The Real Cost of AI for Advertising & Marketing Agencies

Matt (Guest): [00:00:00] why do people pay you money? Matt (Guest): He's like, I have an extreme confidence in my taste, in my decision making. And that's essentially the operator of an AI is like, it allows you to only have to make the decisions on the things that you feel super confident and have taste in, Kelly (Host): everyone. Kelly McGuire of the OU Day Digital. Excited to talk about today's topic and, be joined by our, guest. Matt (Guest): thanks for having me guys. Matt here. I'm a strategist with BuJo Day and excited to get back on the pod with you guys. Todd (Host): Hello, happy Monday, if that is a true saying. Todd (Host): , This is Todd Kelly and Matt's partner, and we're gonna talk about AI again today. , As Matt enlightened us to a recent, , storyline that's [00:01:00] happened and, , it sort of follows, a discussion thread that we've been having internally, which is What is the expectation of a marketing services firm like ours and the use of AI in the minds of a client and the debate around, does it mean we should be cheaper and faster? Todd (Host): Does it mean we should be smarter? Does it mean we should have more value creation to add more output to give? , Does it mean we should be able to do things we've never been able to do before? , It probably means all those things. , But in our experience now, Kelly, you having some specific experience that I think you want to talk about and Matt, some of your recent experience. Todd (Host): , What we're learning is that the AI tools are like, I don't know, maybe it was like having to learn Photoshop 20 years ago. Like yes, it made the ability to do creative so much more valuable and enriched. [00:02:00] Every designer had to learn how to use Photoshop. And that took time, energy, effort, curiosity, hands on keyboard. Todd (Host): , It didn't just change overnight the way that you could take a, photo shoot and make it look better like overnight. It took a lot of time. And so recently, , as Matt you shared, there was an article that , hit the Wire that, , Deloitte, , . And they're a branch in Australia, , was sued by, , an Australian government department for producing a report that they got paid a lot of money for, , that was found to have hallucinogenic errors and other ai, known content in it. Todd (Host): , And so the client sued them for not disclosing, they used AI and sued them for basically giving them not a great product. , And they won. So slap, slap, ouch. That hurts. , But it comes back to what are we expected to be able to do with AI as a marketing services company? And so far our use of AI [00:03:00] tools has meant nothing more than a whole lot of time, more time, way more time. Todd (Host): Spent through self-exploration and discovery and testing and sandboxing of all these tools to see what they're capable of and do they actually produce anything of any real incremental value. And so let's talk about, , where you guys have experimented with that. I know recently you've talked about using some AI tools, , within your strategy journey, working on some clients and thinking about some things long term. Todd (Host): But it's not like you just hit the keyboard and overnight, voila, you got this magic, you know, thing to output to a client and go, look what I did. Here you go. And it happened really fast and, and it didn't cost you anything. So I, that wasn't the experience. And Kelly, I know yours, your journey's been longer and more, , in development and I don't think that's your conclusion either. Todd (Host): So Matt. what's been your recent exploration [00:04:00] with AI tools? , Don't obviously name client names or anything like that, but talk about the purpose of what you're using them for and what is the incremental value creation that you think these tools are going to allow you to eventually pass on to a client? Matt (Guest): Yeah, I think the. The word is enhanced and not not replaced. And we've talked about that before. And actually as you're bringing up the Deloitte scandal, kind of what restarted in my mind was when I was in high school, , SparkNotes was a big thing and SparkNotes were online. And I actually remember a lot of kids would go in and they would use SparkNotes. Matt (Guest): Entirely when they were like doing their paper, which is like against the honor code and you get in trouble for it. But when you did that, you pulled in a lot of incorrect information. It was just obvious that you were doing that. And what I found a lot of other students were doing is they would use SparkNotes on top of reading the book and having a point of view. Matt (Guest): And what it would do is it'd bring in a totally different lens to like a report. Like I literally remember doing this and it would pull in different themes and things like that. [00:05:00] So if I'm thinking about Deloitte in this situation, they were sort of like the student that just. Probably just use the Spark Notes, copy and paste it in the