Disrupt or Be Disrupted: AI in Marketing
Artificial Intelligence is reshaping the marketing landscape with groundbreaking ways to engage customers, enhance user experiences, and streamline marketing operations. Yet, the AI revolution also presents challenges, including data privacy concerns, potential job shifts, and ethical dilemmas.<br /> In this audio from our recent webinar, Joe spoke to Tom Edwards, Director at Archetype to discuss the applications, implications and potential risks of A.I.
Related Podcasts
Transcript
Transcript
Joe: Hello and welcome to the B2B Essential Podcast, brought to you as ever by Lead Forensics. I am your host, Joe Ducarreaux. This episode is the audio taken from our webinar we ran recently called Disrupt or Be Disrupted – AI in Marketing, Are You Ready? For this I was joined by Tom Edwards, Director at Archetype for a conversation around the hot topic of AI in marketing. AI has fascinated me these past few months so I couldn’t wait to talk to Tom all about it.
So without further ado, here is my conversation with Tom Edwards on AI.
Joe: What are some of the key ways that AI can help marketers improve their targeting and segmentation strategies?
Tom: There are so many ways. I think we are only just starting to scratch the surface now really. You will have seen for sure, as I have, all of the different tools that are springing up that are building on the CPT4 API.
For me the big one is just ideas and inspiration. It’s a very creative partner and you can use it as a kind of brainstorming tool in a way that I think that we’ve never had before. So there’s that and the other thing that I think is really powerful is just the volume you are able to achieve now. So with things like AB testing you can just get assets creative copy for a massive range of verticals and segments in a fraction of the time. You are just able to test and learn at a much greater pace with the same amount of people
There are a few different things, the big two for me are ideas and inspiration, getting creative and getting new ideas and then building on those with humans and figuring out the way forward. Then just the volume with testing. We don’t have a huge degree of integration with the platforms yet, that’s not to say that won’t happen and I am sure it will. It’s really all about that volume, getting the different segments built in a fraction of the time.
I guess one other thing to call out is probably chatbots. You can have much more engaged conversations with prospects. I think the degree to which the data you get out of the back end of those chatbots is going to be really powerful and is going to grow a lot over time, is my hunch.
Joe: Something that you touched on in the first part of your answer there, the creative side of things. That’s something I wanted to pick your brains on because if we have things like ChatGPT and Midjourney which can generate images through AI does that pose some risk to the creative teams within marketing or is it more of a collaboration, do you think?
Tom: It’s a good question. I think it does pose a risk to some people. There have been stories about game developers who have basically been replaced by Stable Diffusion or Midjourney. But, it seems to me anyway, it’s much more of a creative partner. Humans are very much still in the loop. Where we will go in the future, who knows? But right now it’s essentially turbo charging the creativity of people who are already creating. That’s my hunch.
It’s the speed that you can iterate on your creative ideas has just been accelerated so much. You can test so many more theories so much more quickly. I use Midjourney a lot, I use it for thumbnails and dropping ideas to visualise what’s in my head. So whereas I might have had a conversation with someone which would take me 15 minutes to describe what I am talking about with a whiteboard and drawing diagrams or whatever. You can iterate over your thoughts so much quicker in Midjourney and throw ideas out there and that ties back to what I was saying about testing and learning. You can throw lots of different visual ideas out there as well and test them.
It seems to me so far that it’s like putting a turbocharger on your creative engine if that’s not too strained a metaphor. I think it really speeds things up and gives you far more options. That’s not to say there’s no risk. I think the big risk is in the size of teams. So there’s risk for designers and people who are actually implementing these things; video editors, that kind of stuff. What you might find is that all of those people who were doing repetitive design tasks will now go away and create their own stuff. That’s the dream scenario I think, is that they can finally execute on all the creative ideas that have been bubbling away in their own heads. I’m naturally an optimist and I think it feels like a very exciting time but it’s also moving incredibly fast, so who knows where we will be this time next year. It’s so difficult to call. What do you think?
