyond Just write me a clever subject line for my email. Which is certainly where we were a few years ago and I think now where marketers are is they're in an interesting space. I think there's a lot of marketers and marketing teams that are just kind of getting to grips with this technology to understand, to understand where it's at. And there are still some that are maybe using ChatGPT to write them some emails and they're at the, I would say they're kind of laggard territory. And then you've got people doing it with automated content campaign testing. I think we're seeing people use AI generated content for good or for ill on platforms like LinkedIn to mixed effects and yeah, it's a, uh, it's a very mixed landscape out there.

Speaker B: I, I, uh, it's interesting. There's sort of two things I took away so far is it's fair to say that marketing M teams are probably a lot like any other finance sales teams in that there are some that are really pushing the gamut and doing advanced applications, getting into agentic AI and marketing. But you sort of hinted at like there's marketers who are still learning how to spell GPT and just get started and try to find applications outside of the basics. Uh, where do you think the teams are struggling the most? Is it the tools? Is it just learning the different applications and LLMs and all that? Is it building out essentially workflows? Is it a mindset challenge? Like where do you think the challenge

Speaker C: lies for marketing, specifically tools, workflows, mindset? I think it's a little bit of all three to be perfectly honest. There's definitely a AI literacy piece that I think is the people that get it, get it, they throw themselves into it and they just get started and like right, this is amazing. I'm um, still working with lots of organizations where most people don't have a paid subscription to a GPT or Claude Codex Copilot, whatever. They're operating on the free version of Microsoft Copilot, which is the kind of IT approved, safe version. But uh, actually if you've got a Copilot Pro license and it's actually really quite powerful when you use the Frontier agent, when you use the analyst agent, um, and you start stringing together some of the more interesting things you can do with it once it's embedded into your SharePoint directories and what have you. But that isn't the version that most people are getting. So there's a tooling version. I don't think they're teams are being given the right tools for the job. If they are, are they being supported with training and you know, the IT team will send out a couple of newsletters, maybe forward on the latest Microsoft copilot eshot that they've received and expect teams to figure it out themselves. So that workflow piece comes into it as well. No one's helping them build the workflows, no one's telling them what the difference between an agentic assistant or a custom agent, like a uh, copilot agent, what that can do versus an automated power automate workflow with some AI prompt steps in the middle of it and how that could completely reshape the way that a team works and certain tasks within a team.

Speaker B: So I have personally three different subscriptions and then on top of that I have two at work. So obviously I spend a lot of time doing this. But there are people who to your point, don't have a subscription yet and so they're not hands on. How do you get them engaged? How do you get them up to speed? And corporations are still figuring this out, like how do I deploy this across the entire company and keep our governance, keep our security, make sure that it's actually being utilized, you know, money's not going to waste per se from your point of view, when you walk in to a client and you're tasked with helping them grab onto AI, uh, make some progress, what looks easy from the outside, but once you talk to them you're like, that isn't as easy as people think of it on the, on the outside. Do you have any thoughts on that?

Speaker C: Yeah, I think AI adoption, where it falls down is often people overestimating what's involved and they, they hear that AI is coming for jobs and it's this amazing thing and then it can do anything. And then you get in a room with people and they go right, well we've got a co pilot license or I'm using Copilot as an example, but it could be, could be any of the others and I've got this deck that I need to put together every month for C Suite, whatever it is, and it takes me all weekend and if I could get my weekend back, can, um, I just throw it in with a couple of prompts and a couple of slides and be like, yeah, I'm done with it. And then you work out that when you get into the weeds on it a little bit, actually, that particular task that they've said, I've got these slides, it's a much deeper task that there's, there's fetching data from messy situations, there's like gathering data from random teams, chats, and telling this person to go on Google Analytics. And what sounds when they say I just throw together a few slides is actually a much messier process. And that's where it's going to fall down. So one of the workshop I was doing just last week, it was interesting, doing a real exercise with the team to show them the limits of where it can get them. Like, it can get you so far a lot of the time. For a lot of the tasks, it will get you 75, 80% of the way there. But you as a human need to finesse it at the end or might even need to do some manual copying and pasting from one system to another, heaven forbid. But a lot of you, it's not

Speaker B: like you're doing that, not doing that today, right? It's this perfect, uh, transition to. The sort of landing question that I want to get to is really this, this idea you've talked about before in using AI is to get to the sort of 80% and then focusing human effort on that final 20%. Can you sort of elaborate a little more on that? Just, I want people to really understand this because I know specifically in my conversations they sort of approach the table as an all or nothing or that elevated expectation that, oh, it's so amazing, it's going to do everything. And yeah, I think folks like you and I sort of after we're done cringing and being like, yeah, that's not exactly, but I'd love your thoughts on, on that. So is it 80, 20 split?

