Artificial intelligence. Everyone is talking about it, across every industry and in almost every role. It's the game-changer we're all figuring out how to harness, including those of us in marketing.
Earlier this year, our very own CEO Scott LoSasso shared his thoughts on the massive impacts (and opportunities) of AI in marketing. Since then, we've been learning, listening and fine-tuning our understanding and skills around the advancements in artificial intelligence and the emergence of so many new tools. While we believe AI can power up the way we work in amazing ways, we also know we are just at the beginning of what's possible.
So, today, our Chief Strategy Officer, David Fabbri, and Director of Digital Technology, Dimo Raychev, are here to share how we're navigating this evolution to deliver stronger work for our clients with human-led AI. From strengthening our internal processes to finding the right tools that can accelerate our workflows to optimizing campaigns efficiently and effectively, we know this will be the first of many LoDown Chats on artificial intelligence and marketing technology.
Here are some of the key takeaways you can expect from the chat:
- Organizations must prioritize having processes and guidelines in place to address the opportunities and risks of implementing and leveraging AI.
- Successful AI implementation depends on its usage and prompting, along with continuous learning and adaptation to evolving technology.
- The approach towards AI should balance efficiency with ethical implications, such as upskilling employees rather than replacing them.
- While third-party AI tools may not always deliver high-quality output, they can still assist in generating ideas and identifying what not to do.
- AI has great potential to shape industries and society, but responsible implementation guided by processes and guidelines is crucial for optimal results.
You can listen to the audio resource or read the transcript below. We've also linked some helpful resources at the end. If you enjoy our chat, be sure to sign up for our monthly newsletter to catch all the latest musings and marketing insights from our team!
Listen to (or watch) the recording
Read the transcript
Quick note: This conversation was transcribed with the help of artificial intelligence and has been lightly edited for content.
David Fabbri: Hey everyone. I'm David Fabbri. I am the Chief Strategy Officer at LoSasso Integrated Marketing. I'm here today with Dimo Raychev, our Director of Digital Technology, to have a conversation about AI. How it's impacting the marketing world and how we're approaching it here at LoSasso.
There are lots of questions from across our staff—and from our clients—really wanting to understand how to approach it effectively. You know, what are the biggest opportunities? What are the watch-outs? Things to be careful of? It's such a new and rapidly changing area, we thought just digging in and having a conversation about our current thoughts and approach would be a great way to help share what we're learning as we work through it. Hi, Dimo. How are you?
Dimo Raychev: Hi, David. I'm great. Living the dream.
David Fabbri: Living the dream! Today, we want to talk today about AI and the impact it's having on marketing. Obviously, AI's been around, but last fall with the release of ChatGPT, the interest and applications of it have just gone ballistic, right?
Dimo Raychev: Mm-hmm.
David Fabbri: It went from something that was being developed, to something that's out in the wild and being added to all kinds of software and programs. We needed to understand as an agency, you know, what the potential implications were.
Dimo Raychev: Yeah, that's an interesting point. You said it's been out for some time. That's key because a lot of people don't realize that artificial intelligence in the form of machine learning is nothing new. It's been around for a long time, and we just didn't have the access or understanding of it—and now it's different. ChatGPT obviously changed everything. I think that actually ChatGPT was like a marketing stunt tool to bring awareness to that type of technology and service. So, companies that are working on those machine learning models can monetize them at some point somehow.
David Fabbri: And I would say it was super effective. I mean, it had its intended result, right?
Dimo Raychev: Yeah. Now everybody knows what it is. People are talking all over the place about the integration of different products, tools and technologies. It's just exploding out there.
David Fabbri: Yeah, as an agency, we felt it was critical for us to understand it because when you look at what it is capable of doing—or seems capable of doing right away—you say, well, how is this going to affect the industry? How is this going to affect the work we do for our clients? And so, I know here we really have taken the approach that says, okay, let's try to develop a culture of curiosity around it, right? An investigation. So that one, we don't have people resisting it. I think there was a risk of that. There's also a risk of people getting afraid of what it could do in terms of, do you have copywriters who think, "What's gonna happen to my job?" So, I think one of the things we really wanted to do was get people thinking and talking about it.