Joe: If you look back to a year ago, I’m not sure that any of us thought that in a few months time we would be using these sorts of things. I think that on the piece of the ideas creation, the brainstorming, it’s almost like having a creative partnership with someone. Right bounce some ideas for this blog post or this piece of content I need to make or whatever. I do hope it retains some of its humanity but I think you make a good point in the size of teams perhaps will be affected.
Tom: That’s the risk, I think. As an individual, as a content creator, someone who produces ideas for clients, for marketing you can just produce so much more. I used to use random word generators, I would get 10 random words and see which ones sparked a creative idea and then just keep doing that until I get five that are half decent. Then I’d take them to some other humans and iterate over them.
Now you can ask the ten pretty fully formed creative ideas and then go away and get test imagery for each of them out of Midjourney and then take all of that and start iterating over that with other humans. It’s just like a step change in speed and volume that we’re only just now getting to grips with the possibilities.
I’m sure you will feel the same way as me that there’s just news coming out everyday that’s just changing things again and again. It’s a very exciting time, I think.
Joe: Absolutely, it’s an incredible time to be in these industries being able to see how people are using it. To that end then, one of the key points you mentioned earlier Tom, was the volume of things. If I’ve learned anything during my time doing these webinars and podcasts for Lead Forensics, it’s that people buy from people and approaching customers with a personalised approach is essential.
How can you use AI, through the prism of it can do huge volume, how can you retain that personalisation with AI for a customer experience?
Tom: I’d almost come at that from the other way which is we are now seeing such a huge volume of AI generated content coming out. My own inbox is full of what are clearly AI written emails. You can still tell the human written ones and I think that we will start to have an increasingly more of a premium on that human element. Because even if the humans aren’t writing every single word of it, if not now then soon those days will be gone of just writing a whole outreach note yourself, but the human editor is still going to be crucial.
The reason for that is that GPT will do whatever you will ask it to but there’s so much nuance that you are not even aware that you know when you are writing copy, that getting it all down into a prompt perfectly takes the same amount of thinking time as it would do to write it yourself. You get to a point where there is so much context and nuance and specific elements you want to include, that if you are writing it all into a prompt you might as well just write it yourself. Certainly I think the way this will develop people will get a generic piece of copy with a very lazily written prompt and then the human edit will be the real crucial factor, where the real value add is as a marketer and where you will notice the difference on the other end.
As a customer we are now being and will continue to be bombarded with generic salesy AI generated content which has been created by a fairly lazily written prompt and you can tell. There’s not any real difference between that lazily sales copywriting done by a human. For instance the amount of time that you spend figuring it out and working with your other stakeholders in the business or your clients or speaking to your customers, all of those different things, you are still processing those yourself, as a human. You are still making decisions as a human.
Again it comes back to that same point, you are essentially just turbocharging your process. I think that we can, will and should continue to have the humans in the loop but perhaps more as an editor than a copywriter going forward would be my hunch.
Joe: That’s the key, isn’t it? Is ensuring there’s still room for collaboration. I heard someone say about AI, what it does brilliantly is in the same way as a parrot can say I want a cracker, it doesn’t know the meaning of those words are I would like you to provide me with something to eat so I can get energy and feel nice and warm and fluffy inside. It does mimicry rather than actually generating something like that.
Tom: It does. There’s a big conversation about to what degree large language models are actually doing reasoning versus just predicting what the next word is. When you use it, it does become freaky sometimes when you start having a conversation with it as if it’s like a freelance copywriter and saying can you do that again but make it a bit more formal or whatever. Inevitably you end up rewriting the bits yourself. The bits that you are struggling to explain to the prompt, you end up doing yourself, I think.
That analogy that you’ve drawn actually highlights another point which is that the level of context in the things that we say, that we implicitly understand as humans, you can’t necessarily convey in a prompt because you don’t know that you know it. There’s a lot of knowledge when you are trying to iterate over a message, or a piece of copy, or an image or whatever it may be. You are aware of context and nuance when you are writing that you almost don’t know that you know. So in that give a cracker or whatever the parrot says, there’s a lot of context. The parrot knows it’s hungry, it knows it likes crackers and it knows if it says this thing you will give it a particular kind of cracker, whatever. We are probably straining that analogy a little bit at this stage but you understand what I mean. There’s a lot there.
