Caution and Innovation: AI in Modern MarTech Stacks

Hannah Kelly: Welcome to the Marketing Automation Moment Podcast. I'm Hannah Kelly.

Mike Maynard: And I'm Mike Maynard. Um, this is Napier's podcast to tell you about the latest news from the world of marketing automation.

Hannah Kelly: Welcome to the Market Automation Moment Podcast. I'm Hannah Kenny.

Mike Maynard: And I'm Mike Maynard.

Hannah Kelly: Today we talk about HubSpot giving AI agents access to its CRM.

Mike Maynard: We ask whether only young people can be AI natives.

Hannah Kelly: We talk about Acton's new data strategy.

Mike Maynard: And finally, we discuss whether you should let AI models control your martech stack.

Hannah Kelly: Hi, Mike. Welcome back to the Market Automation Moment Podcast. It's been a little while.

Mike Maynard: It's great to talk to you again, Hannah. And it's particularly exciting because we're in our new Napier podcast studio, so hopefully the audience can hear us and see us. But, you know, if we haven't got the tech right, it could be quite exciting.

Hannah Kelly: Definitely. Well, I mean, I love the setup. It's something that we did maybe a month or so ago. Um, and last year we actually recorded in our first ever podcast studio. And that was inspiration.

Mike Maynard: Yeah. And it's nice to be able to record, you know, face to face rather than remotely. So that's great.

Hannah Kelly: Absolutely. Now, let's kick off because we have so much to talk about and I'm really excited to get into our conversation. So the first thing I want to have a chat about is HubSpot. Now, actually, in the episode today, we're talking about most of the major market automation platforms. And then interestingly and not surprisingly, a lot of it is around AI. So let's kick off because I want to talk about HubSpot. Now, only a couple of days ago, HubSpot actually released the news that they will be introducing, um, AI agents into their CRM. I don't know. Did you see the news?

Mike Maynard: Well, as a geek, I was so excited by the news. I love a good model context protocol server like everyone else, I think. Um, so an MCP server is basically a way that a system can talk to an AI agent. So if you, for example, want to manage your HubSpot in ChatGPT, uh, or in Claude, then what you need is an M MCP server that interfaces between the two. And this is what HubSpot's doing. So people may be familiar with APIs, application programming interfaces, which is the way that different systems previously have talked. So if you wanted to interface, you know, for example, a Webinar system to HubSpot, it would typically send the data through the API. The MCP server is really an extension of that. And, uh, what HubSpot has said, which is quite exciting, is that uh, the MCP server, so any AI tool will be able to do exactly the same as you can do with the APIs. So you can, you'll have a lot of access into HubSpot through your AI tools.

Hannah Kelly: So what does this actually mean for the future of the platform though, Mike? Because actually in the article it talked about, you know, are, ah, they actually moving away from being a master automation platform and more just like this AI platform? I mean what does it actually mean for the future of HubSpot?

Mike Maynard: Well, I mean from a cynical point of view, I think it means that there'll be lots of stories that start with ChatGPT, deleted my HubSpot database. I think one of the things we've seen in the software industry is that where people are programming with AI, if you're not careful, you don't have the appropriate guardrails in place, then you can get into trouble, you can actually make mistakes. So I mean, I don't want to be flippant about this, but there is a possibility that AIs could delete lots of contacts. So I think people need to be a little bit careful and we'll talk about this later I'm sure, but what it means is you can effectively control HubSpot from an AI agent. And so this means that you can sit for example in ChatGPT and you can ask questions about your HubSpot database and even do things like set emails up using AI. Once the system is fully up and running, um, and you have all that access, you can then basically, uh, through the MCP server, say, create an email and do this. So there's an opportunity to firstly work as a human with AI and use natural language rather than having to click the user interface for HubSpot. But then following on from that, there's no reason why you can't have AI agents doing it automatically.

Hannah Kelly: I mean as a marketer myself, that sounds brilliant, it sounds really exciting. But I also think we have to be realistic and obviously we work in the B2B tech sector, uh, and we know people are very risk adverse and I know that that actually they sounds like they've got a mammoth job of their marketing to do to basically jump over these hurdles and make, you know, sound as cool as you've made it sound without all those risks that are associated with it.