Dimo Raychev: That's right. What the opportunities were and the risks, in general. A lot of conversations are out there. There are some concerns about this technology and how exactly it works. There are also privacy concerns. People are also scared if that's going to take their jobs away. But, you know, I don't think that we need to necessarily see it as a threat. In this case, it's more of an opportunity for what that technology can bring—what this technology can help us do better.
David Fabbri: It is sort of fundamental to how society is really gonna function in a lot of ways. There's a guy named Paul Roetzer, founder and CEO of Marketing AI Institute [NOTE: This is a correction from the original recording conversation], which is a group that's been together for a number of years now and really is at the forefront of tracking what's happening in the industry. And there's a quote I saw from him that said, "We are under the assumption that at least 80% of what knowledge workers—across all industries—do will be AI-assisted to some degree in the next two years.”
So, on one level, that sounds exciting, terrifying… They're building AI into all the major platforms now, right? So, it's in Google Workspace and Microsoft Co-Pilot and Salesforce. All these companies are adopting it and sort of integrating it. Even in the day-to-day work that people are doing with the tools they're already familiar with, it's already touching so many people's lives. I've frankly been surprised. When we did a kickoff meeting to discuss it as part of one of our quarterly team meetings, you know, we talked about how fast things are likely going to move. And I think even thinking in those terms, it's moved so much quicker than I expected.
Dimo Raychev: Absolutely. I think it's going to move forward very fast. We're already seeing that. It's going to be here to stay. I'm not sure if it's going to stay exactly in the format that it is because of some privacy concerns and other things that we'll talk about a little bit later. But I do think the potential of that technology and what it can bring to many different industries—not only for marketing—is just amazing. And I'm personally very excited to see what the next step is.
And then the other thing I wanted to kind of clarify is the whole structure of how these language models or artificial intelligence tools work. For example, ChatGPT is not the actual artificial intelligence model—it is not the language model. There is another company behind it called OpenAI, and the actual product is that. ChatGPT is just the interface of that tool.
David Fabbri: Yeah.
Dimo Raychev: Actually, the abbreviation of ChatGPT is chat generative pre-trained transformer. So where I am going with this, is that we are going to see an explosion of tools—marketing technology tools and other online digital tools—using these artificial intelligence models to bring some kind of functionality. And many of them are language-based. But there are many tools that generate images and videos. We're going to see more and more. So, that's kind of the excitement there. There are six or seven big companies that are working on these things—OpenAI, Meta, Google. They sell that service. They sell that artificial intelligence tool to all these third-party companies that can build thousands of different tools. So, I'm just excited to see what people will come up with as an idea, as a type of automation, as a type of tool to help us do things. I think one of the best company examples is the one that we use here for our writing.
David Fabbri: Jasper?
Dimo Raychev: Jasper, yes.
David Fabbri: Jasper AI. Yeah.
Dimo Raychev: They tapped into OpenAI to use their language model, and what they did with their interface is they just predefined those usage cases for different things that you can do. They created, if you wish, a template that you can use to write a blog post, generate an idea, write certain kinds of social media copy, etc. They just streamline the process of communicating with those language models—how you ask them to do things, and what the prompt is. So, they just offer that nice interface, and it's much faster and more efficient.
David Fabbri: We see it [artificial intelligence] getting plugged into a lot of tools. And we're testing things, right? In terms of how companies like ours and client-side companies think about bringing those tools forward and using them. What are the ways people should be thinking about operationalizing? I mean, we talked about risks and opportunities. What are the risks right now with some of the AI tools?
Dimo Raychev: I think one of the biggest risks in using some of those tools, especially for research, is the accuracy of some of the data. You know, in the industry, they call them hallucinations because they [AI tools] may come up with something—with the feedback, whatever the prompt is—and it may sound very, very good, and very true, and not be at all. That's just the nature of the tools and how they collect data.
David Fabbri: That’s a pretty big risk.