Coming back to the point you were making about human to human, that human touch you have that further down the funnel with video content or once you get salespeople involved there are actual meetings happening. There is so much context that goes into a human to human interaction that you can’t get through ChatGPT spitting out copy that it thinks is going to be highly engaging.
It’s very much augmenting what we are doing as people and who knows where we will get to. It’s already shocked everybody how good it is now.
There’s nothing to say that in 5, 10 years time it won’t be indistinguishable in pretty much every way. Then we will have a different conversation to have which is why are we here, what are we doing, is there any point in business, shall we just let the AIs run the world, they might make a better job of it than we will. But we are not there yet.
I definitely think that there’s plenty of reasons for humans to be in the loop in marketing as with all the other areas of business that this is disrupting.
Joe: I’m pleased that so far, Tom, we are not in Blade Runner territory just yet. Perhaps in 5, 10 years time if we do this conversation again, it’s not even us, it’ll be AI versions of us doing this.
Tom: Have you seen CarynAI, the story? This is going a bit over the top. A Snapchat influencer called Caryn Majorie who has created an AI version of herself and she’s charging a dollar a minute to talk to her. They’ve programmed in all of the … .well I haven’t actually used it but they say that they have replicated all of her quirks and she made $100,000 in a week.
Joe: That’s uncanny. It’s very zeitgeisty I suppose.
Tom: It’s also scary and I think in that context it’s worrying. I think in that sense to a degree it could be preying on loneliness. You can see how that would be adapted and used in a business context in a really potentially quite exciting way. You can see how having your thought leaders, your salespeople, your CEO getting AI versions of them and creating customised messages for all of your different verticals or targeting segments. So your most engaging person is talking directly to all of your different target audiences with a whole bunch of different messages that you are AB testing at the same time using this avatar of your CEO or top salesperson.
There are lots of applications that we haven’t even got to yet. Hopefully it will still be you and me chatting, Joe. Maybe we’ll do a conversation between us two and another one between our AI avatars to see which one gets more listens.
Joe: We started to get into the realms of the ethical implications of using AI, in the context of the social media influencer and would it be ethical if it was preying on loneliness, that sort of thing. What are some of the ethical considerations that marketers have to keep in mind when implementing AI?
Tom: I think the big one is IP, in the broadest possible sense. Firstly from the fact that these large language models and the generative AI for images, Midjourney and Stable Diffusion, they are all trained on information that a human has created. We don’t live in a world of AI created information yet. So they have all been trained on something and there are already court cases.
I think Getty images have started a court case against Stable Diffusion and there are a few others where they are saying the images you are using to train these models are from us. They are our intellectual property, you owe us. You might get to a position in the future where, I don’t know how likely this is, but any business that’s been using content created by Stable Diffusion or Midjourney or whatever it may be, is then subject to having to pay Getty Images licensing for the training data that has been used to make that model.
I think there is a big risk there. I think also as a marketer, IP in the sense of your own IP, as an agency or a business. As we were saying that human touch, having a human in the loop, having an editor is vitally important and if you get to a stage where it is all just AI generated content it then becomes a question over what value you are adding, what the actual morals of the business are and all of this sort of stuff. So there are a lot of risks and because it’s moving so fast these court cases haven’t played out yet. We don’t know what the results of them are going to be. But you see it in some of the videos that have been created. For instance, there was Will Smith eating spaghetti or noodles, have you seen that?
Joe: I don’t think I’ve seen that one.
Tom: There’s a bunch of them but they got the Shutterstock watermark on them because all of the images they have been trained on have got Shutterstock watermarks because it’s public. They haven’t had access to Shutterstocks proprietary dataset, they just skimmed it off the web and all of the versions on the web are watermarked.