Speaker C: Well, that's certainly true. And I think they will get you 80% of the way there. And 20% of the, the polish or the finishing touches come from the human. And that takes shape in different ways for different tasks. But like, the window of what those tasks contains gets, that gets wider all the time. Right? If you were to speak to me two years ago about what an LLM could do, I would be like, yeah, it'll get you the way there on this set of tasks and I might reel off 10 or 15 tasks in marketing that it could do really well. Well now we'll say it'll get you 80% of the way there in this set of tasks. And that task set is now 30 to 40 tasks long.

Speaker B: Right.

Speaker C: So the window of its, its capabilities is expanding but there is still the need for human in the loop and human oversight. Case in point, Microsoft researcher, uh, agent. It's a deep researcher tool, much like Claude Research, Gemini Research, what have you. You give it a topic and off it goes. I think it's actually pretty good. It tends to, from my experience, hallucinate less than some of the others. But in one of them that came back recently with, with Copilot, I'd asked it to research how AI was being used in, in libraries and kind of public libraries and I was trying to find some interesting use cases for some work I was doing. It came back with a, and a case study that was really weak from the outset. You could just read the description of it and I thought that just doesn't sound quite right. The use case, it was almost like um, like it had been written by GPT3 from years ago where you might say give me an example of how AI can help with marketing. And it would say it can help with improving customer experience and personalization, like the most generic phrase. And I researched it and what had happened was this. It had found a citation and a source online, but it was basically just a spammy blog. It was a link building spam network, um, someone trying to sell you some backlinks with all those spammy emails that you're getting in your inbox. It was one of those blogs that it had cited and if I hadn't been on my game and checked that and read the report, there is every possibility that, that I could have included a completely fabricated case study and handed it to a, uh, client or used it as part of a report or something like that. Now that's just one example. There's, there's many others. There's, I mean copywriting is a really interesting one. The amount of copywriting that people use AI for these days and they just take it. You can tell they've just copy and pasted it, posted it without even the most cursory checks and just stylistic changes or I mean how many, how many brand style guide in house tone of voice guides have just basically gone out of the window because marketers are just using GPT voice?

Speaker B: Yeah, sort of a tougher question. If you are talking to a marketing analyst right now and they're, they're tasked with doing copyright. Any of these, you know, website copies, a big one. Um, and they're tasked with using AI in that regard. Is it matter of fact, if we're talking within the boundaries of this sort of 80, 20 rule, how do you approach it? Do you do all the copy in say an LLM? You have Claude write it all out and then you as a human come in and you know, go through it and check it and that's. Are you seeing the best results that way? Are you seeing some people I've talked to who they now call themselves purists but they go out, create all the copy, their own brain, their own words, their own hard work and then they use AI to tweak it. Do you have a perspective on, on um, each.

Speaker C: I think that these are preferences in style and workflow, to be completely honest. And different people have different takes on this. I think a lot of it depends on the volume as well. Right. If you're writing a kind of 20 page brochure site, sure you want it. Maybe it's consultancy. You want it to come across as the voice of the consultants. You want it to feel like you're working with real people. In that case you might handcraft and polish, edit, tweak with, with the LLM. If you're running a E commerce store where you're listing hundreds of products and you've got to ensure consistent word counts and style bullet points here, 50 word description here and et cetera, et cetera.

Speaker B: Super challenging. Yeah.

Speaker C: You kind of programmatically stick that in, stick a data sheet in and you could blast it through and it becomes a speed and volume based project. So I think it's horses for courses. You have to look at what, what you're trying to achieve and I can see strengths and weaknesses, um, for both use cases.

