s02 e07 Marty Samples & Marketing Automation

Welcome back to the Big Cheese's AI podcast. Our guest is Marty Samples, a digital technologist. Today I'm your host, sorry about last week, I was a little tired. Really anti- like marketing, creative professionals that are like super anti-AI, and they're like, they're just like, we don't use it, we don't believe in it. And I think, what's up with that? Because it's like, you can use it, it's not black and white. You can use it to assist, to get you to better ideas, to ideate, and then insert your creativity. But I think is it because everyone's spamming just a bunch of bullshit out there, and it's just seen as a net negative that they have to be superior? Or like, what is that? Like, pure as mentality or why? What? Well, I think some of it is. I don't think they understand. I don't think there's a clear understanding of what is AI and what isn't AI. And really, there's some anxiety about, this is going to negatively impact my job. For example, if I'm a content creator, And we've got six content creators in our department. Let's just say that's true. Of the six content creators, we'll three of them go away because now all of a sudden, 2GUN.ai can not only generate good ideas, but it understands the algorithms of all these walls, gardens, like Facebook or Instagram or LinkedIn. And now all of a sudden I can generate content that fits the algorithm and performs better. Is that going to impact my job?

• I have a question. Since you've been involved in larger organizations, Does marketing sometimes exist too much in like a, like not a silo, but like an area where it's hard to hold marketing accountable in some ways, right?
• It is.
• And is that--
• They're like narcissistic abusers against you.
• Right, right, right. Like you see like--
• You see marketing's always trying to justify their, marketing's trying to justify their value with like metrics that aren't necessarily like tied to the bottom line. like how have that Super Bowl commercial do for Taco Bell sales? You know, I don't know. Like there's no attribution there. So like does- That's it. You hit on a very key word there. And I think this is really important. If AI provides any value moving forward for marketers, it's going to be how do I know what's performing well, what's not performing well and what metrics I should follow? Because I don't think there's always been good data. But what I'm starting to hear more in the marketplace now is marketers are being evaluated by this ROMI, return on marketing investment, and they're being asked to basically defend the money they spend. I don't think that's always been true. No, no. I mean, it's something that it's like, well, how am I supposed to... Well, I've got conversions. I've got website visitors. I've got these metrics. Impressions? Yeah. Oh, yeah. Oh, no. That was a... I got 450,000. Please, Miss Jan, is that right? Yeah, yeah, yeah. And then half the time, but half the time we're sitting there going with these marketers and we're like, how do I set up the goal conversion tracking? I mean, the most basic stuff, right? And maybe that goal that they're tracking isn't even tied to revenue. So I think sometimes maybe when it comes to why marketers may or may not be adopting this stuff, is I think there's probably some fear.
• Well, let me give you a concrete example. So several years ago, we worked with a health insurance company. They're one of the big four. If I named them, everybody online would know, I'm talking about I won't do that 'cause this might be embarrassing. (laughs) And at the time we were helping them rethink not only their website design, rethink their SEO and how they were engaging with customers. And their number one collection was when somebody comes on this website, we want to quote insurance. We want them to go immediately to the quote quote insurance. So we started looking at their SEO. They were ranking fourth, fifth page on health insurance quoting. They were ranking number one in Fettuccine Alfredo with spinach. [LAUGHTER] So here's the thing.
• They had 300 pages of healthy eating content on the website that was not driving anything but traffic and they felt like they were doing great.
• Yeah.
• Right? Well, if you've ranked number one for Fettuccine Alfredo with Spinach and you're on the fourth page or fifth page for health insurance quoting, there's something wrong with the way that you're doing that tomorrow.
• Well, I've seen that in e-commerce. So a lot of e-commerce sites will have a blog, right? And they're like, "Ah, our conversion rate is so bad." I'm like, "Well, half your traffic's going to a blog post."
• Exactly. So that's not really high intent. (all laughing)
• I mean, it kind of reminds me of that shift where everybody was really worried about bounce rates for a long time. And then we started to think about that a little bit more critically and we're like, well, if they got what they wanted and then left, that's actually a good thing. Like they didn't, most people don't surf the web. Like they're like, oh, let me go to CraftitUp.com.
• I think Google tried to reinvent their entire analytics platform to get away from that metric.
• Yeah.
• And I still think that GA4 is a disaster.
• Oh, well, that's a weird one.
• It's been a nightmare.
• I really wanna know what the conspiracy theory behind universal analytics to GA4 is. It had to be something related to the dark hand.
• Thank you.
• Because GA4 is such trash, dude.
• Yeah, yeah.
• Now again, I think that ship is sale. You know, we can all look back and remember like the glory days, Napster. Right? Like, I remember how awesome it was.
• Remember when we had a,
• And we're in Google Analytics, you used to be able to see the keywords.
• Yeah, no.
• They would just show you.
• You know, you can, you can still get to them, but you just have to pay $100,000 a year.
• Really?
• It's to get the enterprise.
• There's a cost 360, you've got to get the enterprise version.
• Yeah.
• The thing I didn't even know that.
• Oh yeah, and it's ridiculous.
• The one factor I do like though about GA4 is now the tracking's all tied to a user as opposed to an event.
• Yeah.