Dimo Raychev: It's a pretty big risk. You know, the way those tools are trained, they're trained on a large scale of data that is out there on the internet, accessible. But we know with data, there is good data and bad data. So, they are just going to scrap all of the data and create that knowledge based on that data. There are certain checks and balances that these companies put in those tools. So, if you ask the tool, "How I can create a bomb or something like that?", they have those little checks and balances to make sure that information is as accurate as possible. But still, it's a pretty big issue.
David Fabbri: I do think that points to the fact that we've seen it's very important to build your processes. We call it "with a human in the loop" for a few reasons. One, if technology can give you great information but also completely hallucinate, you need somebody who knows enough about the topic—subject matter experts—to be able to say, whoa, whoa, this seems like nonsense. I better check this. Really checking, fact-checking everything as part of the process to avoid those kinds of data errors. But the other thing is that so many people will have access to this technology and start to use it in a way that's quantity over quality, right?
Tell me about this topic. And they sort of get the same generic feedback. You get a sea of sameness out there. So, what is the unique point of view that you are going to add to that? I think that comes in a few different ways. We can talk about prompting in a second, but saying this is the general information about a topic, but because of my industry, my company's perspective or my personal perspective, I have a different take on this that says: this is the general information and this is the spinoff. That's what I think is going to get more and more valuable.
I think that AI absolutely is going to help with some of the sort of bulk-writing content. But it's going to be so crucial then to take those copywriters and subject matter experts and get them focused on making special, unique content because they know the audience or the topic well and have a unique perspective. Because otherwise, you're going to be one of thousands of people or companies sort of spouting the lowest common denominator of general information.
Dimo Raychev: Yeah. Actually, I think you make a very good point here. And earlier we were talking about how certain types of jobs and professions may go away because of those language models—a copywriter, for example. But you gave a perfect example of how the human element, is still really important in that process. If we dissect the process of content creation: we have the initial research, we have the fact-checking, we have the writing, we have the editing and we have the publishing, right? All these different buckets have been performed by humans—copywriters, editors, etc. Now there are certain buckets that artificial intelligence can help with.
It can help with initial research, but then a human needs to do the fact-checking. It can help with the writing, but then a human can create that uniqueness to that piece, that voice with the editing. Then it can help with some of the publishing things. But again, the human is part of the process to make publishing more unique, not automated.
You know, how you create different types of content to go to those different places. If you think of that example, it's not taking away, it is giving—it's helping, it's enhancing. So, if we take certain types of tasks of initial research, which is a very intensive task even with the internet, you still need to go research, check, watch videos, whatever. But if that tool can save time in that place, then the human can put more time in a different place of that process, like more in editing. It's a better-refined piece, more in publishing and on more channels. That's exactly how I see it: an enhancement to our job and what we do.
David Fabbri: I think one of the things we're seeing with our teams and our experiments, too, is the sort of front-end input there in terms of how you're prompting the tool. You hear a lot of people talking about prompt engineering and some of those things. You can get a very different outcome, or you do get very different outcomes, based on what you ask for and how you ask for it. Again, back to that sort of person and their role in triggering the right process, framing it up strategically and in a way that it's going to deliver something that is more valuable and more nuanced than just a very generic prompt. That can make a big impact. One of the things we're trying to figure out is, you know, what's the best way to track our successes and failures with prompting?
A lot of people talk about building prompt libraries. If we know to ask for certain things a certain way, we don't want all of our writers and people to have to figure that out every time they go prompt something. So, are there tools that can really normalize a process? What are some of your thoughts on some of these processes and tracking?
Dimo Raychev: Yeah, super important—prompting and creating that prompt library. As you said, we, as a team, started saving certain types of prompts based on our success and experiences, either for some kind of language output or image output. Usually, the first prompt is just not good enough. And then we need to refine it. We need to ask for more specific things. We need to just ask more. Then we have success. Then we save that in a repository where everybody can access it. So, they don't go through the same process of refining. They just have a ready prompt. That's one way to do it. The other way, as I mentioned earlier, is some of those third-party tools are actually created with an already-defined prompt process and templates. You can go and create specific things because the engineers behind it created them for you.
