David Berkowitz on AI in marketing, gaining superpowers, amplifying marketers, and the future of agencies (AC Ep12)

“There’s no substitute for trying everything. The great thing about AI is how immediately usable it is. As soon as you hear something mentioned in the news, a friend’s LinkedIn post, or elsewhere, just go and try it.”

– David Berkowitz

What you will learn

• Leveraging AI as a marketing superpower (04:18)
• Streamlining speaking proposals with AI (07:21)
• Utilizing AI to streamline repetitive tasks and iterations (10:45)
• Discussing the evolving landscape of AI at the organizational level (13:46)
• Balancing AI and human expertise (15:20)
• Exploring the changing landscape of entry-level work in marketing (21:12)
• Highlighting major agencies’ ability to adapt and restructure in response to industry changes and internal challenges (25:08)
• Emphasizing the immediate usability and versatility of AI in various applications (28:57)
• Establishing confidence in the value of AI-augmented work for clients and stakeholders (34:12)
• Recognizing AI tool limitations is crucial for addressing and improving them (35:59)

Episode Resources

ChatGPT

Claude

Marcel

Midjourney

Google

Meta

Microsoft

Adobe

Transcript

Ross Dawson: David, it’s awesome to have you on the show. 

David Berkowitz: Great to see you, Ross. 

Ross: David, you’ve worked for a very long time in marketing. All the time I’ve known you guys, you have been at the edge of new emerging developments in making marketing and marketers better and stronger. Today, what’s most exciting about how we can amplify the capabilities of marketers? 

David: Well, I’ve never seen anything like this wave of AI. A big part of it too is I’ve always gravitated toward experiential learning and just that act of learning by doing. In part, because I fell into the marketing world, I didn’t have the experience, I didn’t have some MBA, let alone from a fancy school or anything, there was just so much I lacked, but I was very eager to try anything I could. Now in this wave of generative AI and other recent developments, there has never been so much that one could actually try.

Ross: You’ve been trying and sharing some of your and your colleagues’ learnings around that. This idea of amplifying cognition, the marketers are amazing thinkers, and very creative, that’s a lot of what their work is, so what are some of the ways that AI is amplifying marketers? 

David: For marketers, AI can be this added superpower for basically anything and everything you’re doing. This is where if marketing is 10% of your job, then it can give you ways to do things that you just couldn’t possibly put together before. I’m working on some very scrappy products and projects right now. If you spend some time getting an idea on paper, then once you have that core of the idea and that core of the message that expresses what you want to express, then there are so many ways that you can take that right now in terms of creative content, building a website, generating imagery that’s appropriate for it, generating presentations around it, and things that you would have needed a whole team to bring to life or tons of resources. Now it’s possible for someone to do so much of that, at least, MVP version, a minimum viable product.

You might not create the best deck ever, and you might not get the best analysis of all this as you would from a well-trained person who has been doing this for a long time. But there’s so much that you can actually do, and some of that is behind the scenes, like coming up with a plan, coming up with personas, trying to understand a bit more about some data you’re sitting on, and making some sense of it. Then some of it can be public-facing, consumer-facing materials, especially once you get the hang of the kinds of flows you’re working with and can spend some time on that editorial phase of it, and editorial meaning for whatever work you do. It could be data-related, it could be image and video-related, and obviously text-related. But as long as you have that sense of the questions you need to be asking for it, like the Toyota Five Whys, and how do you just keep making it work better for what you want it to do?

Ross: I may be misrepresenting but a bit of a character of what you’re saying is a person has an idea or objective, they have a starting point, they use a lot of technologies to amplify the content generation in the various guises, but then humans cleaning up or tidying up or refining so that it is truly fit for purpose. Is that roughly right or not?

David: No, it’s in the ballpark and so much of it is once you get that first input right. There was an event that came up and I was thinking of submitting a speaking proposal for it. All these fields to fill out, I’m not sure if I want to go there. Let’s try using AI to help me come up with it. I started riffing on a couple of themes.