report, when in reality, what they could have done and, and I think this is the way to use the model, is like you can use AI to challenge your point of view more deeply, inform your point of view, and ultimately when you think about a client deliverable, that's just an enhanced report. Matt (Guest): So you're taking your own sort of experience and your own knowledge, and you're using the model to test that a little bit. Really to like enhance areas that I think you're weak at. So for me today, what I've been using AI for today is, you know, we're coming up at the end of the year and we've got a lot of great reporting for clients and what I'm trying to do is tie the reports together and actually use creative tools to summarize 'em in a different way. Matt (Guest): Whereas I know, you know, our clients receive a lot of decks from us and really great work. I've been playing around actually summarizing a lot of those reports with micro videos that can be shared internally at client side. So again, that's something that I probably could do and take me a couple [00:06:00] days to string together an animation. Matt (Guest): But I'm finding that these models are allowing me to, upgrade the deliverable a little bit. Something that I couldn't do based off my experience in an afternoon. So, , you know, I'm not just ripping something online and sending it over. I'm sort of prompting and massaging it based off my business intelligence experience, and that's ending up with a really cool deliverable that wouldn't just be able to be one prompt and done. Matt (Guest): , So it's a little bit of a, of a ramble, but I feel like you can't use AI to cheat, but you can definitely use it to, elevate what you're doing. Todd (Host): I love the word phraseology. Just use, upgrade the deliverable. So this is where I think, you know, a lot of this talk track goes, is that we get paid to produce a deliverable, , for our clients. Todd (Host): And that's, in our case, it's media strategy, media execution, , media analytics and business intelligence feedback loops. , And. Upgrade. The deliverable, I think is the key phraseology because as I said in the beginning, none of this is making our jobs easier or [00:07:00] faster. We're not getting to the executable easier and faster. Todd (Host): In fact, the AI tools are, are. Taking longer that we have to invest more time, more learning curve, more. Again, it's like learning Photoshop 1.0. Nobody knew how to use it and it took a long time to learn how to use it before it paid off. , So I like the phraseology upgrade, the deliverable, which I think today that's probably where the majority of our responsibility lies to our clients, is that these tools are coming out. Todd (Host): I think it's our responsibility too. Do the demos, go to the trade shows where they're being promoted and marketed, test them out, trial them. we have several paid subscriptions just literally to put these tools in a sandbox and try them out. So we as an agency, have a whole new line item of expenses in our p and l. Todd (Host): , That we're not passing on to the clients. They're not bearing the burden of, of these learning curves and expenses relative [00:08:00] to the technology. , We are, and again, it's human learning curve. It's human time, it's physical cash out of pocket costs, but upgrading the deliverable. If that's the ultimate output, then that means the client's getting more value. Todd (Host): , In terms of what their deliverables are, and I think that's probably the best thing that we could hope for. I think the struggle that, I personally face in the operations and finance side is how do we pay for that? Because we're not getting paid for that today. We're not getting paid for more time and labor and materials to use these tools. Todd (Host): We're not passing the cost of these tools on. Our clients are just getting the direct upgraded benefit from you, Matt, , in a deliverable without having to pay for it. That's not a long-term sustainable sustain finance and operations. . Effort that I don't, I, I don't think any agency could undertake. Todd (Host): And so Kelly, you've spent many more man hours, , time, energy, and effort, , through your sort of personal exploration , where have [00:09:00] you been journeying, , again, in generic terms and what are you finding as the upgrades that are coming out of it? Kelly (Host): it's a really interesting thought exercise to go through and it takes you. Kelly (Host): To many different paths in terms of what, what can AI do? And, we've talked about this a lot, and I think about, ultimately things can only serve a single master, if you will. And, and I think what that means to me in this context is there is a world in which. A lot of media agencies, creative agencies, a lot of the world that we work in has traditionally been very hours based. Kelly (Host): It's charged hourly, it's billed hourly. Everything's based on that. And so this idea of faster, more efficient, , you know, getting rid of some of the