That shows you to the degree to which proprietary data has been used to train these things. I think there is risk. I use AI, and I recommend this to anyone I work with, as inspiration but it’s not the final product. They should always be a human editor and then if you are creating or selling design or content you still need to create new stuff. If it’s just regurgitated stuff that’s already existing we are going to live in this recursive world where we are just rehashing old ideas.
So there is a risk taking the human out of the loop entirely. It’s another reason why humans still are very much in the loop otherwise it could potentially run away from you and you could end up getting in trouble.
I think that idea of IP and proprietary data is going to become increasingly important. It’s also going to become important on the training side. Using big datasets, Twitter has ramped up the amount they charge for API access. Google has obviously got access to a huge raft of data from YouTube as well as Google and that’s all going to be used to train the next large language models so having that pool of data is also going to be really important.
Joe: It’s mind boggling isn’t it. I love talking about AI because it’s just so bonkers……the routes can go down. You say you use AI in quite a lot of the stuff you do, have you examples of successful AI powered marketing campaigns, at all?
Tom: The big one is the Coca Cola ad. I don’t know if you’ve seen it, it’s a guy in an art gallery and the coke bottle goes around all of these artworks and they are all brought to life through generative AI. It’s really amazing, it’s a great ad. I recommend everyone to check it out.
There have also been some funny ones. I don’t know if you saw the beer commercial.
Joe: Yes, I was going to talk to you about this. It’s like a beer at a BBQ, is that the one?
Tom: A beer at a BBQ that then becomes a spilling, twirling inferno in the sky….weird, very, very weird. There is big ticket stuff like that. I think generally it’s difficult to identify particular campaigns because most of the time you don’t know that people are using it.
People are implementing an AI driven segmentation strategy as we were talking about before, putting out variations, testing and learning. You wouldn’t necessarily see that at the customer end. I think most digital marketers will be dipping their toes into the water with this kind of stuff and I think it’s actually being implemented in far more marketing campaigns than anybody realises. At the high end, I think the Coca Cola advert is the best one I’ve seen.
It’s only going to increase, we are going to see more and more of them. Things are going to get pretty wild, I think.
Joe: With the adoption of any new technology there are a few fits and starts aren’t there. I’ve not seen the coke advert but I will check it out after we’ve finished recording. But I cannot recommend enough you go and check out this beer advert because it is pure nightmare fuel. It’s very strange, weirdly you can see the thinking behind a lot of it. There is logic to why they have put so many massive Cheshire Cat grins in it because that’s like every advert campaign, oh we’re having a lovely time at this BBQ. It is genuinely scary.
Tom: Totally agree. The thing that gets me on that beer one is the way that they drink, they way that they are sucking on these amorphous blobs of pseudo beer can, it’s freaky. It’s like the Uncanny Valley, it looks similar enough to real to convince your brain that there is something there but nowhere near real enough to be real. So you just constantly looking at it thinking do I recognise that shape? I don’t know if that’s a can or if it’s a ball or a person. Very weird to watch.
You can really see where it’s going to get to. I remember when these large language models were coming out or the first iterations on testing on them maybe five years ago, maybe more. They would write poems that would be almost good but were just weird and didn’t make sense. You can kind of see the same thing now with the video. It’s almost good but it’s weird and doesn’t make any sense. You mentioned the fire, you can see how it got there. You can see because fires are weird and they flicker in unpredictable ways it’s just recursively thought the fire will go on forever and then it just goes all the way into the sky.
It’s the same thing with fingers. Midjourney had this thing with fingers for a long time that everybody was talking about, it couldn’t understand hands. They all had seven fingers or whatever and you can see why. It’s because every finger is usually seen in relation to other fingers so that’s how it’s understanding what’s going on. Clearly over time programming has made workarounds or solutions to that. You are definitely going to get that in video as well.
Joe: To move away from the creative side of things, to go back to data just very briefly Tom. How can marketers ensure that the data they are feeding into AI systems is high quality and unbiased?