• And with the right kind of configuration, you can actually--
• Exactly, the--
• The tracking user problem with my marketers do though. They go, oh, what the hell is this? And engagement rate, like, what's this? I have to click around and make buttons happen. Like, so yeah, it is a, it is a, and that's kind of one of the things I wanted to talk about, which is AI, and we've talked about this. This has been a theme since last season is AI is suited in data ready environments. And a lot of folks go out and buy software. They buy MailChimp, they buy, you know, Clovia, they buy this, they buy that, and they dump their CSV of all their people in there. And they start blasting the list, right? There's no usage of the nuanced features of the platform. So what would make you think that they're gonna use the new AI feature on the platform if they're not using anything more than the basic 101 stuff anyways?
• And what happens too too often is everything that a person does in marketing drives to a website. Think about it, every traffic driver you've got drives to a website. If it always drives to a website and you don't know what happens after the click. >> Yeah. >> You don't really have any good information about who this person is or what they do. All you really know is that first click. Once you get to the website, now from there, it's like a black hole. Where AI can provide a lot of value is now with Salesforce Marketing Cloud, you can actually assign an ID to that person, that ID can be tracked all the way GA4, you can bring it back into the customer data warehouse, you have the ability to kind of see that whole cycle, and AI helps you analyze what's that journey mean. >> Right, yeah.
• I do think so. So let's say you're in that data ready environment. Where AI to me comes into play is summarizing that information and making it more digestible because that's where LLMs are really valuable. But I also think and what we're seeing with, so Clavio is another email marketing company. And what I've always looked at them since we started doing more Shopify work and people are like, "Clavio, this Clavio, "what the hell is Clavio doing?" You know what it was. And we started using it, I'm like, "This is really good." Well, they're saying, you know what? It's hard to create flows. It's hard to do segmentation. And so they're using LLMs to say, just describe what you want. Just tell me what you want and I'll go do it. Instead of clicking these links and putting these flows together.
• Oh, yeah, I'm full.
• And so people are using natural language to help set up those segmentation, to help set up their flows. And I think that's a, that's a maybe more of a approachable way. If so, if you're a marketer and you're looking like and saying, how do I use AI? Check out some of the tools that allow you to just, instead of getting real technical and having to go in GA4 and click all these reports and do all this stuff is just telling what you want. >> Well, that's an excellent point because an observer, an agent, can do these things at scale and in real time and they already have the experience of how do you set these tools up. The reason people don't like GA4 isn't necessarily because it's bad, because it's a new tool, it's doing things in a different way and now I have to relearn this platform, but the value in AI, what they're good at is marketing. That's their task. They're not good at implementing GA4 or whatever, Clavio or whatever. And they're not particularly good at segmentation in general. So what am I gonna do? Look over 10,000 visitors and which paths they chose and then manually segment them. And then if that person decides that they're gonna do some different action, like move them, AI can do that. But what I can do from the segments from there is build the campaigns and like, you know, the next actions or whatever I need to do that's actually more impactful, right? So when they'd be cool if like AI could just like observe my users and be like chewy.com or something. I have two dogs and a cat. And all of a sudden I'm buying lizard food. Well, maybe if I buy a few times, they're like, well, this dude probably has a lizard now, right? And maybe I'll put him in that new segment. No, I don't, but I'm thinking about now. (laughing) No, my brother had a bearded dragon.
• Oh yeah, those are pretty cool.
• They're pretty cool.
• When we look at the other side of, if you look at it from the other side, we talked about this when the Super Bowl happened. we were really interested to see what happened with AI in the Super Bowl, and that was that minion's commercial.
• That was so funny.
• But like the takeaway on the podcast was human creativity killed it in the Super Bowl. Like the top tier content that they were creating, the highest level marketing agencies that made the best commercials, that AI couldn't even touch that.
• No, I agree. And that's what we're, I think that's where we're gonna see is that, what AI is ultimately gonna do is it's gonna force the creatives to be more creative.
• Yes.
• And that's exciting. Like that's really cool. 'Cause like as a creative myself, it's like there's a competition now and we're competing with like this global memory like mine and to be able to outperform it, I think is what we're gonna see. Now again, does that mean a lot of the mid creatives disappear? Absolutely, right? Like they're gone. They're, you know, no, maybe they'll be helpful because they'll use AI or whatever. But no, I think, I think the AI is ultimately going to push us to be way more creative. And I think accelerate the idea generation. And for me, I think that instead of spending days in rooms going through brainstorming sessions, you're going to get to ideas that are executable quicker, which I think is going to be a huge value of AI. No, so one of the things, as we were kind of thinking through this marketing automation-- and I've now had an opportunity to talk to a bunch of different marketing companies for these startups that I'm working with-- the one thing that I think that if you're marketing company today. I want you to come to me and I want you to say, we're going to put your marketing on autopilot. Right? We're going to set this thing up. They're going to land on this page. They're going to fill out this form. This form is going to be sent to make.com. It's going to analyze it using chat GBT to figure out what they're interested in. It's going to automatically categorize them. It's going to add it to your HubSpot group. It's going to follow them up with a secondary email saying, oh yeah, you said you were on Android and you're interested in our thing. Here's how you go and install for Android. this hyper personalized experience that is 100% automated, I don't wanna talk to a single person and we just set it all up with them. And make.com is one that I'm starting to become more and more bullish on. And if you were on TikTok anymore, you'd quit her. You would see that there's a huge make, like a make talk. This is what you'd call it.
• Where they go through and they show all their workflows. I'm like, this is some really bad ass stuff.
• Before I get off there, that was a lot of the automation workflows were on make. And then I'm trying, world.
• Oh, when mill.