Speaker B: Yeah, yeah. It's super interesting because I uh, think this is the sort of big message here that is evolving is there's a lot to learn. We covered that. There's a lot for people to get up to speed on, there's a lot of different tools to play with and then there's a lot of different ways to implement those same tools. Right. And so as much as I know I'm sort of crazy and hell bent on learning every little corner of AI and all the tools, I do have empathy towards folks that have been sort of taught one way, taught an angle and now they're sort of thrust into, to this new world of Just the complexity. And so I'm sort of trying to get into the psychology of this push to let's just AI the hell out of everything, right? Like, you don't need me as a human to do this. I got too much on my plate. Let's just do it all. And I think the sort of, on this point you're really strong about saying, hey, you got, especially if you're in with clients and you're wanting to make sure that your brand is represented well. Like these are all things that still require human. They still require the oversight, if at all. Just someone to check and make sure that things are being done the right way.

Speaker C: Well, you see examples of it with some of the big four consultancies landing in hot water with AI generated reports for clients where they're having to issue refunds because the reports have got fake citations and rubbish in them. It's so, it's so basic. That's your product. Your product is your expertise and your consultancy. The idea that you go, well, we'll just, I mean, how. Talk about devaluing your product, um, and your whole positioning if you just go, well, we're just going to throw it in ChatGPT and it can do it and we're not even going to check it.

Speaker B: I mean, yeah, there's all kinds of horror stories that I think are uh, coming out as fast as the success stories, right? Because people are learning like it's not a all for nothing. Is there, is there anything in this sort of topic that you think is still a pretty human aspect in the marketing realm? Is there anything that you just sort of have a stake in the ground and be like, yeah, don't ever put this out to, to an LLM.

Speaker C: No, there's not. And the reason that there isn't because I think it varies on circumstance. Different organizations are going to have different expectations and the market will respond in different ways for different things. I was speaking with a client the other day who they'd identified there was something of a bottleneck in some of their, their design team. They were under Resourced for Design, but they were trying out things like Copilot, Create, ChatGPT with GPT Image 2 and Nano Banana and all of these things. And they're fun and they're impressive and they can do cool stuff, but they always found that it wasn't quite on, um, brand for things that they wanted. There was always maybe a background texture that wasn't quite as it should be if they gave it to the graphic designer or that little bit of Kerning on something was not quite right, whatever it may be. And um, we had a discussion around it and said well that's where they left it was maybe actually what we need to do is rethink how we approach design for different applications and say for a LinkedIn post and you know, if it's a client testimonial where we are slapping some words on the background sort of thing that you might churn out in canva very quickly, we'll use it for something like that, for a piece of content that maybe has a shelf life of 36 hours at uh, most and uh, reach of sure it's, it's important to get good reach. You don't want to devalue the content, but we're not going to pour our heart and soul into it. But our designers, when we go to our exhibitions, we want to make sure that we've got the great motion graphics and we want to make sure that our stand design is amazing and we want to make sure that that printed collateral that lands on someone's desk, that feels premium and then the two things can exist in the same world in the same team. On the flip side, there'll be other companies that say, you know what, I'm churning out loads of print collateral and I'm going to make it in ChatGPT. And well, in fact if I go to my local community center, there's posters on the walls all the time telling me about a uh, yoga class or something. And it's a different application, it's a different market, it's a different use case.

Speaker B: Right.

Speaker C: It's great for that. But would I use it for a high end premium B2B product? Maybe not.

Speaker B: Especially if you're working in a huge company of, you know, where there's a pretty substantial marketing team. The idea that that company's brand is tied to the quality of their materials, the quality of their colors, the quality of their design, their quality of that sort of implement, how the brand shows up, super, um, important. You don't want to outsource it. Now this is actually goes back to one of the key things I think marketers really got onto. And uh, I remember this because I have some friends that are artists and artists sort of freaked out when AI came around because they're like, you know, are we done? Is AI going to, you know, mid Journey was having his heyday and there's this sort of conversation around, hey, does how does AI affect design? There's the principles and the elements and the components of it all can be Very much ideated via AI. You and I can go to a Claude or Gemini or Copilot and get all kinds of crazy ideas. We want this color, we want this statement, we want this font, all of these elements. These two work together great. These two don't. This one conveys solidarity, this one conveys boldness and new ideas, whatever. Yet the componentry, the architecture, the putting all these stuff together in a cohesive manner actually seems pretty human. Right. Seems like, um, AI could probably take a pretty good crack at that. I'm not naive to think that AI can make some inroads on that, but there's still something about the fact that the branding is influencing humans and that humans portraying that is still the way to go.