low end work. I can see why there's a world that sort of gravitates towards that. I've seen this before at the entry of programmatic. It was all these tools went from negotiating individual buys and, working with different reps to, we can do it faster, we can do it at scale, but is that the master you wanna [00:10:00] serve? Kelly (Host): you know, we've challenged this in a different way for a long time. Should clients be simply just paying for hours? Is that the thing that matters is hours or is. Are you paying for outcomes and expertise and, and skillset? And that's, we've had this talk about being labeled as performance marketing versus agency models and why we don't track hours and, you know, 'cause again, we've traditionally come to the conclusion that if the outcome is as good at two hours as it is at 20 hours, what is the difference in there? Kelly (Host): Right. It, it's, you're getting those results. I think that is so easy to basically say, well, AI can. Eliminate the quote unquote grunt work or the low end work and it can kind of level up and you can outsource all of these things and you get stuck in that idea of efficiency and speed versus. Kelly (Host): What I saw in programmatic, and we made this decision and, and I see in AI now, which is the word that Matt used, which is the enhancement and what does it unlock? What, does it give you that you would not have otherwise had you can now use to look for those better outcomes? And that's [00:11:00] where what you said, Todd comes in where. Kelly (Host): when you serve that master, when you try and use it towards that goal, now you're adding in more time and energy and effort. And the reason that I've experienced that, that has happened is because the analogy that I use is I'm not the most creative person in the world. Right. And, and you and I have talked about this before, Todd. Kelly (Host): Like if we're thinking about a website or a logo or something like that. There's almost this element of like, I don't know what I want. I'll know it when I see it type thing. And that has always worked really well where it's like I could try and describe what I want, but it's hard. But if you give me an array of different options, I can start to formulate really quickly in my head of what I like about that one, that one, that one. Kelly (Host): If you can combine these together, this really gets me and you get to what you had in your mind because you started with all these. Iterative examples that you could kind of immediately crystallize your mind what you do and don't like and, and you can give guidance. Whereas without that, you're operating the dark and kind of guessing your way to an outcome and you're just going after one vision. Kelly (Host): AI is very much that, to me, where I can explore these [00:12:00] ideas. I can describe my thinking process. I can have it trained on the outcomes and what I typically like. I can put it in there. I can kind of see what comes out of it, and then that starts to spark the idea. It starts to give me sm, I like that thought. Kelly (Host): If I combine that, but I want to change that now. I want to iterate here now, and it starts to give me all this exploration of, call it 20 different possible paths to an outcome that I can explore and take the best of all of them as I go through the process. Whereas before, I would've had to have picked one lane, put all my attention in that outcome, and whatever the end of that path would've been the best it could have been. Kelly (Host): Now I get the best of 20 different thought processes that come together in the best possible outcome, but I have to have the knowledge and the expertise to spot and iterate and pick and choose. It's not just go be smart for me, AI and I'll just. You know, this Deloitte thing you described is just like, regurgitate what it put out and, and just hope for the best. Kelly (Host): Like it's enhancing what I'm [00:13:00] already feeding it when I'm iterating it. I know to look for when it's hallucinating, when it's wrong, when it doesn't really make sense. But it's enhancing my ability to ultimately get to that outcome. And then what I really love is the, personalization, right? Something that I've learned really deeply is I have one voice, right? Kelly (Host): But all of the people we work with, they have different things that matter to them, that iterate to them, that, that go through. And I've started to learn to use the systems to say, okay, how do I convey this thought? Whether it's in reporting, whether it's in ideation, whether it's in presentation. To fit what I've learned and seen about this person's point of view and what matters to them and what they care about, and start to form my idea in a way that matches them. Kelly (Host): And that's an iterative learning system. These are the agents that you can build that can start to have that feedback and start to skew these things. We just did it for a client where I built this agent that was an amalgamation of all of our team, sort