Tom: I think that most marketers aren’t training their own AI. There will be some but for the most part as far as I can tell anyway, they are either using one of the big large language models, Google’s Bard or ChatGPT and then plug in to something built on top of that. I don’t think the data you are feeding them on is a huge issue at the moment.
Although obviously, bias in general is a very difficult issue and very subjective and I suspect the way to eliminate it is the same way that we have always talked about limiting bias, is to clean the data. Good data in, good data out. For the most part, the way we are talking about bias in AI at the moment is ChatGPT, and that has already been trained on a whole load of data that has an end point. Although, now it can browse the web so that’s starting to change.
Most marketers, myself included, haven’t seen the data that’s gone into training it but I think in general removing any bias from any data that you are putting in is in the same way that you would have done it before AI.
Just clean data essentially, in my view.
Joe: To start to bring this conversation to a slight close. I know we said previously we couldn’t have seen a year ago where we were a year ahead but I am going to ask you this question anyway, Tom. I just want to pick your brains on it. How do you see AI transforming marketing in the next 5-10 years, maybe even just a year ahead? What do you think is coming from AI? How can we stay ahead of the curve on that, do you think?
Tom: It’s a massive question, isn’t it. One of the big ones in my view is SEO. You are already starting to see huge amounts of SEO copy written by AI. I was talking to a friend who is the head of marketing at a tech startup. He was saying he has been using, I can’t remember what platform it was, I think it was Jasper to write general SEO copy that would have taken a whole team of people weeks. He’s basically done it all in a week on his own.
All of the content on the internet is going to be increasingly written by AIs. The thing that humans will still do is make the decisions, set the direction and probably still come up with the ideas or certainly make the decisions on what ideas make the cut.
That’s where the executive power will remain with humans and what you will see is just a lot of noise. It relates back to what we were talking about earlier, maintaining that human connection. I think that’s the best way. I also think that if you can become, within your organisation, the translator between the humans and the AIs you’ll be in a powerful position. Because although yes it’s very easy to talk to ChatCPT there are actually thousands of tools out there that people are building on the GPT4 API.
So it’s not just about talking to the ChatGPT prompt bar anymore, it’s about which different tool that has been built on this should I be using for my particular function. Which one is best, which one has the different functionalities that I need.
There are lots of people, clearly you and I are really interested in this stuff and we are probably keeping up to date with all of the different plugins that are being released, different products that are being built, all of that stuff, but a lot of people aren’t. I think most people in business know that AI is coming and it’s going to be very disruptive but they are not necessarily following every single piece of news that comes out about it.
If you have a genuine interest in this stuff I think keeping on top of it, getting to grips with prompt engineering…. I know they say that Prompt Engineering is a job title now. I think anyone can be a prompt engineer, it’s not like you have to learn code. You just have to get very good at being very specific about what it is that you want and understand that human nuance.
Start understanding the things that in your head you don’t even know that are there. If you can do that effectively, I think you will be a very useful person in your organisation not just because you are going to be hugely productive turning out loads and loads of work, but also you are going to be able to train, help and support other people who aren’t so in the weeds on it.
One other thing that I would add to that you can use AI to teach how to do loads of other stuff. I’ve been getting back into the Adobe Suites, working on Premiere Pro and Photoshop a lot more just recently, for fun more than anything, just as a side hustle. But it’s so much easier to figure out how to do the thing that you don’t know how to do now than it ever was before.
I remember when I first learned Photoshop, spending half of my time scouring Google to find the right blog or the right video to teach me how to do this specific little thing that I wanted to do. Then I would have to watch three and then piece together the precise function that I want because no-one has addressed the exact specific issue that I need. Now I can ask ChatGPT and it will tell me exactly what it is I need to do and if it gets it wrong or if I am on a different version……. Often the version is a problem. So the video on YouTube might be on a previous version of Premiere and the newer version that you are on doesn’t work anymore. You can just say to ChatGPT No that’s wrong, try again or I’m on this version of Premiere, try again or I can’t see that button, what am I doing wrong? And it will tell you.