Give me ten ideas ChatGPT, Claude, could be barred. Of these ten titles, which do you think would work best for an audience of X, Y, and Z who are going to be showing up to this event? Then once you got to that thing, you’re like, Okay, I could speak to that one. Can you give me a synopsis of it and the three main takeaways from this talk? Now, if you know that you can deliver on those three takeaways, then it’s like, write the full-on event description. Now, pretty much the only people who have read the full event descriptions are likely the event choosers. I’m not even sure if speakers read the event descriptions that they’re submitting. It’s often done by a PR firm or someone else.

It’s this odd habit of mine, where I realize before every talk, I literally did this today for a virtual one, I had to go back and read the description of the talk, because I’m like, Oh, wait, are there some things there that I wasn’t prepared to speak about that are actually in the event description? Because this is probably what people who show up want to know about. You have all of this and all of a sudden, something really daunting, is actually fun. I like writing. I don’t like writing proposals. I don’t like writing pitches, especially for myself. But now I’m like, okay it’s just a little bit more of a process. I’d rather just skip ahead to, if I am given this talk, actually doing the work on that, I’ll spend time there. I can now get all these ideas and come up with a better theme, then maybe that’s just something derived from the past ten talks I gave, that I can actually make more exciting. It’s something you’re looking forward to creating.

Ross: Fabulous. It’s nice to have very specific examples of how you’re using that. Thinking about marketers and agencies, marketers are either usually on clients or the agency side, but just looking at agencies, there’s a whole array of roles. You can think about the individual frame or the organizational frame. How do we amplify the individual? How do we amplify the organization? For the individual, is it a case of simply giving the tools and giving some training or telling them how to use that so they can amplify that or how much of that happens around giving individuals tools, or building this into how the overall organization shifts? How does it work to amplify what it can do with the power of AI?

David: Some of it involves giving individuals the tools to use them within reason. No matter where you are, there needs to be some protections like, say, avoiding inputting customer data in there, avoiding personally identifiable employee information, all these things that should be flagged, even by a more open-minded organization, that’s all for making things more efficient with the AI. But having some room where people are encouraged to go and use these tools, test them, just try to ask better questions of them, see if there are better insights that could be derived from them that you couldn’t have gotten otherwise, or you’re hitting on way faster. Having some recommended starting points, because even if you’re looking at content generation, or if you’re looking at some kind of data analysis, there are so many options right now.

A lot of them are built on top of one of the major language learning models, ChatGPT being the most common now. But there are a number of these very well-trained models that so many other things sit on top of so you can sometimes just skip the middleman in these cases and go right to them. Some of it’s just trying to see where this can fit in. You probably don’t want to get rid of your whole creative team just because you can use tools like Midjourney or Adobe Firefly to create all these images and manipulate them in ways that were so laborious to do so. But you might be able to, say, slow down in some of the hirings for more of the busy work and the iterating that happens because there are plenty of ways to go and test a lot of different versions of the same creative, that you want to resonate with different segments.

It’s like where can technology give you some of those superpowers and just also make the people using them feel like they’re just getting so much more done and making more of an impact. Where I think we’re also going to be going at the organizational level, that we’re just starting to see, talked about more commonly  AI, and it’s going to be a huge driving force on the b2b front of AI, are these private language learning models and private instances that aren’t trained on the entirety of human cognition, but are trained on the smaller set of, say company data. Or if you’re in the healthcare space, then every state’s healthcare policy could be an input into this. Then you could start understanding how does this work with this kind of insurance claim and all these things you need, but you’re not looking at the entire web. That’s also going to reduce if done well, one of the biggest challenges of AI, and that’s hallucinations, or just making stuff up.

Ross: Yes. This is one of the reasons why you always need humans in the loop. I’ve seen a number of small micro agencies essentially, take this up wholesale and be not quite AI first because they are human first, but very much AI amplified, are there any examples you’ve seen of agencies that stand out, that are really on the forefront of bringing AI into how they work?