of. Data and history and learnings over the course of a couple of years, and I sort of built this agent [00:14:00] engine where you could ask it questions and the way that I was describing to people as I was showing it is it's not that different. Kelly (Host): If you were to ask my agent a question and you were to ask just a blank instance of chat GBTA question, you're gonna get somewhat similar results. They're not gonna be night and day different. What the agent is able to do, because I've given it that lens and that point of view and that perspective and that history is to tailor it in and to use data sets in a move in a way that is more relevant to that narrow scope. Kelly (Host): And that's just an example of, again, it's not that the chat GBT version would've been bad. It's just not as good as it could be if you take the time and the energy to think about it in a unique way and to go in a certain angle. And that's where the more you lean into that side of things, the more powerful it becomes, but the more time and energy and effort. Kelly (Host): And so now the only way it's worth doing that is if the output is more impressive, is more valued, and ultimately [00:15:00] you get to a better place. Would it in fact be cheaper to go the other route and say, I'm just gonna use chat GBT. It's gonna be fast, easy, and cheap. No one's gonna test my work and there it's gonna be good enough. Kelly (Host): And this goes into that. What is good enough versus what is great and how do you pay for the differences between those and how do you know the differences? it opens up a whole world of discussion when you think of it that way. Todd (Host): Yeah. So a little bit about what you're bringing up, I think is there's a. Todd (Host): Phrase or keywords that are popping up on my radar in the last couple of weeks. And it's, and I just downloaded this article, I'll talk a little bit about, but it's around synthetic audiences and synthetic personas. , So you talk about an agent, you're basically creating a synthetic agent or person or, or whatever AI that you're training and programming, , by feeding it. Kelly (Host): Yeah. It's a tailored point of view. Todd (Host): Right. . But it's influenced by what you are feeding it. Sure. Largely. And so then you have to ask yourself the question of, well, am I, [00:16:00] is what I'm getting outta that? , You know, is it skewed because what I'm getting out of it is what I'm feeding it. , So the question I wanna ask is, and Matt, you can answer this next is, what is your belief? Todd (Host): In terms of the viability and of these new synthetic sort of audiences and, and what I read recently, , there's a company that got focused on research companies. So think big research firms that put together big panels, big focus groups. all this kind of stuff to do research from brands, well those, those sort of exercises can cost tons of money and take a lot of time actually to schedule and produce and then, and then synthesize and, and report on. Todd (Host): They can take a lot of time. So there's some companies that are coming out that are creating at mass scale what they call sort of [00:17:00] synthetic or semantic similarities with humans. . But with the ability to execute on focus groups. So imagine Kelly, if you had a panel of, or Matt, let's say in your case, you're thinking about the stuff that you're thinking about in business intelligence and strategy. Todd (Host): Maybe it's even for retail marketplaces, which is one of your specialty areas of, , work Is there a place for semantic or synthetic. Groupings of, cohorts or, people or panels or whatever that you, you think you would rely on or trust? Matt (Guest): I don't, I don't think I would do it immediately though. I definitely think that's where it's gonna go. I mean, I think in three years, if you think about the math of how quickly these things are accelerating, getting, better, I would bet on in the future. Matt (Guest): But today, in the moment, no, I, I don't think it would. Fully replace a focus group for me. It would get you started though. I mean, you could kind of use it as the prefo group. Maybe you kick it, a couple quick reactions or something. You use it from there. But, Todd (Host): this company that I've [00:18:00] read their white paper on, is claiming that it, achieves a 90%, , human test, retest, reliability. Kelly (Host): not to get you on Matt, but there's a thought here because you and I have really had this debate, Todd, about the value of research and the outcomes, especially when it comes to personas and profiles and customers and what you're talking about to me, I think feeds exactly on the point of view I've always had on that, which. Kelly (Host): I've inherently always sort of dismissed the value of that because I've always believed that any level of experience or intelligence or the ability to think about a company, a brand, a product, or a problem, you can get pretty close to pretty much laying out or guessing what the outcomes of, a focus