• When mill as well. Which I just--
• It's an open source, it's basically an open source version that's similar to make.com, but you can run it on any longer thing.
• But just like what we were talking about earlier with engineering, 'cause we've been doing some AI stuff, developing stuff that we don't even know the programming like this is where we're looking. I feel like the person that is really good at execution needs to still, Those are the people that are going to be successful because it still takes a mindset, it takes an ability to get things done 'cause there's these capabilities that you guys are talking about, they're just insane. And finding that tool and being able to stitch all that stuff together and not go, you know, Bobby and IT's, he's gotta figure that one out for me. I can't do this, I can't figure this out. There's gotta be a can-do mentality or else those, you're talking about these mid-level marketing. There's almost like a AI tenant marketing engineer kind of person that has enabled now. Well and I would agree with that and I think what you're going to start to see is you're going to start to see fractional marketing organizations play a role in that. I think IT's going to largely be bypassed in some ways because what's happening in our IT group and they're real forward thinking but there's risk involved. Their number one task is to reduce risk. There's number one task is to introduce risk. (all laughing) You've ever been a marketing, you know that you're a task. And you enjoy it, right? So you're looking for somebody who's gonna help you get past the risk to something that can be executed. And I'll tell you the truth, most organizations, even large ones, don't have a marketing staff that's both trained well enough to do that, or even an IT staff that's got the bandwidth to do that.
• Yeah, we saw it in the development world where a lot of technologies shifted to the front end or were hosted and enabled people who are front end users to use them. So, and the whole point of that was to do more with less, right? >>Yes. >>Now, a lot of the front end developers didn't have the context of the backend developer's experience, so they were unaware of some of the risks that were involved in doing those things. So, I think it's a similar situation. >>Agreed. >>And to your point about if something's fully automated, I think the risk there for people is to say, oh, everything is just going to be done. I'm done with my job. that can move on to something else. Well, now their task, instead of spending all that time learning how to use the tools and set it up, they need to monitor, analyze.
• Yeah, 100%.
• Way more opportunities for our bats, where you can say try and test hypothesis, get it out in the market faster, fail faster, you know, and all those things that we think are pretty good, right? Like we want as many opportunities to get in front of the customer in different ways and double down on what works and throw away what didn't work, right? Instead of taking six months to do that, I don't know, it can be done in six days now. Yeah. Well, and then here's the other, the antithesis of this. You're talking about stitching products where you even know what they're called. There's companies out there, and this to me is a big risk, what's happening right now in the market. And Brandon was the first person to bring evidence to this, which is you've got a company charging $60,000 for a CMS per year just in terms of licensing. It's complete black box, right? There's no, it's a done for you. Oh, we got AI, we've got everything. Here's all the tools you'll ever need. Just sign this contract with us and we'll make it all happen for you. And that is, don't do that. (all laughing)
• This one really popped up and this was a Reddit, started with a Reddit thread and then I started searching around And it's basically a company goes and they build their entire product on webflow. Right? So low code, no code platforms. Everybody's like, this is the way to go. This is the way you should, you should be really be building. So they go and they invest a bunch of money and time into building out this enterprise kind of like high level application built on webflow. Well, the traffic that they were getting became a problem for webflow and webflow reaches out and says, Hey, listen, we're going to have to migrate you to an enterprise plan. or you're gonna have to find a new CMS. All right, so just, here's your two options. Either migrate where we're going from, I mean, it was literally like they were paying 160 bucks a month to something like, it was like 35 times as much for the enterprise offering. So it was basically a breakable bill that they would get from them. And/or they end up having to basically go find a replacement. You can't. Like if you've built your thing on Webflow, you're stuck with Webflow forever. And so that is like an interesting thing. I think marketing people really do need to keep in mind when you're thinking about these no code, low code platforms, even make.com, right? Like with make I can go and I can glue these things together. The fact is you don't actually own your application anymore. You don't own it. And you're not gonna take a Webflow thing that you've built and be able to migrate that to a next 15 or whatever the hell it is without basically just starting over.
• There's companies that are good and we have clients that are in that position now that whatever the marketing team dreams up can happen. And it might not be the tip top tier, but they got what they wanted pretty quickly 'cause they owned it all the way down to the metal. And sure, there's no code look of platforms out there, but if you relinquish this control, first of all, you're not learning anything. You're not getting better at anything. You're in really, especially when you get to the end, You probably know this from dealing with enterprise sales guys, is like when you have an issue with the company, a lot of these companies aren't prepared to service enterprise customers.
• No. - You know what I mean?
• Not at all.
• Yeah, you're not talking to a human.
• No, their whole business model was built off of no one ever contacting them.
• Yeah. - Right? It's just a SaaS product, right? You don't have a sales engineer, let alone a customer support dedicated specialists who's in like that stuff costs money. Our margins just went down because you call this. Right. Right. Yeah, for sure. Exactly. Yeah. And we saw it with Theraprew. So Theraprew's current website, we built it all on Wix because again, it was just a matter of we just need to get something out there quick, get it out there. And so we have a website. But the reality is is like the more that non-technical non-designers interact and they start like moving stuff around on the Wix site, it just ends up looking horrible. Like just bad. And it's just like, it's no shame on them. it's just they don't know how to do web design. And so it is, it's like, there's this weird median that we're at where you can get people who don't necessarily know how to build the things to be able to build them using these low-code no-code tools. But the experience sucks. And then once that company is like, "Oh yeah, we realize that we can't offer this for free anymore, we need to get $50,000."