Speaker C: Yeah. And I think, um. Yes, yes, but.

Speaker B: Right, please, you are always allowed to disagree. I expect you to.

Speaker C: I really value the craft. So my, a lot of my initial work in, in marketing agencies was within a creative brand agency. They did work that AI is never going to be able to do.

Speaker B: Right.

Speaker C: And there is, there is a finesse to it and there was, there was a quality to it that I do not expect AI to be matching. There was a price tag attached to that as well. And I, again, I think we will see that. Okay, so Claude Design just launched recently. Claude Design is what, a month or two old now? At this point? It's really, I mean, it's not very old at all as we record this. And that product from Anthropic isn't at the moment amazing. It's kind of cool. I've had a play with it. I've gone, I do like this. But I remember when Claude Code launched and I went, this is kind of cool. It can do some interesting things. But you know, I'm not going to live in here. I'm m not going to spend all of my time in the CLI interacting with Claude Code. And now do you know what I do? I spend all of my time in Claude Code. Right. And the difference. And I think that is because of the model improvements. Right. So when Claude Code launched, it was Opus 4. Was it? And I think then Opus 4.5 came out and everyone went, oh, this is interesting. And I think by the time we get to the five series of Opus, I think Claude Design will make people go, maybe these things that we thought were human domains after all were not so human because we've seen it in plenty of other domains so far.

Speaker B: I think this is for people that have been doing this. I think this is our default response to these things. Like Maybe there is no definitive anymore. Right. It's uh, just a matter of time. It's just a matter of time until, you know, Cloud or Copilot or Gemini or OpenAI gets to that frontier. Um, as a matter of fact, I think OpenAI sponsored something where it sort of has a graph of all the, the, the spider chart of all the areas that it sort of getting into and where it sort of ranks itself on a, I think a scale of 0 to 5 or something like that, which was super interesting because there was, there's a lot of stuff where, ah, you know, it's not going to get into plumbing very soon. It's not going to get into, you know, a lot of these mechanics. Right. But you, you bring in robotics and maybe it's just a matter of time. I think this, the interesting thing and I want to sort of switch gears here, but is connected is in the technology space. This is another area. I think marketing has sort of been on the forefront of a lot of technology sort of landscapes. Do you feel like they're, that's still the case? Are they still at the forefront? Are they, or do you feel like we're, they're, they're starting to need to catch up a little bit?

Speaker C: No, I do think, I do think that as well. Um, there is, they've been always trying to find the competitive edge, right? So you look at things like search engine optimization gave rise to huge amounts of SEO tools to track your rankings, identify keyword gaps, tell you where your competitors are doing well and give you all those kind of insights. So you certainly stick that into your tech stack. Then you've got PPC advertising, so you need the tool to manage your pay per click, manage your budgets, to do dynamic bid pricing, to a b test your copy. So you buy that tech stack because that's integrated into, into Google Ads and then you've got your email marketing and you need to manage a database and, and, and, and so as channels have become more digital themselves, so have the tooling to, to go and service those, those channels. I think it kind of is a natural extension of that. So the rise of digital marketing led to the rise of all of this digital tooling, which led to a lot of marketers being quite digital. First someone's got to manage the website and integrate it with the analytics system, the forms, the CRM, the so on and so forth.

Speaker B: Do, do you think this is the sort of landscape of skills that, that marketers need to be paying attention to? Is it simply like, hey, learn everything you can about Claude and copilot. Or is there other skills? Because I've been in some conversations where they're saying, yeah, spend some time in AI, but the ability to think strategically, for example, is actually more important. Do you have a play on that? Like how. How would you. If you got a. Had to get me up and running as a hardcore marketing analyst? Digitally forward, of course. Like, where would I. Where would I put my eggs? Yeah.