David: It’s hard to say. Publicis, for instance, made a big push a number of years ago to roll out this voice-powered AI assistant, Marcel. The agencies have been building automation power tools for quite a while. There were some colleagues I was working with around seven, eight years ago, we were building tools to rapidly create different kinds of generations of images, and copy, and then run those against very small segments, and try to test which are the best-performing segments for which copy of which creative, and then optimize all the ads around the handful of winners. This has been done for a while. Now, it’s just the thing that first of all, not every agency needs to invest in and do themselves. It becomes less of a competitive advantage once everyone starts using things like this so there’s more of that expectation. But I think it’s just going to be a lot of that opportunity to see when that’s going to be helpful and when it won’t be.

Every agency probably has some like how we’re AI-powered now in our deck; and if it’s not explicit, then it’s going to be implicit, and if it’s not in there, then you’re already seeing some of that backlash of like, here’s how we’re not using AI, here’s how the kinds of humans that you get to work with front and center and the kinds of experts that you have on that team, and all that real insight that goes into it. There’s already some healthy backlash, it’s not just this knee-jerk Luddite-ism. But, , as a consultant my work is AI-powered. I think my clients would be disappointed if I wasn’t using that in some way. But I think they’d also be able to tell if I was just spitting things out that they could enter themselves in an AI tool, and then they could just cut me out if it worked like that.

Ross: You said AI is not a competitive advantage. I’d argue that how you use the AI as a competitive advantage. Yes, everyone’s got the tools.

David: Yes, and it’s so hard to tell until you see just the kinds of results and the thought process that we’ve worked with, and you’re probably not going to know that that well, even over the course of a pitch process. Some of it is that X factor, it’s like you know when you see it, but also, it’s only so often that you realize you’re working with a great agency that gets you and that there’s someone that you need to have this relationship with and that isn’t going to be RFP it out every year.

But this does add to some of that opportunity for just the kinds of insights, the kinds of creativity, the ways that agencies can cut data and surface things in surprising ways. I’d argue that it just gives agencies more opportunities to showcase their value, showcase they’re smart, precisely if they’re spending less time putting things in a spreadsheet and going through lengthy review processes, and more time actually doing stuff, putting things in the market. I tell clients, just the thing we need to do right now, it’s just like, let’s get this out there because whether we spend three months on it or a week, we’re going to know if it works if we actually do something. Now you just have more chances to actually do something and see what resonates.

Ross: Yes, absolutely. Looking forward a little bit over the next year, I think you’ve always been interested in the future of agencies for a very long time. Now I’ve got a factor, which significantly changes the landscape or could. What might you see in five years from now and the agency landscape? What might have changed? What might be the leading agencies? And how might they be different?

David: I think that there’s a big challenge for a lot of the entry-level work that’s going on. I think that a lot of firms are going to be less focused on this very headcount-driven model. That’s just so much of how accounts are structured right now. Sometimes, it’s like a big part of the pitch process. One shop is like, we’re going to put 100 people on your team, and other ones are like, we’re going to put 125, but we’re still going to charge you the same rate as the ones I put 100 on. I think that there’s going to be a lot less of that and a lot more of what you’re getting out of it.

I hear this from agencies, I hear this from other kinds of marketing tech companies, that they’re trying to think what don’t they need full-time staff for, what kind of work can be done either in-house or just by some specialty shop that knows how to use either their own tools or some others and that can go in and fill some of these needs. That probably does mean a fair amount of attrition. Will these people be redeployed in different ways? I think there are also a lot of opportunities in common for all kinds of businesses that weren’t doing much marketing, to do way more. 

The math, it could be done with video right now, AI-powered video editors right now. If you use them, a lot are likely your first forays into it are probably going to be disappointing, even if you’re using the best tools. But I also wouldn’t sign your long contracts, even if you think you are using the best one. Because three months from now, there’s going to be a way better one. In six months, this is going to be accelerating so quickly. If there’s anything you see that isn’t good enough right now, you have to assume it will be. Everyone was making fun of the Midjourney six months ago because the fingers came out all gnarly. That was like the thing to say six months ago. AI, Why can’t it do hands? It was the thing everyone was complaining about.