Mike Maynard: Well, I think this is happening a lot. So what we're seeing is we're seeing a lot of different um, systems, say for example in software engineering, trying to allow people to access things in AIs. And it's going to have some interesting impacts. I mean the first is, is that potentially people are going to do some silly things. So that could range from deleting contacts accidentally all the way through to uh, creating endless streams of emails. And I think we will see like a lot of AI optimized campaigns that potentially may go a, uh, little bit off the rails. So particularly with small data sets, that can be the case. But having said that, it offers huge potential. You know, it offers the ability to, you know, really talk to an AI about your data, segment data using uh, natural language rather than having to actually go in and use filters and then send super targeted emails with much less effort. So I think it's really exciting. I mean the worrying thing for platforms like HubSpot is of course people then stop working in HubSpot and they start working in the AI. What's going to happen in the long term? Are people still going to be loyal to HubSpot or are they not going to care about what the database is behind it and they're just going to worry about how they interface in AI?

Hannah Kelly: I think that's a really great point. Mike and I actually want to move on and talk about another one of our topics. And this is around Salesforce because it really feels like HubSpot is leading the charge. But they aren't the only market automation platform in this thought process. Um, and that takes me basically to Salesforce's CEO actually announced on Twitter, or X whatever you like to call it, that they will be hiring a thousand grad native AI users. So this is basically people, the young generation, but they have to be AI, they have to know how to use AI properly. And I thought this was really interesting because you know, I was looking at and I thought, okay, that's a bit strange. And then actually I came across a complaint and someone's also been complaining about how in Salesforce they can't get hold of any humans, they can't get hold of their representatives, they can't get hold of the AI agents, they won't help them and they've actually had to go to the very top to get a simple answer to a product that they were being billed for. Now it kind of relates, but it's, you know, if they were just going down this AI route, uh, are we actually losing that customer service element of market automation platforms as well?

Mike Maynard: Well, I mean, I think there's a lot of questions in there. The first thing I'm going to do is Start the whole age debate here. You know, I remember it was only what, two or three years ago where people were talking about hiring young people because young people are digital natives. And I read a great quote. It's like talking about young people being digital natives is simply a way for old people to be lazy. And I don't believe that young people necessarily all have a huge advantage in terms of thinking about AI. In fact, there's been studies that shown that older people quite often use AI more frequently and more effective at work. So to me, I think, um, Salesforce has come out with or invented this term, uh, AI native and then tried to create a press release about what they're going to do around the term. I think it's a lot of it is marketing. Salesforce obviously hire a lot of people. They're a big company. I mean a thousand people in the scheme of Salesforce is not actually a huge number of people. So I think they're hiring a lot of juniors and they're hoping that AI is going to help those juniors upskill a lot quicker. Um, so that's my cynical view on age. I don't know if you agree about that.

Hannah Kelly: You know, actually speaking as part of like the younger generation, not quite Gen Z, but I actually say I have to agree. But also as well, they're not as native as you think they may be. And this is because, you know, universities, colleges is they actually ban the use of AI in work. So yes, they might be using it in their day to day, but they're not actually using it where it counts. And so I think there is this misconception of like, well, the younger generation know all. They're going to be so skilled, so equipped, they're going to know how to use the different AI platforms. I guarantee a lot of the younger generation don't actually know that there's other AI platforms outside of chat GPT.

Mike Maynard: Yeah, and I think it's a great point. I mean without doubt there's some young people who are incredibly talented and really understand, uh, AI. So I, I think it's unrelated to. I think you can be really knowledgeable and proficient at using AI no matter what your age. It's people who've really embraced the technology. I think that's Salesforce's mistake, associating it with a particular age rather than an attitude. And then let's go on to the comments about Salesforce and the issues which you picked up. I mean I feel sorry for Salesforce, but if you run the world's biggest CRM, you're going to get criticized. And I think one of the comments in the LinkedIn post you referred to was that when they finally got an answer, having failed to get anything from either a human or an AI, um, agent, they actually asked when the renewal date was for the Salesforce platform. And the inevitable comment was, maybe they should get themselves a decent CRM so they could track that. I think it's difficult and it really shows that, um, actually, this CRM and this marketing stuff, it's actually quite hard and it's really easy to get it wrong. And if Salesforce can get it so wrong, obviously everybody else can. So whilst everyone's getting excited and rushing towards the new technology and trying to integrate AI and doing all the cool, shiny stuff, actually it's also a great reminder that we need to go back and really think about the basics and make sure we get the basics right, don't you think?