Speaker C: I think there's still the foundations of good marketing principles. The four P's of the, you know, the marketing mix, right. Product price, promotion place, all of that kind of stuff comes into it. You've still got to have a good product and know who you're selling it to and at what price and how are you going to distribute it and all of that. So that's kind of core. You've got to have, have that. Beyond that, I think so not to skim over that. I think that having the general marketing fundamentals is critical. Beyond that, I think the future generations or someone looking to get ahead in marketing now throw yourself into AI literacy in the round. Play with these tools. I've been speaking to hiring managers at my clients in the past couple of months and in a team with. I was working with a team, I think they had around 10 or 15 marketers distributed across a few sites. The hiring manager was saying, for their next hire, they want someone that's got really good AI, uh, literacy and someone that had very good cv, maybe even a better CV on paper than another candidate that had good AI literacy would miss out. Because the AI literacy is a, uh, skill multiplier and it opens up that person's ability to do more. Maybe they can't do data analysis, but they can throw a few CSVs into a tool and throw a few charts together and read the charts and go. There's an insight to be gained from that. Maybe they can work quicker, they can iterate, they can ideate, they can proof

Speaker B: the AI literacy is not just using the tools. You gotta learn how to open Claude, write a prompt. Those are all good things. But AI literacy, as you're sort of feeding it and hinting at it, is understanding why it works the way it does. And that seems to some where that feels overwhelming. But AI literacy is really about understanding all the stuff you and I talked about. Where does it break down? Where is it really useful? Where is it good to use, where is it bad to use, where is there legal concerns, whereas there sort of governance concerns, privacy concerns, all these. Like, that's AI literacy. Is there anything else you would add. That's super important there.

Speaker C: Yeah, I think that's a, uh, really critical part of it. The AI literacy is about understanding what it can and can't do from a capability perspective, but also what it can do in terms of risks it might introduce into your business, into expectations you ought to have, with its ability to successfully and consistently. Consistently being the key word there, uh, perform a task over various time horizons. Having that understanding of the failure modes is a key, key part of it. And I think you only get that through experimentation. Technology is, I think Excel is a, is a good analogy in so much as it's a very, very versatile. You go into Excel, it's rows and columns and cells. But what do people use Excel for? People use it for financial planning, for making birthday card lists, for making to do lists, for booking holidays, for budgeting,

Speaker B: like building out applications.

Speaker C: Yeah, it's super varied and everybody finds their own use case for it. It's an incredibly versatile technology that you get more from the more you play with it. And I think LLMs have that similar capability. And if you're coming at it, if you've never put a prompt into a language model and you go in today, there is a decent amount of catching up to do. And I think sometimes if you're in the weeds with it, like I am and you are, ah, you might take for granted that they have tool calling now. And I mean, tool calling wasn't a thing when, when ChatGPT first launched.

Speaker B: Uh, I think the tool calling is a good one because I love the look on people's face when I'm like, oh, yeah, it checks my emails, it can manage my calendar like these, it can do these things. But at the end of the day, I understand and I look through and I pay attention to what are the guardrails. We had a podcast previously and I have a, uh, couple scheduled that we're talking about governance. We're talking about those things because they are part of this concept of AI literacy.

Speaker C: Where do you think that sits within an organization? Because I think this, for me, this is a really interesting one. IT teams that I'm coming up against aren't all over this, right? This, for a lot of them is a, um, burden. It's another thing to manage and roll out. And they're not really sure of it themselves, but they know that they've just read headlines about this thing called Mythos, which might be coming down the track that's going to be giving them a cyber security headache. And now they've got These people that want a, uh, Copilot Pro license or uh, they want a Claude code deploying into their Microsoft tenant and they're scared because their natural position is risk mitigation and minimization a lot of the time. But then you've got the marketing teams or the. Obviously this conversation is about marketing, but it could be another department, hr, finance, whatever, that have seen a demo of it and gone, you know what, let's get the shadow IT system, let's stick it on our departmental, uh, credit card and get ourselves a couple of anthropic Claude Pro licenses and see what we can do. Uh, like you are 100% right, that governance is a huge part of it. I just think it's a messy world out there right now.