Once that’s fixed, then you have to start nitpicking further and further, and then you see that the people who really know how to use Midjourney are creating these incredibly nuanced crowd scenes, and every last detail looks like, it could have been in some Ansel Adams photo. Especially if you’re looking even one year out, you have to start saying, what if the tech is really good enough for us to do X, Y, and Z? Because give or take a little bit of time, it probably will be.

Ross: Yes, frustrating. Will you suggest that I think that this could lead to more niche players so that we start to get experts in particular domains around using particular tools or particular outcomes? Do you think that’s moving away from the more monolithic structures to specialists or niches? Will that be one of the potential outcomes?

David: It’s really hard to count out the major agency holding companies. There are few fields where the most dominant players have that much staying power, especially over the kinds of threats they faced and from all kinds of shifts the past 40 years in terms of media consumption, to shifts in how creative and media are done either separately, or together to all kinds of pricing models changing, to the rise of consulting firms, to all kinds of technological enhancements, it’s like, time after time, there’s been some reason to say this is going to cause the death of the holding company of the major shops, and then a few get folded around and molded together.

I worked at Publicis. I don’t know how many internal murders I saw and reorders in just three years there. But it was a lot. But the Publicis is still here. Publicis at the time almost had their merger with OmniTom, but that thing still would have been here if they became one thing. There is some of this rich get richer. The much safer bet is that they’ll find ways to deploy these tools at scale. If you look at the rich-get-richer side of big tech, and you see the investments that Google, Meta, Microsoft, Adobe, all these major players are making, then that also plays into holding companies’ hands because they have those relationships to get first access and to scale those across all their teams and accounts and countries and everything else.

But at the same time, I’m a big fan of Newton’s law where every action has an equal and opposite reaction. At the same time, short-form disappearing content became really prevalent. It also had a greater uptick in these long-form medium articles and lengthy podcasts, and let alone even a resurgence of local bookstores. It’s like the opposite things can be true at the same time. There are ways to play in niches. People who weren’t product developers, people who weren’t coders used tech in creative ways that then filled some kind of need that marketers have, and that will meet folks in various verticals, so just being able to take your experience and build something that fills one of those holes. I think we’re going to see a lot of that, and it’s a pretty exciting time to be an entrepreneur.

Ross: Very exciting. Switching gears a bit to what you do. You mentioned at the very beginning that you’ve got to where you are through experiential learning, which I call learning by doing. Is it simply that? Do you just learn things by doing things? Or how do you go about that in a way that you’re effective as possible in learning?

David: The great thing with AI is just how immediately usable it is. When I was leaving my last in-house rollover Mediaocean, I started just buying a handful of domains that I didn’t know what I’d do with them. I bought one that was AIsayshireme.com. I then used ChatGPT which was like the only game in town, and said, based on this very brief bio about me, what are the reasons that someone might hire me as a marketing leader, and I put those on the site, and it’s like really simple, really low-tech, really low-budget but it connected a couple of those dots.

I then created my AI resume. It was just taking my work experience, and having ChatGPT generate what some of those bullets would be for someone without experience, and even showed you how you can actually do that yourself. My friend, Leah Moron, who runs the AI marketers guild with me, wound up creating a version of that, that you could just go in and try right there on that site. But then it’s just constantly trying to find ways to use it. I just recently published a book using AI. I did not write a word of these 150 pages. I put in a short synopsis of a project that I’m working on. I took the copy from the slide on the marketing deck, put that into the site omniscience, and 30 minutes later or less, it sent me the ePub file in PDF and a cover image, and I popped that on Amazon and the Kindle publishing store, and then I had a book. I don’t think it’s the best book, the couple of snippets I’ve read, it’s not the worst. I’ve read a couple of worse ones lately.