• So I think anyone who's like trying to figure out, okay, what tool should I use?
• Locode no code is great for experimentation. You don't know what you're trying to do yet, so you wanna play around, learn the space, learn how to use the tools. Great. But when you're ready to go to scale and production, you need to think long and hard about the long-term solution. And what, is there value in building something and owning it yourself? Or is this something you're gonna throw away in two years anyway, so why would you go that route?
• I think if you're gonna stick with the no code low code platform, you need to get on that user group. You need to become part of that community. You know, you need to tie yourself into vendors that are specialized in that, right? You cannot just say, oh, let me call my rep, right? Like that's your answer to everything.
• Well, and you can't even assume they'll be around two years from now.
• Exactly.
• I mean, to be fair, how many AI companies have folded in the last 18 months?
• Right.
• I mean, I mean-- - Big Gs? (all laughing)
• If you made bets on an AI company 18 months ago, You may have made the wrong bet. Yeah. Well, yeah, and a lot of those startups, even if they're successful, they'll get bought and killed by Google. Exactly. Right. And there's risk there. I think that just, it comes down to, I think marketers, the execution marketers, it's very similar to a developer. Yes. It's very similar. You're really getting these tasks done that are specialized and there's platforms and technology involved, and but there's content and there's creation involved. But the ability, what makes a great developer or a great marketer to me is someone that understands the foundations, right? They're lifelong learners, but they're actually also really good at executing in the modern environment, meaning they know how to use the tools, they understand the tools, they know enough to be dangerous in things that are even tangential to them. And it doesn't matter what platform they're on, they're gonna be able to evaluate it, right? They're gonna be able to evaluate the work, they're gonna be able to say in understanding like how this is all stitched together so that if they do change platforms or if they're always gonna be valuable. So like it comes down to, if you're in that space, you just need to get good.
• Yeah. (laughing)
• Yeah.
• Just get good.
• And you have to understand the platforms that are significant to the marketing execution role, right? It would be crazy for me not to have some experience with Salesforce Marketing Cloud if I'm gonna be in the marketing space. It'd be crazy for me not to have some experience help fund, not know how to do data extensions in marketing cloud, not know how to do basic things that are going to help me with my marketing automation. That's going to be important, right? I might not be an expert in CRM, but I know what they are, I know how they work, I know how to set them up, configure them, use them. I think that's going to be true of any market, right? Yeah, well, then you can actually be more educated as you talk to the people who are executing it. And you know, you can sniff out if they're blowing smoke up your ass or, or, you know, >> Yeah, yeah. (laughing) >> So there is one that I just started playing around with today that I wanted to bring up, which was anything LLM is what it's called. >> I've got it up here, yeah. >> Yeah, so it's anything LLM, I think you can just Google that. And it's an application that you're gonna download, it's available for Mac, Linux, and Windows. And what I like about it is that, so it's basically, you know, imagine that you're installing like your own version of Chat GPT, but you can pick what, you know, AIs behind the scenes. So you can say-- - That's by Y Combinator.
• Is it Y Combinator? - Oh, no, no, no. Okay. - That's good.
• So you can go through and you basically be like, you know, I want you to use chat GBT or I want you to use O llama. And so like, if you're on a Mac, you have a llama, then you can just, you know, run it all purely local. Where they've actually done some really interesting things and this actually ties to some of the things that we talked about with Aaron last week on about prompt privacy is they do have now the idea where you can go and you can connect different integrations. So you can go and connect confluence, right? You can go and connect an entire GitHub repository. You just give it the URL and it's gonna go start indexing the entire repository and basically generates a huge vector database that allows you to chat with it. And so that one's actually, it's a bit--
• This is sick.
• It's very cool. But again, where I think, where we've run into a problem is the same thing. And this is what Aaron had solved, is that, so again, I go and I connect that, my GitHub repository. Once I do it and it indexes everything and it basically builds out the vector vault and I can chat with it, if I go and I update anything, the vector vault is an updated, right? And so, and that's where everybody goes wrong with RAT.
• That's where our entire conversation with Aubrey stalled out, was because it's impossible to keep these things up to date.
• Exactly, but that is what, And that's the mode that prompt privacy has, right? That prompt privacy is--
• This is your weekly prompt privacy commercial brought to you by now one. (laughing)
• Brawny by Brandon, who's a shell, because he's my client.
• So what's wrong with rag?
• So nothing's wrong with rag, right? What's wrong with the way that we do rag today is I go and I say, hey, I want you to go, so any LLM, another thing you can do, is you can actually do a web scraper. So you can put in a URL and you can be like, I want you to go four levels deep and it's gonna go and start just scraping the website and basically generate a vector vault based on all the content of the website. Great, prompt privacy has that as well. That's very cool. The only problem though is then what happens when they go and they update the FAQ? The vector vault's not updating. It does, it's a one shot. Like all of these RAG solutions are pretty much one shot. They're not getting updated as the content is changing. And that's the difference, right? So we go and we have all this rag in there, we can go have a conversation with it. But the fact of the matter is, is we're out of the moment that we store it into the rag database, we're now stuck in time at that position. And the data has changed, right? It's changed. It just didn't get automatically synced. So anybody who's doing rag stuff, you're going to have to have an incremental, continually updating every time that file changes, the vector vault needs to go get updated as well. And that's the piece where prompt privacy, I do believe it has the moat because they're constantly, their vector vault of your entire data estate is continually being updated. It's not just a static thing. But any LLM or anything LLM allows you to get something close, but again, it's still that one shot. And so you can go have conversations, like what you were doing with trying to get Chatsubitea to write your Rust code. So you go and you update your code and you output it blah, blah, blah, blah, then you go ask it again, well, all of the past files are the old files. So the code's no longer up to date. And so that I think is the biggest problem with RAG is that people are gonna, you have to get this to where when I change the file, the vectors have to be updated as well.