Speaker B: Yeah, if you've been around 15, 20 years. You know, we didn't really get solid into the sort of data governance conversation until then. I mean, we're still like the idea of true data security and true IT governance much better than it ever was. But they're, they're still fighting the good fight. Like they're still figuring out new methods, new frameworks, new best practices. And that data is a very complex thing, but it's a pretty well understood thing, right? Like we, we know what a database is, we know what a data warehouse is, we know what those technologies do, we know how systems connect to data, how they could be compromised. Like these are all sort of the known knowns versus the unknown unknowns. And literally one of my other guests said this is like AI governance is not the same as data governance just because it's got the same word. The concept of it, of being able to manage it effectively is similar, but it's not a one for one. And I think IT teams just either they knew that right off the bat or they're learning very quickly that it's not a, uh, it's not an easy task. It's, it's absolutely not because there's CIOs in the world that are spending an absorbent amount of time and effort and capital on, on trying to get their teams hard secured, build up their security infrastructure, their, their capability. They're bringing in the right people. And now you're like, oh yeah, yeah, you. While you got the data governance, let's go. We got a new one for you called AI. Good luck with it. You know, but it's one of those things that has ramifications to our, you know, can bring it back to our point, is a single user not having the AI literacy, not using AI Uh, in the sort of right way, can be dangerous. Like they, it just can't, can't be.

Speaker C: There was a great story. I read it this afternoon before recording this. So the NHS in England, the National Health Service, they've instructed their data and digital team to turn all of their GitHub repos private and that there should be no public repos except with exceptional requests and use cases. And the reason that they've put that in is because the rapid increase in the use of AI is meaning that people are pushing to repos with vulnerabilities or, you know, they're leaking API credentials or whatever it may be, um, and they are introducing all sorts of potential risks and the ability of AI to read those repos and identify those risks has, has gone up tenfold. And I think that takes us back to that point about the. So I was talking about the generalist marketer, but I think this is the case for the generalist, whatever, right? If you work in hr, if you're a data analyst, if you work in sales, having AI as a skill is going to be a force multiplier for you. If you've got the ability to use Claude code or Claude cowork or codecs or whatever it may be, you are going to be able to get more done. However, at some point you're probably going to find yourself straying into territory that you're not super, uh, familiar with and going, you know, it's that classic with Claude code. Just hit accept. Yes, yes, yes, yes, yes. And you're going to commit something to a repo that has your credentials in it, or you are going to draft a contract with a clause in it, uh, which is going to come back and bite you in the ass. And you have to be the 20% of the human that actually still checks and validates and makes sure you know what you're talking about.

Speaker B: Yeah, I, I, I love it. I, I think this is the sort of battle that people are going to have because you have the pressure from the company, pressure from the bosses, pressure from, you know, just society in general to perform and get this stuff out the door and ship it. We've always sort of had that. Now it's at a fever pitch. Be warned, if you're listening to this, Martin has warned you you have no excuse. He's told you that, like this stuff, there's stuff that can happen, there's real consequences. And that's something, um, I tend to think about nearly every day. Every time I'm interacting is like having that awareness that uh, understanding that this stuff can go south, it can hallucinate. Hallucination is still a thing. The other one that I know, just getting into the weeds a little bit, is just the. It's hallucination, but it's that confident hallucination. AI will tell you it did something, and it absolutely did not. Oh, yeah, I created a spreadsheet for you. It's got 15 columns and four, uh, hundred records. And you open it up, there's literally nothing in it. Like, it's that, that sort of thing requires you to check it, you know, requires you to get after it. And so I think, you know, if, if people are walking away from this podcast, we're advocating for learning AI, we're advocating for literacy and understanding how it works. But I, I want to sort of get, get into another section where if we fast forward 6 to 12 months. Where do you think marketing goes?

Speaker C: I think that's, uh, an incredibly broad question just because of how much marketing encapsulates. But I think that I think what we see within teams is AI first teams. And by that, uh, I mean teams with real people doing real work that are AI enabled. I think they will be moving faster, particularly around areas of, uh, research and analysis. This is the biggest win that I see when I go into themes, is just showing them how quick they can move and how quick they can present. I've done workshops with teams where in the space of two hours, we've, from start to finish, kind of done some research, built some interactive dashboards or portals and had something that they felt that they would take to another stakeholder, uh, that would previously have taken them a week or more.