Hannah Kelly: I love that, Mike. Get the basics right. I couldn't agree more, and I love that point as well, is that it's easy to make mistakes. Salesforce is this huge company, but if they haven't got that core right, then everybody's going bad press.

Mike Maynard: Yeah, absolutely. I mean, you know, with all their resources and all their expertise, if they can't get it right. I mean, if you've had an issue with Martech and you've had a problem and it's been down to not getting the basics right, maybe, you know, give yourself a little bit of slack here because everybody struggles. But equally, you know, I would say as a marketing professional, if you're worried about your Martech, don't go rushing off and spending all your effort on trying to integrate AI into all of your systems and your processes. Actually, go back and think about the data you have, how it's available and how it's used.

Hannah Kelly: Well, I'm going to put you on that point, Mike, because you said the word data. And I don't know if you want to talk a little bit about the Acton report that we saw around their data strategy.

Mike Maynard: Yeah. And I think, in a way, you know, act on is very similar to HubSpot. HubSpot's always had a great AI. I mean, at Napier, we've done a lot of work, you know, interfacing into their API and helping automate things. Acton's not been quite as forward in that. So whilst HubSpot is talking about creating an MCP server, they, I don't believe have anything available yet, but they're talking about creating this MCP server to link into AI Acton is kind of playing catch up and they're building a better and stronger API. And obviously one of the things you need to do is if you want to get this, you know, AI interface and these MCP servers and sorry about the geeky language, but you know, if you want to do that, you've got to get an API that works first. Uh, so act on is putting a lot of effort into doing that and catching up. I think what it is is it's a reflection and this is probably less of a dig at Acton than It is of HubSpot. Sorry guys at HubSpot, but HubSpot tried to position themselves as the ones platform you need for everything and clearly that doesn't quite work. Clearly HubSpot recognizes they need to interface to different systems. And so we're seeing more and more this idea that, you know, the kind of one stop universal solution isn't quite enough and people are having to use multiple systems and we're getting to the position now where those systems are integrating and talking to each other rather than needing a human in between to kind of carry data or indeed being completely separate. And I know, you know, in the podcast before we've talked about the issues of, you know, having multiple systems and not being able to pull reports together that that's a huge problem when things don't talk to each other.

Hannah Kelly: Absolutely. And I think you're right. You know, act one is playing catch up a little bit. They're ones that haven't maybe been as. Not eager is not the right word, but not so intense on like let's get this AI, let's get the API set up, let's do this. So it is really good to see as well, you know, not necessarily the smaller, but they aren't, you know, as big as ones like HubSpot or Salesforce really trying to implement and catch up on the AI side for their customers.

Mike Maynard: Yeah. And I think, you know, APIs have become more and more important and we can talk about, you know, the way people are vibe coding to um, generate little bits of software that link things together without necessarily having a huge amount of software development knowledge. I think that again is pushing this need to give access to the data, ah, to outside systems. So again, maybe it's a bit of an AI story.

Hannah Kelly: I definitely agree, Mike, but let's move on because I want to talk about something a little bit different now. And this is a report from Martech. Now martech, if you don't know, actually, actually report on the marketing technology landscape. So these are all the tools that can be seen in the martech stack. It tracks what people are using. And, um, we actually came across some interesting stats, so that they've just done some research, they've just analyzed what, what are actual people using. And it's actually become quite clear that people seem to be actually canceling subscriptions in their martech. And you think, great, people are finally condensing, they're finally using like, you know, what they've actually got. But actually the same amount of people who are canceling are also actually just replacing those tools with different martech platforms. And I think that's really interesting because, you know, we've talked quite a bit on the podcast about, you know, you have to condense that martech stat, stop using all your budget on things that you're not using. And it seems like people are being like, this isn't working for me, but actually this new shiny object would be great. And I do wonder if part of it is, you know, perhaps GEO tools. We've talked about this before, you know, but I don't know, what are your thoughts? Because to me it just seems that, you know, us marketers are a bit like magpies and are going after the next shiny platform.