• What's the frequency of update for privacy?
• Every 15 minutes.
• So it's going out and retrieving updates. But there's not any kind of push from the other end, right?
• Well, so it kind of depends on like where the data source is.
• Right.
• Right, so like if it's a database, well then the database, you know, you might have to hit every, you know, you might set that up to only hit it every once an hour, or whatever. You know, Google, you might have to be every 15 minutes. So it really kind of depends on the data source that you would kind of do that pooling.
• Okay.
• But that is, that seems to be the biggest thing where, and we see it all the time, like people go and they set up their chat stuff, they go and they add a bunch of files, they're like, "Oh, that's cool." But data is not static, right? the data is constantly being modified, so we need to be able to have chats with it. And so I think that that's where RAG pretty much gets it wrong.
• I think that there's a normalization of RAG in terms of going on, in terms of people are really understanding the flow of taking your structured unstructured data and vectorizing it, and then the concept of bringing in an LLM to use it to actually figure out how to make it, you know, make a human to be able to ask it questions. And I just think that there's also a little bit of acceptance of the little leaky leaky that's going on where it is kind of going over to that LLL.
• We're here.
• Right?
• It's just an accepted risk because, you know, because either it's, you know, it's just obfuscated enough or it's de-sensitized. but I think that you're seeing that that's just gonna be here to stay. Yeah. Yeah. And maybe people are gonna get better. We talked about, Jacob talked about people moving technology back to on-prem. Mm-hmm. Yeah. So if you have a strong IT team, you can do this stuff on your own without it ever going over. Yep. And these models are gonna get pretty ubiquitous. So maybe it's only a year where you don't have to do that stuff anymore. Right. you pretty much can guarantee that a lot of people, the way that they interact with documents, whether it's a spreadsheet, whether it's a PDF, a research paper, the way that people interact with these things is not going to be reading it anymore. It's going to be uploading it and asking questions. And you're having it summarize it for me. Well, it does raise a question, though, in regulated industries, am I always going to have to be concerned about leaky models? If I'm always concerned about leaky models, am I ever going to get comfortable enough with LLMs that I'm going to make investments there. I think most regulated industries are already well down the path of just on-prem solutions. We are certainly a lot of our solutions. Yeah, but yeah, I mean, yeah. And I think the other issue with especially in regulated where you have to be accurate, back to you. The document won't we just put up a line?
• One of the, another common issue with RAG is I have version one of my documents, so I did some research on some people, and six months later, or two years later, whatever I do, more research has changed. So guidance has changed.
• That's cool too.
• So you're gonna upload the new document, and now you've got an issue where, and there are lots of different solutions already out there in the market that are trying to tackle this, like, you know, waiting this one more, or, you know, reindex, it's expensive, but reindex the entire thing after you remove the old one. but that's gonna be a problem, right?
• Those are Aaron Shaver problems. To me, I just wanna upload this random document, talk to it for five minutes and never come back to the end of the video. I don't need everything that I've ever created to be in this thing. I just need this thing that I'm working on today. And maybe I'm being--
• Yeah, so it's a different use case for me.
• That's you as a user having a very specific use case, right? And I think that where we get to the autonomous agents that those autonomous agents have to have the latest real data.
• Right, well, and I think there has to be some way of versioning, there has to be a versioning algorithm that somehow knows, not only is this the latest, but let's say for example, I go to chat at GPT or any kind of search and say, "I need the previous version, the version that took place on July 1st, because now I've screwed something up and particularly in regulated industries where there was a mistake made, and now I've gotta go back in time that a version that was already authorized and approved.
• Yeah, these Autonomous AIGents are kind of like the Karen in the office that just, won't, just butts into every meaning.
• Excuse me, turn on the recorder.
• I remember missing the data here. It's like, come on, you just butt out. Okay, so wait.
• This week, this week we went through breaking down that feature spec for the very specific for prompt versioning. And it's not just you or even your prompt versioning, It's what version of the model are you using? What version of the prompt are you using? What grounded data are you ultimately using? And at any point we should be able to say, "Hey, when we executed this autonomous work execution, "it was using this prompt at version two, "and it was using this model at version 1.3 or whatever." But you have to be able to at any point with these heavily regulated industries.
• Well, or any industry, let's say for example, I changed the prompt and now I put a conjunction in there instead of using or I use a hand. And now all of a sudden I get a completely different response. Now I've got a different response, a different prompt. I've got different types of content, different kinds, different kinds of primary research I need to do to confirm that this is accurate. There's just a number of things I think that are gonna be impacted by first.
• 100%. - So anyway.
• Yeah, 100%.
• Yeah, I'm still embarrassed at some level on the omnipotent enterprise AI solution.
• Right, and I think that for, I think bringing it back to marketing, I think that if you're in that mid-tier execution role, like using those one-off tools, like within your company's guidelines and policies, is gonna just help you get stuff done way faster. Sure, your images might look the same and have misspellings in you. (laughing)
• We're like terrible.
• You know what I mean?