Again, it’s just these kinds of possibilities. For me, the scariest stuff I’ve experimented with so far is voice cloning. Because that gets very good, and there’s something like seeing written words that could be yours, seeing an image that could be generated, there’s just something so scary, really, of hearing your own voice that’s not quite you, there’s a little bit of that uncanny valley going on but that could be you. Somebody says are literally with 30 seconds of training, and it’s creating something that say, over the phone or on a slightly tinny Zoom call, no one would probably have any idea it isn’t you. This stuff, it gets eerie quickly. But then it’s also not just experimenting for experimentation’s sake to see what I could do with it but how do I apply this to clients. How do I get it to help me write drafts of content that I’m building? How do I go in and scale some of that output?

The one thing I’ve seen over and over again is that so many of these tools are almost like whatever you’re using if it’s gotten to that point that it’s ready and out there for you, there’s probably so much that it can do. But you do need that really good input. It’s still worth spending almost too much time crafting that initial about statement, the mission statement, and values, whatever those elements are that you need, go into it. Once you have that in place, and it’s something you’re happy with, then so many of these AI agents can run with it and say, now create a PowerPoint, now create the copy for one sheet, now, I’m going to pop this into a DIY website creation tool, now, I’m going to use this to create a script for a video that an AI engine and you could just do this almost infinitely, as long as the inputs are good.

Ross: It’s fantastic. The nub of that is that you’re putting it to practical use immediately. That’s where you’re rather than I think a lot of people just play with all sorts of stuff but they’re not actually trying to make something useful out of it. That’s when you really start to learn how to put it to good use, of course.

David: I don’t want to sound overly self-congratulatory here, but I will say just even from my experience, it takes getting over some of that fear, fear of this being great or being judged. Someone could easily look at some of this stuff and say it’s a stupid idea, or why are you putting out this crappy product or things like that. Even if you have something that you’re showing to a client, I think fear can help you in a good way. It’s like, is this something that they could really do themselves just by spending a bit of time on ChatGPT or did you do something with these tools that not everyone can do, and this is why they’re hiring you and whether it took you 10 minutes or 10 hours, they’re going to want you to be the one who keeps doing that. I think most people know, when they get to that place, when they’re mailing something, or when they’re delivering something of substance. It’s still though getting over that hump and saying, Oh, well, is it bad that tech helped me do so much of my job here? If you still put You into it, then it’s not bad at all. You probably just made yourself way more valuable to that.

Ross: I think that’s a fantastic framing for it. We’ve already got there too, in a way, but just to round out, is there any advice you would have for marketers or anybody on how they should be approaching, bringing AI into their work?

David: There’s no substitute for trying everything. As soon as you hear one thing mentioned in the news, or a friend’s LinkedIn post or something like that, just go and try this thing. Almost all of the main tools and tech out there, it’s either free or it’s like 10 or 20 bucks a month or there’s a one-off thing for like 10-30 bucks, there’s all this stuff that you can try now. Be a little mindful; don’t rush to put customer data into it, you’ll have some of your guardrails; you don’t need to publish all the stuff that you do, but just open up that box, and tinker around a little bit. Then, you just have to train yourself to ask better, tougher questions. You’re also doing something right if you get to the point where you break it. It starts spitting out gibberish, or it’s like, coming up with something totally off.

I put in a play that someone was working on as an input to just go and do some analysis about characters, and I realized when I was using Claude, that Claude was basing everything by the file name of the play and making everything up because it didn’t want to tell me that it couldn’t read the text in what I uploaded. Only when I really pushed it did it admit what its technological limitations were. But it was good because I knew where I broke it and then could work on fixing it.

Ross: It’s fantastic. Thanks so much for your time and your insights. David. You’re already having fun, no doubt, but continue to wish you fun on the journey. 

David: Thanks so much. I look forward to continuing to learn from you and your other guests. Thanks for having me here.