Mike Maynard: Well, I think it's interesting, I mean, they said two things. Firstly, that the number, uh, of new martech tools is not growing as quickly, which is probably not surprising. Although, as you mentioned, GEO is an area where we're certainly seeing more. But also this situation where previously martech, uh, users were signing up for a lot more tools than they were actually canceling subscriptions for. And this year the balance is even. And I think we've got a bit of sanity coming to the market now. People, as you say quite rightly, are thinking of, um, you know, how they use tools much more than just randomly going out and, um, pulling in every new exciting tool. I mean, I'm interested. You're our martech buyer at Napier, so, you know, what's your view? Are you looking to buy more new tools or are you more thinking about what tools can we cancel?

Hannah Kelly: I would actually say we're looking at more tools that we can cancel. And this is something we've done quite recently. You know, we were a bit like, we're using all these tools. Is it actually giving us the benefit? And we actually have to be really honest with ourselves. And I think as a marketer that can be really hard sometimes. But you want this tool, you go, this is amazing. And you know, with my role, I, I get to try the new exciting Tools. I get to try it to see if it's any good for our clients. And we've actually had to be like, no, we don't use that, or, oh, God, do we still have a subscription for that? And it's a bit embarrassing in times, but it's also allowed us to really look at our tools and go, okay, we've got some fantastic editing tools for podcasts. We actually use some great AI platforms to help us generate our videos, um, help us generate our social media snippets. It's allowed us to really go, okay, are we using our market automation platform to its full potential? What can we use of API to really bring in and enhance the data? So, you know, we actually use Apollo. That sinks into our Shop3 multimation platform, and that's really helpful for us to enhance our CRM. And so it's allowed us to actually go, actually, I'm going to pat myself on the back. And our tool stack is actually pretty cool, but it's also generating exactly what we need. But we have fallen fit into that. Oh, this is really cool. We should try this. And then actually not admitted soon enough that it's not working for us.

Mike Maynard: Yeah, I mean, I think, you know, to be fair, as an agency, we always want to try the latest tool anyway because all our clients want to know how well it works. And, you know, if we're not trying these things, we can't advise the clients. But I really agree. You know, you're right, what we're thinking about now is not features or, you know, the number of tools we've got, but much more effectiveness. So, uh, it's much more about optimizing what we've got and spending the time and the money on things that work and maybe not wasting time on things that are much less effective. And I think that's really important. You know, our market is very specific in terms of who we're targeting, you know, B2B marketing professionals and clients. And some of the things that will work for marketing chocolate to consumers aren't going to work for our market. So I'm really glad. I think we're focusing on efficiency and we're focusing on results now.

Hannah Kelly: Absolutely. But, Mike, I'm conscious of time, so let's move on to our top tip of the week. Now, one of the things that we wanted to talk about, we've talked about a lot, is about AI agents. And I wanted to ask and have a bit of conversation about whether actually companies should be allowing AI agents to access their Martech stack. You've talked quite a bit about API, you've talked quite a bit about the different platforms talking to each other. You know, we've talked about this in the sense of the market automation platforms. But if you look at the martech stack as a whole, is this something companies should be doing?

Mike Maynard: So this is really interesting and I think, you know, there's kind of, um, a range of answers. So there's a concern about confidentiality. And obviously if you're um, using a public AI model, those AI models train on data that they're given in the prompts you supply. So if you're pulling data in that's confidential and you're using a public model, you've got to be concerned. So I think, you know, where people have their own private models that works really, really well because you've got control over the data. And I'm sure what we're going to see is for things like, um, martech more and more push to having, you know, your own private model, whether that's um, on premise or in the crowd, but something that's owned and controlled by you, rather than just going to Claude and trusting Claude to do everything.

Hannah Kelly: But how easy is it to do that, Mike? So if you want to build something yourself and you don't want to use something like claude, is that like, feasible for like smaller companies? Like, what kind of capability do you need to have?