• No, so you know, honestly, one thing, So I think people sleep on the GPTs, especially with creating GPTs. So here's my tip for you guys. If you wanna use Dolly to generate images for you, you're gonna go create a new GPT, and you're gonna basically be like, here's what we're gonna do. You're gonna be an image generator for, we'll use prompt privacy just because I've got one specifically for prompt privacy. That is, I want it to be 1960s style animation with these four or five different hex colors or our primary colors, blah, blah, blah, blah, blah, blah. You're always gonna have it be horizontal unless the users ask otherwise. And so now I can literally go in and just put the title of a blog post into this GPT and it's gonna generate me an image that will be very consistent for our brand, right? And people sleep on that. And I use that thing every client I've got, I've got an image generator for that client.
• So I'm not using GPTs, but I will say people are sleeping on using Dolly for UI design. >> Oh, yeah. >> That's what I would say is the most important feature for Dolly. And again, it's because we don't think about the overall user experience when we think about AI, but AI has the ability to be able to give it strict controls on what you want to see in the user experience. And now all of a sudden, you've got really good workflows. >> Right. >> Are you using it a lot for UI stuff? >> I use it for inspiration. >> Yeah. >> And I use it for like, you know what I'm using it for. Captain MBA. We have our MBA analytics website that we run. And what's the website? Craftednba.com. Okay. So we've got about 75,000 monthly visitors to this site that's grown over time. Are you advertising on it? We have some advertisements, yeah. And we're looking for sponsors and... Drop it! ...investors. Yeah. Wait, so you... Hold on. Let's delve into this just a hair more. Sure. You have 70,000 a month.

75. 5,000 unique visitors every month.

• Yes.
• Coming to this website at CraftedMBA.com. And the website is basically a--
• It's curated NBA analytics from our team, which is Jacob, really Dan, Kavanaugh, and I.
• Is there any fever stuff in there?
• Not yet, but we're working on it.
• Are you?
• Yeah, yeah, yeah. So basically the problem that Dan had, who's actually my first cousin, he was my first employee back in 2013, their moms or sisters. He always had this issue where people would kind of spout off about the NBA and like certain players. And they would have an opinion about somebody and he as an NBA expert would be like, "You don't have any idea what you're talking about. "You don't really know basketball. "Like you don't understand, you know, "like the metrics that, "like most people talk about someone's offense "but they're a terrible defender." And so he sought out to like, and they're in these group chats on GroupMe when they're talking, they have the NBA group chat with each other and they're sitting there going, having all these arguments. And he's like, well screw you guys, I'm gonna bring data to this conversation. We started putting spreadsheets together and putting data, 'cause he's just like his dad, has to be right, right? He has to, he can't lose an argument, right? And he will never let it go, right? Well, that's his dad, he'll let it go. But anyways, he's like, I'm starting to accumulate this data. And it's not team stats, it's not box scores. It's like your reliability on defense, right? your ability to shoot, you know, corner three. Like really specific curated stats. And so we built Crafted MBA and it started to get organic traffic through SEO. But built by a, as a side project, the UI design has always been something that bothered me because I'm like, it could always be better. So I started using chat TPT 'cause it's a retro vibe. And so I'm like, hey, here's what I want you to do. designed me a retro theme for this site. And Dolly and ChetCPT do a really good job at proposing different solutions and things that you would never kind of do on your own. But the other thing is that Dan figured out a way to take those images and turn them into the vectors that he can use. Dan. Dan Cavan, those four of them are crafted MBA. I don't know what the heck, what was he using by the way? Do you remember? I don't know. So he was taking the mockups. He was taking the mockups, the PNGs. The PNGs and he was able to get high quality vectors out of-- Was it maybe a recrafted.com or was it-- I don't remember. I need to follow up on that. I'll post something in that in a minute. Yeah, because that, again, that's one thing too that people sleep on a lot is AIs are really good at obviously generating PNGs, but you want them to be vector. You want them to be SVG so you can scale them, You can edit them, you can modify them, recrafted, I think it's recrafted.com, is one that we've talked about over the last year or whatever. That's actually a pretty impressive tool to be able to take those in general. But then there's a few other ones that are out there that allow take your PNGs and generate a vector. And those are actually pretty cool too, and they're all using AI, but you're gonna pay like 15 bucks or whatever it is for.
• Yeah, but I think that if you're not using that is like almost like a idea, like mood board kind of palette.
• It's good for that, really good bright.
• Yeah, right.
• So a use case I heard the other day that was really good was because AI is not terribly good at being creative. I mean, image stuff is good at like ideation and like get you some ideas, but like if you wanted to create a marketing strategy that was really innovative and different, it's not gonna do that because it doesn't know anything different. It only knows what you taught it. But what people do is they say, they give me a marketing plan, strategy or whatever, then they use those ideas as, okay, these are overused ideas. This is what everybody's using. So I'm going to throw those away as like probably not the path, if we're trying to be truly, you know, out there on this campaign or whatever. And I was like, oh, that's a good idea. 'Cause just do the inverse, right? Like if it's bad, if it's all--
• It's really good. - Yeah, yeah.
• And you can probably even just give it to him, be like, here's, feed it right back to itself.
• That's what I was thinking about what you would do. like here's the five things we shouldn't do. Now give me some other ones to go through and do.
• Yeah.
• Have you guys had any experience with pattern 89? It kind of grew up here in Indianapolis by high alpha. Are you familiar with what it does?