David Fabbri: One of the other things that we've found important—as we've been going to some of these conferences and are keeping up with how people are handling these things and trying to get a better handle—is to make sure that there is an organized focus around it. In our case, we have this AI committee, basically, which has leadership from all the different departments so that we're coordinating our efforts in terms of potential use cases. I think for companies that are saying, okay, how do I get an effective way to approach this, that it's pretty key.
Dimo Raychev: You're exactly right. That committee is really important. The conversation happening between team members is really important. The process that we go through to research the different uses of those tools is important. And I think involving everybody from your organization—in our case, members from Account, Creative, Media and Web—the idea is that we have a representation of each possible use. You know, when I look into those tools, I often look into how I can optimize certain types of data cleansing or data research and input. I look into how we can optimize certain media spending and buying processes. My view of those tools and use cases is very specific. And that's where I concentrate. But then that could be completely different for a copywriter from the Creative team. They need to look into those tools in terms of content creation.
That's why it's important to involve everybody to get the best out of those tools. And the other thing is, when we go through the process of looking at the use cases, we also try to optimize that in terms of how much time it's going to save us as a team, as an agency. How repetitive is the task that I'm looking to automate? Is this a tool that I use every day, every week? Who is responsible for it? It's all these other things that we put into consideration when trying to identify how an artificial intelligent tool, um, language model will be used in our workflow.
David Fabbri: And of course, doing all this at the same time. We're doing all the work we had to do already, which is another reason something like a committee can be so important. It's easy to think about it and then forget about it for a while. And I do think, even if you're doing a good job with that kind of a committee and building that process, there's a real risk of thinking: "We figured it out now, so we can stop thinking about it." The pace of change is going to continue to be so rapid that it's just going to have to always be: "What's next? What are the other tools? This is working great for now, but what else?" One of the reasons we wanted to really get focused on it was that it's going to give agencies—and companies across the board—who leverage it most effectively huge advantages.
You know, this is just a little example, but we had a new client pitch a couple of weeks ago. We had some concept ideas that we were able to execute visual representations of with Midjourney, which is an AI image-generation tool. Now, we're not using any image generation tools in our final client work right now because there's still a lack of clarity around copyright. And with images versus text, there's a lot more of what appears to be plagiaristic activities happening. But that said, you know, we were able to go in and do some really cool dynamic executions on this concept that really had some impact in the meeting. Then, we can then go set up photo shoots and style and, you know, do new work to generate something of that sort of concept and style. But it just had a completely different sort of impact than I think it would have if we just described the concept and showed it—or showed it with generic placeholder art. So, stuff like that, just sort of like idea generation and concepting, is kind of huge for the creative side of our business, for sure.
Dimo Raychev: Oh, yeah, absolutely. I think that that was a great example. And when I saw those images for the first time, they looked completely real and good to me. The quality was that good. But you are right, there are concerns of how those language models were trained to create those texts or to execute those visuals. There are companies out there like iStock or Adobe that are training similar models based on their proprietary data. Technically, if that model is not using copyright-protected data from the internet, you can teach such a model to create images based on proprietary data, so…
David Fabbri: So, they are owned images; basically, they have rights to them...
Dimo Raychev: Yeah, now they're good. I know that there is still a gray area with the whole privacy and copyright protection issue with data. But there are companies that start training those types of models with the right type of data. They're behind Midjourney because Midjourney had the library to train, but the problem there is copyright protection.
David Fabbri: And they're looking for that stuff. This is just a small personal example, but in addition to working at the agency, I'm in a band. And we do promotional materials for gigs, and I used Midjourney to make some really cool illustrations to use for a promo for a local concert. Those were posted on Facebook, and I got a notice from Facebook that there were copyright infringements. I think the risks are really too high still, for sure, in the image area.
Dimo Raychev: Currently, there is no law on the federal level that is saying whether those are copyright-free or not. There was research from the US Congress around that. They're not recommending any actions at this point. They're just waiting to see how things will be out there to make a decision of how they want to act. But, I know, that sooner or later those things will be addressed. The way we train those models, the way the work that they create is copyrighted, who is the owner of that work....I'm expecting that in one—maximum two years—all this will be put in some kind of legislation for us as guidance.