If you are so inclined, and I suspect most people have better things to do with their free time than me sitting around on a Saturday morning talking to an AI but here we are. If you are so inclined in your spare time, you can really learn a lot of new stuff. You can become a video editor, you can become a designer and you’ve got a co-pilot there that will help you with all of those functions that you didn’t know how to use.
I don’t know if that’s an especially great answer to your question but the answer is the possibilities are massive and I would recommend any marketer who isn’t getting really quickly up to speed with large language models and image generation is going to find themselves behind the curve. Whereas if you get ahead of it you can really put yourself in a powerful position.
Joe: Yeah, ever so slightly future proof yourself. If you can become the in-office expert.
Tom: I think so. That will remain the case until we all get replaced. The one thing I fall back on with this is it comes back to what I was saying about proprietary data and IP and the fact that it’s not in Google’s interest to let ChatGPT talk to Google Ads, let’s say. It’s also not in Linkedin’s interest, owned by Microsoft who also own Open AI, it’s not in Linkedin’s interest to let Bard talk to Linkedin and deliver ads on Linkedin.
Due to capitalism and competition you won’t get a big convergence into one prompt that you can talk to, to do everything. You will more likely get through that competition with a whole host of different tools and different businesses trying to win and in the middle of all of that chaos there is still space for consultancy of a person who is up to speed with all of this stuff. Because people won’t know what AI to go to, to ask. They are going to become increasingly reliant on their own datasets and they are going to become increasingly more reluctant to share functionality between each other. Having a human being who can advise on what’s best suited, I think is going to be still a very relevant job for quite a long time. There is a huge amount of opportunity for any marketer, I think.
Joe: We are remaining optimistic.
Tom: Very much so.
Joe: Tom, you have offered up a lot of insight into AI and I’ve really, really enjoyed this conversation and I really related to your comments on Premiere Pro. If there was one key take away that people, when thinking about using AI in marketing, could take away from our conversation today, what is that golden piece of advice you would like to offer them?
Tom: Just get stuck in. You have to sprint I think because it’s happening so fast that if you’re not sprinting you will get left behind within a few years. My advice, this is what I am trying to do, is running to keep up all the time. So just dive in, pay for ChatGPTand use it everyday, fit the plugins, find out how other people are using it and experiment. It’s not going anywhere anytime soon and the opportunity is massive. So that would be my final thought.
Joe: Fantastic stuff. Tom, thank you so much for joining me for this conversation. Did you want to just let everyone know where they can keep up to date with you?
Tom: Sure, yeah. You can find me on Youtube: https://www.youtube.com/@tomedwardstechnews I’m on Twitter tomedwards and you can find me on Linkedin and I am sure it will be linked on this post, its linkedin.com/in/tomedwards100
Joe: Tom Edwards, thank you so much for joining me on this webinar all about AI. Thank you to everyone watching. Do keep an eye out on Lead Forensics’ social channels for news of our next webinar and we will see you again very, very soon.
Well there you go, Tom Edwards on applying AI in marketing. Here are our key takeaways:-
- Embrace large language models and embrace image generation. Stay ahead of the curve by learning about large language models and image generation technologies to leverage their power in marketing strategies.
- Ensure high quality and unbiased data for AI systems. Clean and pre-process data following standard practices to eliminate bias and ensure accuracy and fairness in AI systems used in marketing.
- Become a translator between humans and AI tools. Understand the functionalities and act as a bridge between the technology and others in your team, helping them navigate the AI landscape effectively.
- Dive in, experiment and sprint with AI development. Keep up with the fast pace of AI development by actively exploring and experimenting with new AI tools and techniques in marketing campaigns.
- Collaborate with AI tools as creative partners. Use AI tools such as GPT4 AI for ideas, inspiration and fast iteration of creative concepts. Treat them as creative partners rather than replacements for human creativity.
Thank you to Tom for joining me and thank you for listening. Remember to subscribe to the Essential B2B podcast and rate us 5 stars where possible. We’ll be back next with another excellent edition of Essential B2B.