• Well, so I think you and I were on a call with them years ago. I remember sitting at the speak easy and we were on a call with them. They just switched to, 'cause they were whatever their brand name was before to pattern 89.
• And then they were acquired.
• And then they got bought by Adobe, right?
• Shuttersock.
• Shuttersock.
• But does Adobe own shuttersock? They might. They might have. But yeah, I showed her stock that bottom. But no, so tell us a little bit more about what they were doing. So Petter and E9 originally had, let's say they had 6,000 different customers or whatever that number was. And each of those customers are advertising, spending, you know, millions of dollars online, Facebook, Instagram, LinkedIn, wherever it might be, social. It takes all of the images that have ever been generated by all those clients and creates It's a database of all of the elements in each of those images, and then it identifies how each of those perform, each of those images perform with specific audiences. Wow. So eye color, glasses, dress, shorts, socks, shoes, color shoe, shoe laces, type of shoe, brand to shoe. And then comes back and says, based on our evaluation and analysis, these are the types of images you need to be putting up on social in order to get better engagement. And they're using AI. They used to be on as quantified.
• I want to fight.
• I want to fight.
• Yeah, yes, yes, yes.
• But dude, this was literally like--
• I knew that too, 'cause it ended up on Crunchbase.
• Yeah, but this was, I mean, this had to be seven years, eight years ago.
• They were founded in 2016.
• They were well-headed.
• Way ahead.
• Well, this goes back to Caldwell. This stuff's been around for a long time.
• It's not new. Well, I remember at the time, we were working with a client who was trying to break into a market where the competitors were already ahead of them in terms of market share, And they were trying to identify ways that they could be more competitive in their way they advertise in social platforms because that's where they were going to get their consumers. And so they were looking for something new, something different. And I built a relationship with RJ Taylor, who was the, at that time, the CEO. And he said, we've got something you might be, might be impressed by. And he said, Ogilvy uses it for their clients. I'm like, OK, if Ogilvy uses it, there's probably some value or some merit to this, right? So we started looking at it, it was like, oh, shit, this is exciting. Yeah. Yeah. Anyway, I wonder what Shutterstock did with it. Same. I haven't. I haven't. Yeah. Yeah. You most likely, whatever every startup that gets acquired, it just turns into some poor bastard corner and everybody who's still there is like, man, you're sure never, never, never underestimate the, the, the difficulty in integrating. Oh no. Cultures companies technology. No, it's so funny. You get acquired.
• Yeah, you end up getting bought, and then literally, like, if you can make it for the three years that you're supposed to vest, and then you, but most of them can't, literally, like, I can't tell you how many people I know that were like, fuck this, I'm out! And they're bolting, and they're dropping like $50 million, or like, I do not care, I don't want it, I'm not here to be involved in this corporate, you know, rig 'em a role that's going on, I'm just like, damn. And I couldn't do it either, like, I guarantee you, if I was ever in that situation.
• No, you're day two, I'm like, I'm outta here!
• You didn't have to tell our audience that way.
• They know.
• Why we should have stock, I mean, that tool right now is probably creating stock images.
• Oh yeah, for sure.
• Well, no, I mean, think about how much, like, Dolly is like, how much of these different models like, wait, what should our stock has?
• Oh, Adobe's the same way.
• Oh yeah, yeah.
• You should see the percentage of AI generated images now that are part of their stock photos.
• Right. - That is insane. But what I liked about what Adobe did is so Adobe, and I'm not an Adobe fan. So again, I was a Photoshop like 2.0 guy, 'cause 1.0 really didn't exist. So I've been at, yeah, I've been there. But then they started kind of doing their, moving to the cloud and wanting to charge you while this stuff, so I ended up leaving them. But I will say what they're doing with their Firefly models is they're basically saying if you get sued, we're going to bat for you.
• Exactly. in your ass because everything that we have trained our models on is stuff that we actually own, unlike OpenAI, which is just raw dung in the entire internet.
• I got the same time, aren't they? Aren't they making you, if you use Photoshop, sign up the fact that they own your shit?
• Oh, I know, yeah, yeah. Well, that's it, that's this, Ms. Ejubloy. (laughing)
• That's just a little, no.
• That's just a little thing in the bottom of the contract.
• We won't use this.
• No, we just want you to know.
• We will definitely train our models with every image that you create.
• What?
• Oh, maybe if we wanted to, we could come step on your throat and take all your business.
• But they have user-generated content that users must be signing off on for some kind of royalty. 'Cause I've noticed that they're identifying that the users who created those images in Adobe's stock photos.
• Oh, that's generous.
• And you think they're feeding it back up into the repository?
• Well, they definitely are because I'm seeing those photos. I don't know if they're doing that based on somebody who's doing that in Photoshop or somebody submitting these for royalties, but yeah, that's interesting. Like, would they be hypothetically sitting there, like constantly analyzing every photo that you're kind of working on? And be like, "Ooh, that's a good one." And be like, "Pibble, pop up, hey, are you interested in making some money off this image?" Click, you know, and like pull it in and all of a sudden, next thing you know.
• They're doing attribution, they're giving credit to the food designers.
• Yeah, so I do, so I'll give Adobe props on that. I won't give them props on anything else 'cause I think their business practices are dog shit at this point. But the fact that they're willing to stay and say, "Yep, we'll go to bat for you." And I think all these other providers are gonna have to ultimately do the same. Like chat GBT couldn't do it, right? Like if you've watched their CTO, try to explain where the training for Sora came from. And she's like, they're like, "Did he come from YouTube?" She's like, "Well, if it's on the public internet, maybe." And it was just like the most awkward piece.