Speaker B: Yeah.

Speaker C: To pull together. That is a huge, quick win. So I think that's a big part of it. I think the other part is on the technology side itself. Yes, there'll be more people doing more with the likes of Cowork and Codex, which can operate within the computer use space. So for listeners that aren't all over in this and in the weeds, computer use is the ability of an AI to actually use its own computer or maybe use your computer to do tasks, Open files, copy files, click around, do stuff. And this is one of the areas where language models are getting a real focus at the moment. They are getting a big development boost at the minute. I think this is going to do more tasks, more really simple tasks, and I think people will be elevated. I, uh, will posit it like this. It'll be elevated. Elevated away from some of the more mundane tasks that they do. But then in terms of the technology like the Martech stack specifically. Uh, Now I'm a HubSpot partner. I've been working with and um, reselling HubSpot since 2012. But if I look away from HubSpot and into the general stack, agentic use within the platform of choice is going to be massive. And I think we're going to see this with HubSpot have a thing called um, customer agents. They've got prospecting agents, we've got. I wouldn't be surprised to see similar agentic campaign development and creation coming out in the likes of mailchimp. Manus from Meta or uh, Manus which was acquired by Meta, has um, a meta ads agent management tool now. So it integrates with your Meta Ads platform and it will do campaign optimization, it will do end to end management of it. I think we're going to see a lot more of, of that and that's probably in the six month timescale. I'd go to the end of 2026. Beyond that, who knows? I mean we could have models that uh, are brand new with whole new capabilities by that point I think in house teams, this is another interesting shift. In house teams will be able to, with agentic tooling, bring some things in house that were previously outsourced to specialist agencies. So if you've used a PPC agency, I think PPC agencies are going to come up against it. What do they charge on? They have domain expertise of the channels with budget allocation and bid optimization and things like that. And then they know how to navigate the Google AdWords platform or whatever it may be. You know, the programmatic platforms. That feels comfortably within the computer use space. For me that feels pretty, pretty doable.

Speaker B: Anything that requires a distinct level of knowledge, the power of the person knowing a thing is sort of being undercut. It just is. It's, it's the ability for an AI system to learn that thing. Personally I sort of stepped back from like, oh this is definitely going to happen and this is definitely going to happen and these people are definitely going to be you know, impacted because you even said yourself like outside of six months, who knows? I do like being a sort of futurist and imagining what five years out will be and how big my house on Mars is going to be. But like at one point you're like, I have no freaking clue. This stuff is moving. So the models are moving fast. Software that is built on the models is moving fast, the capabilities are fast, new technologies coming out or like good luck, good luck. All right, so this has been easy so far, but we got what I call Lightning Rand. I have. I think I have five questions for you, but they have to be first response, quick, instinctive answers. No, it depends. Or sugarcoating it. Or. Oh, it could be this, this and that. It's like what your gut ran.

Speaker C: Yes or no? Or am I allowed a small experiment?

Speaker B: You can elaborate. You can elaborate. But. But it can't be a, uh, flimsy answer. It has to be sort of stake in the grand. Okay.

Speaker C: Piece.

Speaker B: And we'll keep it brief and quick.

Speaker C: One of my answers now with a, uh, this might not be my actual position. If you want nuance, come and speak to me on LinkedIn.

Speaker B: There you go. Good. See, he's smart. You're smart. Spoken like a marketing person. Like. All right, so the first one, I'm going to go slow and then speed up, but one task AI should absolutely own in marketing right now.

Speaker C: Right now, for me, it's research. Research for, uh, nearly every client.

Speaker B: Good. One task that must stay human.

Speaker C: One task that must stay human. Ah, editing.

Speaker B: Good. Yep. We talked about that one skill marketers need to develop right now. Sort of talked about it. But what one skill specifically?

Speaker C: I would say so I don't want to talk about just prompt engineering or something like that. That kind of feels table stakes. Right. I think the understanding agentic and getting familiar with that. Uh, so for me, that would look like dipping your toe in first with building copilot agents or you even, even like for real basic stuff, like get familiar with chat GPT projects or Claude projects or Gemini gems. They're a really easy entry point, but do it.

