Subscribe
Share
Share
Embed
Demand-Geniuses is the podcast for revenue-focused B2B Marketers. We bring you the latest insights and expert tips, interviewing geniuses of the B2B Marketing world to bring you actionable advice that you can implement to accelerate growth and progress you career. The role of Marketing in B2B go-to-market strategy has changed drastically. It's more important to revenue generation than ever as buyer engagement becomes more digital. We equip you with the information you need to thrive in this new, revenue-critical role.
Tom Rudnai 0:00
I'm an ordinary marketer. I know that it's my job to make sure that we're using AI as well as we can in order to achieve my goals, but like we come back to that issue of how do I stay on top of all of these, on top of everything that I should and could be doing, prioritize it alongside my day job. So I think that's probably a good place to come back to. What's your advice to a person in that position about how to start, and how you would approach that problem that they find themselves with.
Oren Greenberg 0:26
Well, to prioritize how to get on top of AI
Tom Rudnai 0:30
to make sense of all the noise. I guess what you're saying is that this isn't something that necessarily needs to be tackled with large scale change programs, it's something that actually should be put on the plates of your department heads. How was a department head who already had a full-time job, as well as keeping up with this fast, fast evolving trend? How do I go about that?
Oren Greenberg 0:54
One of the ways I went about solving this problem is, you know, I've created a program called the AI Marketing Lab to help marketers with that specific problem, so my approach is training and upskilling marketers to learn everything from the foundations of prompt engineering to automation to agenda KI, so they feel like, oh, I'm confident and comfortable with this. So I think that's a solution. It doesn't have to be my training program. Go find a training program that you want to do, so you're upskilling and training, but you know, you, that's like, if you want a simpler, easier way, if your stuff did active, then there's a huge amount of content on YouTube, TikTok that's widely available. I mean, like, if you look at NA 10, which is growing in popularity, they have some really fantastic videos on YouTube that show you how to do stuff, but, like, it's like it's just a time sync, an energy sink, like, where's on your priority list? I think the problem is, there's so many software now, with the fragmentation of different areas of marketing that they cater to, that you can't do all of it, so you just got to focus, and you got to think, well, what is the problem I'm trying to solve, is that a specific tool or software? How do I go down that rabbit hole, or is it another problem that I don't have, and how custom is that problem? And then you can either, like, you know, reach out to someone who knows that, you know, someone who is positioned as a, you know, either a data scientist or AI consultant, and you can ask, how complex or hard is this problem I'm trying to solve? Actually, for example, the other day I wanted to write a book on prompt engineering for B2B marketers. So now I have a crew of six agents, and I've set it up in cursor, and I'm using Crew AI as a framework, and I'm building that myself to solve to scratch the itch, because there was no off-the-shelf solution that, and I was looking for that could help me do that in a more efficient way. So, you know, sometimes you will end up having to build custom solutions, and if you can't, then you need to outsource that, and that comes at its own, like, you know, can of worms, or you veer and work with a safe, well-defined route. I mean, there's a lot of tools out there. I don't want to name any of them, because I don't want to ban that. There's a lot of very popular AI marketing tools that are just so ludicrously expensive that I can do what they do with a Google Collab Python script, at which it takes me 10 minutes to whip up, and it's a 100th of the cost, and the problem is the reason marketers veer towards these sources because the UX and UI is so beautiful and easy, but that's not really what the issue is. The reason the problem is the cognitive load and the time sink and the juggling of multiple responsibilities. So marketers are time poor and energy poor, and they need quick solutions, and they're happy to pay a premium for something that's easy to use and quick, rather than try and learn how to do it, even if it only took 10 or 15 minutes of fiddling around with a Python script and Google Colab to try and get that result. So, I think I think the problem a lot of marketers have is they're kind of split into two groups. There's been someone the other day, 100 marketers in the organization, 30 marketers are AI native, 70 of them aren't, and apparently all the product marketers are the AI native ones, which is interesting. I never, I wouldn't have thought that product marketing specifically would be the first sub-specialization marketing that would pick up and become AI native, but there you go, and that's another survey representative, just that organization. But what was really interesting is the ones who are more AI native tend to be a bit more technical and more performance marketing oriented, and it's the marketers, and you know, the bulk of marketers who are more brand
Oren Greenberg 4:30
centric or content centric who aren't as technical or aren't as comfortable, and I think for them it's just a lot easier to go and find those tools off the shelf and try and use those rather than try and build something that's more efficient or cheaper, but obviously it requires a lot more mental strain.