• Sometimes I feel like opening eyes like, we don't know how this shit works. (all laughing)
• What do you mean, you think it is? We figured out this thing, and then it just kind of started running itself.
• Kind of like the web.
• Yeah.
• And what I think's happening with Adobe, and you're gonna start to see this Microsoft, and probably all of your enterprise companies, if I'm servicing enterprise clients, I'm gonna have to provide some kind of legal backdrop for them to be able to protect their property.
• Yeah.
• I just think that's gonna be a standard approach.
• Right, no, I agree.
• Yeah, well as the market size increases, and they can command more, you know, they can just 20% pay the lawyers.
• They're gonna win.
• Yeah, you can win.
• So before we go, did I talk about anything LLM already? Did I already talk about that?
• Yeah, you did.
• We're gonna get a blind.
• Jump back in.
• Jump back in.
• Here we go.
• Well, hey, if you're a marketer, I think learn the tools, learn the technology, but get your data ready.
• Get your data ready.
• Yeah, stand up.
• Get your data ready. Evaluate your current usage of your platforms. and really look inward because if you're just still on the surface level or you don't even log in, you don't even know what you're doing with this stuff or you're not using AI to augment even your day-to-day workflow, ask your boss, figure out how you can do it within company practices.
• Yeah, for sure.
• So.
• I mean, I'm one tool I've used and it's been really good for idea generations called 2gen.ai. What was it?
• 2gen.
• How do you spell that?
• T-U-G-A-N dot AI.
• T-U-G-A-N dot AI.
• And never even heard it.
• Well, if you go out and looked on LinkedIn a number of people are just up front. We're using content that's been generated, ideas that have been generated through toogan.ai. It helps you with your content creation workflow. And it basically tailors it to the algorithm of the social platform. And so it identifies that this is a professional audience. They're looking at more educational content, whatever that might be. Then you can both feed it a topic and then keywords. And then it will generate a list of topics. And then you choose from the topic five ways to increase lead generation as a marketing organization. something like that, but it's very specific. And then many times it even provides primary sources for--
• So, I might do that and I might be like, I want you to generate the most offensive LinkedIn posts that you can possibly come up with. And I'll just go for a week.
• I can't just see the most offensive things on LinkedIn to see what happens. Like that would be hot funny. 'Cause like I can't get canceled.
• Well.
• I get more.
• Just don't let people kind of ploy anywhere.
• Just don't let me.
• Yeah, I'm gonna tag Marty.
• This is a single one of my money.
• I did not recommend hard.
• This is what it came off with.
• I did not get brain in the ass. It just came out real quick.
• Has anyone, as we try to end this podcast, has anyone tried, so Rory's, going back to Rory's idea from podcast one season two, was getting a AI to provide you feedback on your product.
• Yeah.
• Hard implementation, 'cause I asked to use it, but is anyone tried to just, instead of say, "Hey Chatshape, give me the stuff." Hey Chatshape, T, you're this audience.
• Yes, yes, yes.
• That was one of them.
• I've done that.
• Yeah, yes.
• So, well Matt, Matt, sorry, I'll let you go.
• Go jump in.
• Yeah, so Matt McBath, who I went and--
• Oh yeah, that guy.
• Right? So he's kind of built a very similar thing to that, where it's the model. Now the other one though is Corinna. So Corinna and I have worked together on and off for like 25 years.
• I love from a Corinna.
• Yeah, so-- Yeah, you worked for China, right? She's a bot. No, no, she's a person. She might be a bot. She's like, she's so damn smart. She might be a bot. But Corinna works at Farm Bureau now, and so that's where her and I interact the most. But what she did, she actually set up agents that act like Jarvis, who's one of our bosses, who's Jolie, who's one of our bosses, and she can go and present an idea to them, and they're gonna come back and they're gonna reply how they tend to reply. You know, Jarvis is gonna come from a very heavy architecture standpoint. Jolie's gonna come from a business standpoint. And then it basically allows her to kind of practice how she's gonna pitch ideas to the bosses. I'm like, that's a fucking genius idea. Yes, that's a great idea. Yes, she's been doing that there and it's been working. I just use personas basically to help it better understand my writing style. I'm a 60 year old marketer working at the intersection of marketing and technology. I've got a little bit of a rogue cowboy approach to my technology. This is the way I talk. Sometimes I'm a big fan of rap. That kind of persona. And then it will generate content that kind of speaks in much more of an original voice, which is great.
• I use it for-
• That's a good idea. I haven't done that.
• My private LLMs, I can put them into all sorts of different scenarios for anything that I would potentially want to chat with. (laughing) So then most, the most requested, or the most feended, requested, demanded for product that has the least amount of competition is AI girlfriends or boyfriends.
• Yep, yep, and I've got it down. And if I could, you know, we could probably use Grock, and we could probably use Open Hermes to basically build this thing out.
• Open herpes?
• Open herpes? (laughing) That's what it totally needs to be called. That model, no, no, no, that model is open herpes. And it is, yeah, the digital herpes.
• I have a little bar rule. (laughing)
• We did not have a cure.
• That is so good.
• Whoa. (laughing) If you're a marketer, you--
• That's not so good, baby.
• And here we go.
• If you're a marketer, you skipped out of here after 25 minutes anyways, and went to the bar. So, we'll see you guys next week.
• Bye, guys. [laughs]