group or an audience group would be. Kelly (Host): You know, the value proposition of the product, the targeting, all of those types of things. And so for me, I've, never quite seen. The value in the cost benefit to truly getting that in most cases. I, I won't make a totally blanket statement, but the interesting part of the, what you described on the AI side of this is [00:19:00] what I believe it lets you do is if I can think through who the likely audience is, who should respond in all those things. Kelly (Host): It gives me a system to stress test those assumptions. And most importantly, if you have that knowledge base and the expertise to use that system the right way, you can start to formulate that into actionable outcomes. Okay, if this is my audience, what is the messaging that should resonate? What are the things that should work? Kelly (Host): And again, I go back to this ideation. It gives you the chance to get those 20 different logo iterations so that you can start to see the array. It starts to paint the picture for you and you can say, okay, yeah, I really like that output, or, that's an insight I didn't naturally think about, but now it's in my brain and I, you know, it just gives you these puzzle pieces that come together. Kelly (Host): And if you're smart and you have that expertise, you can put that together into an incredible picture where, again, if you didn't have that feeding engine to be able to use it, you again, you may have one single thing that [00:20:00] is on your mind then, and you just go down that path. So I, I see it the same way that I just described, which is. Kelly (Host): do, I see a world anytime soon where there's a tool so good that literally anyone could just go in, type in a few things and just use the put and it's being useful. No, I still think you need people with the experience and the understanding of what to put in, what to take out, how to iterate and to go through it, but having access to a synthetic focus group that you can know how to use and know how to. Kelly (Host): Really understand the outputs and stress tests I think becomes incredibly enhancing and valuable overall. And I think it also, the last thing I'll put on that is it opens up, my other thought around this, and this was a big subject at the conference I was having in the way that agencies and things are trying to use ai. Kelly (Host): There's this whole concept around AI replacing the entry level employee and how if you're coming outta school and you have no experience in the field, it's hard to break through because AI can do sort of that, again, that historical grunt work element, but that's gonna create a [00:21:00] really bad problem where you have a very unbalanced workforce of people who have the experience that can use these tools and actually make use of them and get value. Kelly (Host): But eventually, you're never gonna get anyone that breaks through and gets their first 1, 2, 3 years of experience so that they can make it to the other side of the moat and be the users of the AI versus the being replaced by AI element. And that creates a whole kind of fundamental existential crisis in the next couple of years if we're not careful as well. Todd (Host): Wow. Yeah. That's happening across the board. I saw a headline today that Amazon's gonna lay off. 30,000 employees you'll see tremendous Kelly (Host): productivity and value out of that in the short term. But when you look at the medium and long term, what happens when you need to grow the skilled workforce that hasn't been backfilled into that, and you run outta things, you're drying up your own future pipeline, right? Kelly (Host): It's, it's pumping the oil out of the well faster than it can replace. And it's great when you're. Selling those barrels of oil, but what happens when it starts to sputter and run dry? Todd (Host): I don't know. I just think this is an interesting new frontier of, of [00:22:00] synthetic focus groups, you know, synthetic personas and using them as, you know, the field studies for how is messaging gonna resonate? Todd (Host): How is creative gonna resonate? what and how, and why. Do certain purchase habits exist for certain types of products and services? Again, Matt, you've spent a large part of your career in e-commerce on the brand side. I mean, is there a realistic moment in your mind if you were still back at one of your e-commerce brands where if you had a choice between a real human focus group of 5,000 people, or. Todd (Host): A synthetic focus group of 5 million people, which would you choose? Matt (Guest): If I'm selling the good to a human being? Like a human consumer outta all, well, yeah. Todd (Host): You're, you're working for one of your e-commerce brands, a fashion and apparel brand that sells clothing and gear, right? If you're still working for that brand, Matt (Guest): yeah. Todd (Host): are you more likely to trust 5,000 humans? Kelly (Host): let's use [00:23:00] realistic numbers here 'cause, because you're skewing them too high. 