David Fabbri: I think for now, for us as an agency and other companies who are utilizing the tools, it's how do you get advantages from them while you're protecting yourself? So, making sure that the people who work for you understand that these are the guidelines. This is how we're gonna use it. This is how we're not gonna use it. Keep it all very transparent and out in the open. Because the risk is in not having the conversations is then you've got people who are leveraging tools that make their jobs easier, but potentially exposing proprietary data to platforms or doing things that end up causing copyright infringement. Getting those processes and sort of guidelines in place so that you can get the advantages and avoid the problems, I think is pretty critical.
Dimo Raychev: Any organization at that point should be looking at those processes and internal guidelines. Any organization cannot just ignore it, cannot just block it from their Internet access and their computers. They need to address it somehow and to give guidance to everybody that they're working with to explain it a little bit better, how it can be used...You know, those conversations should be happening everywhere.
David Fabbri: Yeah. And I think there's a broader humanity ethics piece of this as well, which is how are companies going to approach this as they learn how to pick up efficiencies with it? Are they going to say, "Great, I'm gonna get rid of all these people and use AI to do this task to save on labor expenses." Or, are they going to say, "Okay, this is great, now let me figure out the 'yes, and' of this, which is refocusing energies in these areas or uptraining them to do different skills as these things evolve." I think that copywriter example is interesting, you know, in somebody who was more of a straight writer. Do they become more of an editor type? Really focusing on an area of expertise, so that they can help drive high-level content, get it focused in the right direction, validate it, and pull out the nonsense that AI is generating. And get more out of a team versus trying to cut back a team, and do more with less. Seems like a negative way to approach it.
Dimo Raychev: Yeah, and that brings me to another conversation. We've been testing some third-party tools out there. We've been opening accounts, testing them and seeing how good they are. Like there's a tool out there that claims you can create everything from a landing page to all the ads to the actual creative inside the ads. So, pretty much like a full-blown marketing campaign. You just give it the name, the prompts, colors and logo, and then they're gonna create everything. And we tested that and it did that—but it didn't. It was very, very generic. And if a company thinks that they can go and just use that tool to create something, it just doesn't work that way. I don't know if those tools will get better, but for sure the output is not ready for prime use.
David Fabbri: I have to say too, at this point, I found in my own experimentation that some of the most effective help that I get from AI is seeing what I don't want to do. And I mean that in all seriousness, in terms of like, I'm investigating writing a blog post about a certain topic. I kind of prompt it and then it spits out a bunch of generic stuff. So, I'll say, okay, there are some general ideas I want to pick up, but I definitely don't want to go in this direction. Sometimes it's also seeing what you don't want—or don't want to do—that can be beneficial with the tools, too. It's kind of interesting.
Dimo Raychev: And again, there are so many use cases. It's really up to who is using it, how the tool is used and having the right type of prompts. Just really think about how you're using it.
David Fabbri: Yeah. I guess to wrap up, these are really high-impact tools that are going to affect our industry—and the world in general. You know, making sure that your organization is looking at it with open eyes and embracing it in ways that say how can we use these things? How can we be efficient in our learning, avoid risk and reap the benefits? That's the optimal approach for the time being. But any other last thoughts?
Dimo Raychev: When we work with any kind of technology, things are always evolving, so I see the whole process of understanding artificial intelligence and implementation as ongoing. We just need to get there. We need to have the right mindset. Keep moving, keep learning and start figuring things out. I do think that technology is great, and in a way, it works like magic, especially for some people who've never seen it. I just see great potential for it. We just have to be there, test it, talk about it, implement it, create processes around it and create guidelines, so we make sure that we can use it in the best possible way.
David Fabbri: Yeah. Well, thank you Dimo, for taking the time to sit and have a chat with me. I think we'll probably keep this going as we learn things. We're more than glad to share our experience and put it out there in the world.
Dimo Raychev: Thank you, David. It was great. You know me, I can sit and talk about these things 24/7.