Speaker B: Yeah, I. So what I do is I. And maybe this helps folks on the call is if you get into Claude cowork. Claude cowork is. I think of it as the mashup between the sort of Claude that you would prompt and cloud code that you would, you know, build out applications. It's sort of the middle ground there, but very prompt, friendly. Literally write a task, say, hey, I want you to create this Excel sheet, but tell it to use an agent. Tell it to say, hey, I want you to deploy three agents to research these three different tasks and bring it back as one paper, one PDF that is well structured, formatted, blah, blah, blah, and you'll start to. It is not what it is agentic. But it's not like when sort of advanced practice of agentic AI. But it gets you to understand that you're now deploying one prompt, seeing three different bots go out and do these things come back and you're going to see the nuances. You're going to see if they all wrote exactly the same thing, different things, all that. Cool. Next one, we got a couple more. One AI tool or trend you're watching closely and I would even add you're cautiously optimistic.

Speaker C: AI Voice Agents is definitely the one.

Speaker B: Yes, absolutely. That's a, that's always a big topic of conversation. All right, last one. One mistake that you think most marketing teams are making, I'm going to, I'm

Speaker C: going to expand that. So I think one mistake that I think most AI users are still making is they are not being deliberate in their use of reasoning and thinking models.

Speaker B: Ooh, there you go.

Speaker C: I think most people are using Copilot with auto mode turned on or they don't even know what extended thinking or reasoning mode is. Um, and if that's it.

Speaker B: Absolutely. Tell just we have a bit like, give me 30 seconds of why that's. Why would you say that?

Speaker C: Why is it important if you're unfamiliar with this? Instant mode in a language model is where it gives a response straight away and, uh, what's going on under the hood next? Token prediction is happening immediately. And it's a bit like I say to you, what, 956 divided by 72, go give me an answer. And you've just got to say a number as quickly as you can. You might get in the ballpark of that, but if you really want to work it out, you need to sit down and write out the steps involved and do the maths. Well, extended thinking or reasoning mode, is it doing the thinking beforehand so you can give it a tough question and it will go away and think. And it can think for a few minutes, 20 minutes. I mean I've had it. The longest one I've ever had was 55 minutes where it was going away and doing one particular task, which was crazy. And this is before it gives a response. And the quality of the output is night and day. I liken it to instant mode is kind of clever. Guessing and reasoning mode is like your well informed, well read friend taking their time to carefully consider their response to you.

Speaker B: Yeah. So good. Yeah. So the message being spend some time in understanding these different modes. And every model has sort of most of them have extended thinking instant, but there's, you know, different models that have different capabilities. And I think I probably have to have a whole new episode on just token utilization and understanding the use of tokens. That's. That's a hot topic everywhere I go But Martin, thank you so much. I absolutely love that you were able to really open this window from the marketing world and AI. Like, that's super interesting, super fun. I'm going to share this with all the marketers that I know. Thanks again. I appreciate your time.

Speaker C: Thank you for having me.

Speaker B: Absolutely.

Speaker A: I love speaking with Martin. What stood out is how honest he was about the state of things. Most marketing teams are, are not in the middle of some clean AI transformation. They're dealing with too many tools, half rolled out. Copilot, Claude, Gemini, pick your flavor licenses. There's not a whole lot of real training out there yet and there's no clear ownership of who governs what. And he kept coming back to this idea that people getting real value out of AI are not the ones with the fanciest tools. They're the ones who actually understands where it works, where it breaks, where a human still has to step in. That framing sort of reset how I think about AI adoption inside of teams, especially teams that are not exactly built for fast change. We spent some time and dug into the 8020 rule. A lot of different variations of this, but for this one, it's where AI gets you most of the way there, but the last 20% still needs a human. We talked about hallucinations and the gap between real true AI literacy and just basic prompting. They're different. We talked about governance and we also talked about why editing and brand level design still belong to the people. We also learned looked ahead at agentic tools, things like computer use and how in house teams are pulling work back from agencies. So please do me a favor. If you like this conversation, follow alter

Speaker B: everything, wherever you listen.

Speaker A: We're on Apple Podcasts, Spotify, YouTube, and I don't want you to miss the next one.

Speaker B: We got a lot to talk about.

Speaker A: Thank you for listening.