'cause I'm, I'm in the process of a research group with a client that, that we're not in there right now. You're talking about usually like 50 humans against, you know, 5,000 synthetic. Kelly (Host): Like, that's the real deal because I just had this thought in my head as. Can we really tra is 50 representative of enough of a sample to make critical decisions on the outcomes of this? No, Todd (Host): but that's what I'm talking about. The big research firms that do have focus group panels of thousands of people. Todd (Host): They do. So is that the norm? Kelly (Host): I, I, I, I've not big research Todd (Host): side stuff. Yeah, absolutely. , It's not 50, it's thousands. , Because the statistical relevance isn't enough with 50 people, as, you know, pining towards right there. But Matt, like, I'm just curious, if you were in the e-comm director seat back on the brand side and you were being challenged with this sort of thing, or you were gonna challenge your agency with this sort of thing, would you go human or would you go synthetic? Matt (Guest): I do both. I think Kelly kind of said it. I mean, if you go back to like a logo for example, what I would do is instead [00:24:00] of showing a human focus group a hundred different iterations of it, I would use the synthetic group to narrow that down to 10 and then really work the focus group around 10 to find like the magic. Matt (Guest): So I would make it easier to use the human group, but. And maybe that gets better. I'm just thinking about present day, the way that would ease it maybe in three years that, you know, it totally can replicate the human. But I'm also thinking now, if everybody's gonna begin doing this, whether that's like a logo or a go to market plan, I actually think going against the model is gonna be actually the way to differentiate. Matt (Guest): So I'm thinking about logo design of. Whatever the opposite of that AI generated is, is actually what's gonna get people's attention in 10 years. So there's also some sort of interesting way back to the human, , once the whole market adopts this approach, Kelly (Host): well, once everyone's using the same system in the same way, now you've created Homogenization again and yeah, now you've gotta figure out a new breakthrough. Kelly (Host): 'cause it's just raised a whole new. Differentiator and it's noisy and it's the same thing over. I'm curious about when you think about it on, you know, 'cause your background's a [00:25:00] lot in CRO and. You know, some of the branding stuff and some of that decision making. I mean, that's where it gets really interesting is, you know, traditionally where we've worked with clients that struggle, and Todd, the reason that I bring up the small sample sizes, again, a lot of times we're operating in a space where clients can't afford a research group. Kelly (Host): They can do thousands and thousands, right? So we're trying to operate. What I would say represents the more broad spectrum experience of companies that aren't Coca-Cola, Starbucks. Right. That's a small, you know, one percenter experience and we're looking at basically what can everyone else get access to and use, which is the beauty of AI is it's supposed to level some playing field there, but I'm curious about when you think about it that way, like how does that ability to again, properly understand how to use the AI start to allow you to maybe do. Kelly (Host): Iterative testing on creative or landing pages or you know, those types of things, like that's where I start to think that it could be interesting. But again, there's no easy button. There's no, just press it. It does it for you and you apply it. What's the Shakespeare thing about like a thousand monkeys on a thousand typewriters for a thousand years could write, write [00:26:00] shit like it's not just gonna work like that. Kelly (Host): It just magically happens, right? You still need guidance and control and application and analysis, but you can fuel it so much Matt (Guest): better now. Yeah, it would just produce better work in my mind. I don't think it would totally replace anything. I think I'd probably be looking across way more attributes on a CRO level and have like a better outcome I would get. Matt (Guest): Stat six quicker and everything. So, , I don't think it fully replaces anything. I think it just enhances or elevates the amount that I could do within a project. It makes that outcome that much better. Todd (Host): Maybe that's the upgrade, right? That's what we talked about originally. That's the Matt (Guest): upgrade. yeah. Matt (Guest): What would've taken six or seven multi-variate tests can probably take one or two, Kelly (Host): maybe 10. I like your, I like your example around like turning a hundred iterations into 10 good ones, right? Because it feels like there's a chance for it to. For it to kill negative tests faster, right? Yeah. Find landmines and potholes faster so that you can concentrate your actual investment and efforts into a informed [00:27:00] test subject group versus sort of a, a randomized one. Kelly (Host): Like there's an element of that that gets really interesting and in intriguing too. Matt (Guest): Yeah. Yeah, that's, so I was trying to think with like logos, like instead of looking at a hundred different font types, maybe you're looking at 10 font types with different density of font or something like that, you know? Matt (Guest): And that's where the magic is. Kelly (Host): Yeah. And it can running that system through and stress testing and saying, what about this logo with this group? Like the system can maybe point out something that. whether you agree with it or not, it can give you a point of view that your brain's just not naturally floating to the surface. Kelly (Host): 'cause how can anyone think of everything? Yeah. And once you see it, it's that light bulb of like, oh, I, I never thought about that, but yeah, this font wouldn't work that. Or, oh yeah, this is too close to this other one, so I need to make this change. Like, that's the power of AI to, to me, is again, it's, it's enhancing that brain, right? Kelly (Host): It's that ability to say for whatever reason, that idea just wasn't the one that floated top of mind. This thing surfaced it. Now I have the expertise to apply that, and now my next iteration is that much better. Whereas had I [00:28:00] just been relying on myself if that thought didn't float in, I would've just kept going without that input. Kelly (Host): And it just adds all these inputs that you know, at any given time with all the distraction and the movements we're having, we can't have it. But if you have these nice systems. it's like giving you your, your moods and your feelings and the way your brain works if you train it right, it's giving you full access to that at all times. Kelly (Host): Right? It's now it's formulating in my mind, if they say, you only use 10% of your brain, right? you're using a different 10% at all times, this maybe is giving you 50% of your brain. 'cause it's giving you iterations over the last six months. And it's learning on those types of things. And that's the enhancement element if you use it right. Matt (Guest): It's like the Rick Rubin interview quote. Have you guys seen that? Where he is like, he's getting interviewed and, and the guy's like, so what do you do? And he's like, well, I have no technical ability. And he is like, well, why do people pay you money? Matt (Guest): He's like, I have an extreme confidence in my taste, in my decision making. And that's essentially the operator of an AI is like, it allows you to only have to make the decisions on the things that you feel [00:29:00] super confident and have taste in, which is maybe your example, Kelly. Kelly (Host): and then that goes full circle back into the should clients be paying someone who can use this tool to produce these incredible outcomes because it's informed by their system and they know how to use it. Kelly (Host): Or do you just want to pay someone because they now took 10 hours of work with it into two hours of work, but you end up at the same place. It's just faster. Todd (Host): So Kelly, on that point we will wrap up here. How many hours, if you had to think about dating back to however many months it's been since you started going down this sort of deeper experimentation journey, , how many hours do you think you've put in to this? Todd (Host): Just your own creative, curious. I'm going to put time and effort, energy and to figure this out, and, and it was largely tied to some client sort of background, , prompts for you to do this, but how many hours do you think you've invested in this Kelly (Host): 2080 working hours in a year, minimum of thousand hours. Kelly (Host): I'd easily double that Overall. [00:30:00] So what does that put me? A 10th to a fifth of the way to the 10,000 hour expert mark. what does it take across that? No, I mean, in, in all seriousness, I mean, easily a thousand hours in terms of, but again, it's not just replacing work, right? Kelly (Host): A lot of it is that experimentation and that learning. So there's, oh yeah, there's fun to it I'm adding all the hours between. Doing client-based enhancement work to just, I'm curious and enjoy learning and experimenting thing. 