Tom Rudnai 4:46
Yeah, and the challenge I think is you often end up creating downstream impacts in terms of the workload that you have. Like one thing that always strikes me about these tools is that, like, product fundamentals within them, in terms of integration and things like that, it's really poor. So the amount of time. I spend copying and pasting and downloading this as a PDF, feeding it into this AI. It's really bad, and that's where I guess you need to be careful that you don't just leverage every single tool and end up just being a middleman for all your AIs, and you've forgotten to. I often find myself thinking I've forgotten how to think, but I think one thing I was thinking as you were talking was also like the way that we set goals as organizations probably isn't that helpful, because all of our goals, for a marketer at least, revolve around output, right? So you're judged on what you put out and the impact that it has, you're not judged by efficiency, so it doesn't really encourage you to go and play around and say, hey, we've got a way more efficient way to do the same, which is still very valuable in terms of what actually creates enterprise value, right? So, I think it's something, if there's one thing to take away as leadership that you can do to help enable that kind of departmental ownership, that's probably one thing, is create goals around
Oren Greenberg 5:57
it. Yeah, I think all executives I work with, they only care about outcomes, and outcomes are always translated commercial KPIs, and marketers, regardless of seniority, have the channels of translation of how activity input translates to a result output that translates into an outcome, and executives really only care about the outcome, and they don't care about the way you, you initiate that process, right? So, the more you, you need to, when you're communicating internally, you need to say, I'm doing this because I'm trying to get this output, which will lead to this outcome, and then people go, "Oh, I can understand what you're trying to do here, but, like, "Hey, I'm using AI to be more efficient, so I can undertake more activity, which will shift their KPI on marketing qualified leads, or, you know, sales on e-commerce, and that will result in increased revenue. Now, the executive can understand why and what you're saying. Marketers tend to talk in jargon, they tend to talk in, like, click-through rate and impression share, and you know, share a voice, and a lot of executives, especially like COOs, CTOs, CFOs, CPOs, they don't understand what this means. Some CEOs do, because either come from sales or come from marketing, and they kind of get it, and by a lot of other CXOs don't get it, and the marketer keeps falling into that trap of talking about time and efficiency and activity, and they're task lists and to-do lists, and you know how busy they are, and executives already care. If you could do it half the time and still get double the result, most executive would say great, do that.
Tom Rudnai 7:34
Yeah, now you're talking our language at Demand Genius, everything comes back to revenue impact. We don't care how much traffic, how much engagement your blog post has generated what was the impact that it had, and that's what you need to be reporting upwards. Can you maybe just give us a little bit of an overview of, like, what makes a great prompt, because I think that's that's the skill that comes back that you said earlier a lot of people are lacking, and probably if you can do one thing today to get better, is that
Oren Greenberg 7:58
yes, the one thing I'd recommend you do today is go to my homepage or in greenberg.com take the AI gap analysis for your competence to see where you are in your competence with AI, and I'm going to be sending an email in the next couple of weeks with a free email educational course on prompt engineering, so rather than like try and skim, I'm going to give you a nice structured email course you can consume, and that will really help you understand that. So, it's like a really pragmatic solution, but just to answer the question immediately, in the very short term, the biggest issue is like multiple, there's lots of issues, but if I went through the biggest problems, they are prompt stuffing, so giving the prompt far too much context that is irrelevant for your request. It's not structuring the prompt in the right way, like who's the audience, background context, what you're trying to do. Give the AI some sort of specific role, get it a very concrete output that you're expecting. Ideally, give it an example of an output. Another really common mistake that I see would be not going through this process of understanding that it's iterative, like the AI is a generalized model, most large language models, and as a result it's giving you the response that's average, so the person thinks that AI is default, but what actually is typically happening, the user hasn't given enough specificity to get the right parameters and narrow enough to get the desired output, so what you need to do is you really need to bear with the AI as it gives you a wrong answer and give it an opportunity to understand what it is that you want, and sometimes I can take three attempts, sometimes I can take 15 attempts, but what you'll always find is it always came back to you not thinking about some edge case or variable or way to view the lens on the problem that the AI has taken into consideration, because it has so many lenses they can view at the same time, so like I see people always approaching with a very specific lens. They believe that's truth, and the AI has 15 lenses, and they're just viewing the answer through the wrong lens. I still look at my analogy or metaphor here. Hopefully, that's come across clearly. There's lots of other problems I can talk about. I'm not sure where. Do I just produced a video that talks about the order of importance? So, if you put the wrong order in your prompt, it can impact the quality by 14% So, if you put the most important piece of information at the very bottom of your list in your prompt request, you're not going to get as good as a result as putting the most important thing on the top of the list, and that's just because of the way they're the transformers work with the order of words and the tokenization of those words and just the way large language models operate to understand meaning within patterns of language, so it's linear and it doesn't read backwards in the same way, so it doesn't take everything and evaluate over equal weighting, it gives more precedence because most of the content on the internet tends to focus on what's most important first, and that's been the training data that's the hypothesis as to the reason this is happening.
Transcribed by https://otter.ai