Mike Maynard: Well, it's a gig, you know, it doesn't matter how easy it is, it's just fun. Um, so I think the answer is that it is possible to run AI models locally and more and more people are doing that. Um, but the issue is you need some decent computing power. And depending on how much you're trying to do, if you've got a very large database, um, and you're looking to do a lot of analysis over a lot of, um, contacts, it's quite hard to have that level of computing power locally. So you might choose to go to the cloud, but then you're starting to spend a lot of time, you know, a set of servers and you've got a big IT project that's very different from buying like a Mac Studio, which is currently the highest end Mac, and just running a local model there, which maybe for an SME would be okay.

Hannah Kelly: Well, I'm actually going to say if any of our listeners want to speak to our resident geek over here, please do email him. We'll make sure that it's in the show notes, um, because that's something that I'm actually really interested in. But I think you'll be the best one to give advice to it, Mike.

Mike Maynard: Well, I'd love to talk about that. You know, about that. I think also the other question, and you know, this is the perennial problem with AI models is, um, what, uh, the AI, uh, companies like to call hallucinations and what I like to call mistakes. We've just had something. I mean, we obviously use AI in outbounds. You know, remember we're working through this project and I had to talk through the fact that, you know, a, uh, image sensor, an image sensor module are two very, very different things. And missing out module, which, you know, I've got to be honest, the AI, um, model, uh, that we're using, the LLM kind of didn't care about actually makes a huge difference to what you're talking about. And it's a completely different product. And for engineers it could be different. So it's really hard to get AI in some markets, and that's the markets that we and our, uh, clients work in, particularly to get it precise and accurate. And engineers really care about that. Again, you know, if you're trying to persuade people to eat chocolate, it's probably an easier thing to do because people are much freer with language. This doesn't have to be so precise. So I think those AI issues, they're going to produce some really embarrassing poor emails in certain sectors. And then of course, you've got the really bad hallucination issues where the AI decides to go and delete your database. You know, and I can see people sitting there using HubSpot through the AI and just typing in something like, please delete all contacts that, um, aren't delivering a good return on investment and they find there's one contact left in the database. So, you know, I think you've got to be really careful about how you roll AI out. And the way, you know, I always recommend doing it is sandboxing it. So you run trials, you run tests, you run things that are not on the live data. Uh, well, before you actually go and actually start using live data, uh, and that's something, you know, we're seeing a lot of people doing. And I think it's going to happen more and more as it becomes more important to use more, uh, AI. I mean, you know, the last thing to say from my point of view, I don't know if you agree with this, Hannah, is that, uh, AI works in a lot of cases. You know, whilst there's problems and issues, AI really does work. And so it's important not to look just at the, uh, disadvantages, but also to make sure you take advantage of the benefits. Don't you agree?

Hannah Kelly: I fully agree with that, Mike. And you know, I think just to add and to finish, I think those mistakes come when people and marketers maybe get a little bit cocky and they let get a little bit like, AI is brilliant and it's just being realistic. You know, we use it so many ways. We see our clients use it in so many ways that, uh, is really, really successful. It's just being realistic with actually what it can achieve and ensuring that you're just not relying on it solely to get the result that you want.

Mike Maynard: Absolutely. And I think, to be honest, you know, we've got to say, let's go back to Salesforce and they're hiring these, you know, young, inexperienced, um, AI natives, as they're being called. Some of those people are going to make big mistakes and lack of experience with a human is, you know, equally a problem as lack of experience with AI. And I think, um, you know, whilst we can sit here and we can be concerned about the issues with AI, we've got also got to understand that people make mistakes as well. So I think it's something to remember, you know, if you're bringing in, um, you know, very young AI natives, don't give them unlimited rights on your martech stack until, you know, you've trained them.

Hannah Kelly: I think that's a brilliant point to end on, Mike. If anyone has any questions about what we discussed today, please do reach out. You know, we're always happy to have a conversation, but otherwise, thanks for such a great conversation, Mike.

Mike Maynard: Thank you, Hannah.

Hannah Kelly: Thanks for listening to the M Marketing Automation Moment podcast.

Mike Maynard: Don't forget to subscribe in your favourite podcast application and we'll see you next.