'cause I use this a lot for my own passion projects and my own types of things, but all of that utilization plays into the expertise of knowing. Kelly (Host): How do you engage in LLM engine? I mean it all the way down to where I'm still not even an expert is I primarily use chat GBT for the most part, and there's. 20 other systems that specialize in different forms. There's what Matt's doing with the video side of things. There's the image generation side of things. Kelly (Host): There's the research side of things. I literally just in the exercise I did for a client, brought my own self down a rabbit hole where I was trying [00:31:00] to develop an agent with an interface and I didn't even really know the whole engagement with Python and React. But because I was asking a certain way, it started to suggest to that it could build a web interface. Kelly (Host): And so that took me down. Accessing an add-on with a React code and a JS note. Things that I didn't even know, but it taught me through, and then I created an interface and I could take that even further and make it and doing it offline and connecting. And so I guess the answer is, is it, feels like every little thing becomes, if you care enough, it becomes 10 x the effort. Kelly (Host): But it's so cool on the outcome you get to, if you're curious enough and excited enough, I could have just stopped at build an engine, read this stuff and let me ask you questions. But I found it so interesting to be sucked down in that rabbit hole and learn all these things and enhance that, and the next time I'll be better at it. Kelly (Host): That it creates that expertise. And I would be looking for people who use it in that way versus I don't really feel like running this report, so now this does it for me. Todd (Host): Well, in conclusion, what I've heard [00:32:00] is you two are doing a lot of experimentation to add a lot of upgraded value delivery, , through our marketing services company or agency, whatever you wanna call us, , that our clients are gonna benefit for. Todd (Host): , As on the agency side, , we have a real challenge and a real problem in that nobody's paying us more for this. It's just an expectation that it should happen. And nobody's paying, paying us for it. Kelly (Host): think the bigger problem is people not understanding so much that yes, it's about paying more for it and all of those things, but it's that it is a replacement. Kelly (Host): This idea that, you know, we've been in these pitches in these talks lately where it's like, how do you use AI in your process? And it's the answer that I think they want is this magic bullet that somehow just exists versus this enhancement conversation of knowing how to use it. And I think the expectation or the hope is misaligned with the reality where everyone wants this magic box that is the solution to everything We're [00:33:00] in this world where there's this lack of full understanding that the shinier it sounds and the prettier you make it appear. There's just this, that sounds cool. I want it. Versus the diving in and being like, what does that really mean and what does it really get me? And until that changes, I think we're in a bit of a snake oil environment where people can sell vaporware in anything they want. Kelly (Host): And you just get this buy-in and I think there's gonna be a lot of disappointment. For a while before people start to catch on of like, okay, I need to look at the substance behind this stuff. Which again, circles back around to, get the substance people have to pay and dig in and all those other things. Kelly (Host): But just wanted to clarify because I think that's important that there's this idea of people exploiting the lack of understanding and the, the shiny object that is creating a big problem in the way a lot of this is being pitched and sold. Todd (Host): Yeah, I, we've seen recent, , award-winning agencies. A headline on their, on their homepage and maybe some of their category award submissions where AI powered agency AI power. Todd (Host): I don't know what [00:34:00] an AI powered agency is. Does that, does that mean, I don't know what that means. I don't know how a client could know what an AI powered agency is and how that differentiates and adds that incremental accretive value to the outputs, Matt, as you were describing. But anyway. , there's always good stuff to talk about here. Todd (Host): again, what I'm hearing is our teams invest a ton of human energy, time, and curiosity into learning new tools which provide upgraded value for our clients. , Lucky them because, as I said, none of them are paying for it. , And part of it is just probably that expectation and, , sort of our fiduciary. Todd (Host): Responsibility to deliver more. , But I think as it gets more complicated and more depth of use, Kelly, to your point, , it's not sustainable to expect that as just some cost cutting, freebie efficiency, creating, , whizzbang, you know, shiny box that just makes everything easier. it makes everything. Todd (Host): Better and more difficult and takes more time and [00:35:00] energy to get more out of it. , So we'll keep, doing that and I'm providing more value and hopefully. That will help us do our work much better. , Thanks for going down some rabbit holes today, Matt and Kelly. , I'll think of some more rabbit holes for us to go down

Outro next time.

Outro Thanks for listening to contrary to Popular Opinion where we challenge you to think independently, push the limits, and take uncommon action so you can achieve game-changing results in your business and your life. To stay up to date with our latest episodes, be sure to click the follow or subscribe button now. intro: Thanks again for listening. See you next time.