This week on Stacking Growth, Matt Chinello hosted Carl Yay for our December B2B marketing roundtable. They spent an hour on exciting new ways to add AI into your workflow, waiting through all the AI hype that's been crowding your feeds. They cover AI efficiency versus opportunity in modern B2B organizations, specific use cases for different tools, and how to automate reporting and analysis. Hope you all enjoy.
This is Stacking Growth. I'm excited to walk through this because we got some really practical, I think, examples for what you can do with AI, and we're going to walk through really the gamut of JTBT, Gemini, CoPilot, and Clos. We're going to have a few examples across each because I know depending on the business that you work with, you may have to do a certain LLM. Then the other thing also is you may be trying to do things with a certain LLM, not getting the result you want, but you could probably get the output more reliably with a different LLM.
Carl's got a super interesting example around that. All right, man. People will trickle in here this normally happens, so I want to go ahead and kick off because we're going to get into doing a lot of screen sharing. So just welcome you all to our lab event.
We're going to go ahead and run through. If you guys have questions, by the way, at any point in time, just put your hand up and stop us and say, I want to double tap on that. Also, super interested as we run through this. If you have examples that are similar, it's going to what you're getting into here.
Raise your hand. Speak up. Tax staff. We'd love to bring you on to talk through those because with me today, I have Carl Gay, who is the co-founder of Zero to 60 AI.
And Carl basically does nothing but implement AI solutions for companies, and it is not just marketing. It's all kinds of AI solutions. And Carl's been working with us on trying to implement AI solutions as well, walking us through frameworks and just ways to use these LMs properly in order to get the outfits that you want. And so I'm excited to bring him on to let him show you the kind of stuff he's able to do.
And he has done for companies, not just us, but others, related to marketing, operations, reconciliation, all kinds of things like that. So Carl, I want to give you a couple minutes to talk through you and your company, and what you guys are doing. And then we can kick off a little bit with our salvo and then get into some practical examples here for people. Yeah, yeah, just really briefly because I know whenever I attend webinars, it's like, get to the point.
So I can't get to that. So I run a company with my business partner, Zero to 60 AI. Our goal is really to get companies tangible benefits out of AI. So we want to cut the hype out.
What are the tangible benefits today? We work, like Matt said, we work across the entire gamut, accounting, finance, operations, HR, administration, yes marketing, yes sales, but pretty much operations, supply chain, very different elements. And hopefully today while I did kind of tailor this a little bit towards marketing sales, some of the techniques that we have, some of the things that we put in, I use across the board too, different tools, different. And we do, like I think an hour is probably not enough just because of what we can show, what are the possibilities and yeah, things like that.
So Matt, yeah, where do you want to begin? I think I just want to talk real quick just about the hype around AI versus what people actually do around it. So I think if you follow LinkedIn or you follow mass media, AI is coming for your job, AI is going to take your work away. Everybody is using AI at a very competent level, except for you, that is from my conversations with people.
And even from the conversations I have with companies who are VC backed and going and diving into being AI first, couldn't be further from the truth. Like so much of the AI kind of hype cycle around LinkedIn and other watering holes is coming from founder and board mandates to become AI first without products or teams necessarily being fully fluent or equipped with them. So that is created this huge industry of consultants and companies who are trying to who are maybe farther along than others trying to get them up to speed. But just because you're not feeling like you're using AI at a high level, doesn't mean that the peers around you who are acting like they do are either anything.
Another hard truth is that very few people actually are paying for AI, aside from maybe having a chat GPT or a cloud license. I think a lot of people have Gemini and don't even know they have it because it's part of your Google workspace. If you have Google workspace. So I think that a lot of things that you hear and see, not necessarily the truth when you kind of get them in the close conversations about it.
And I think that's a lot of people are saying that, but that's definitely the things that I'm noticing as well. Carl, I want you to walk through a little bit of what you talked through with us when you were going through just getting people start up in AI, understanding the different levels of the models. And then we'll get into our examples. And we have a few different examples across a few different industry that I think will be instructive for people.
Yeah. I'm just going to share my screen here. There's a lot of stuff obviously that we go through with our clients. I think the really one of the big points that we want to emphasize too is that in the space right now, there's a lot of focus on AI for efficiency.
So you hear a lot of, hey, we want to be more efficient. We want to be more productive. And yes, you can do that. But if you think about it, the processes and the workflows that your company is looking to integrate AI into or bolt on AI into at the core, it's still designed for humans, for people.
Right. So yes, you can get your efficiency, but that shouldn't be the do this to replace this. But the whole workflow and the process is still human designed. And honestly, the question I ask is, are we just building mechanical horses, or are we actually doing true transformations?
Are we actually going towards a vehicle? Right. Because what I've started to see a lot more of is, yes, while you're adding AI and making yourself efficient, more productive, if you look at the whole, they're just automating in efficiency. So yeah, that's fantastic.
But I want companies and people and employees and teams and departments to think of, once you get through the AI for efficiency, you also have to think about AI for opportunity. So the key message is, yes, do those efficient tasks? Yes, try to be more productive, use AI. But look at a bigger picture.
As some of the questions you should ask is that process or workflow you're automating. If you made it AI centric, how would you make it AI centric? Or is it even needed? Right.
To me, how would you rebuild these processes and refos to be AI first, rather than just bolt on AI solutions to it? So that's one of the key things. And the slide says value comes from reimagining that work. And also, I think, too, it's very short-sighted for companies to be or the thought or the hype around.
It's like, oh, companies are going to lay off per people or you're not going to get your, we're going to reduce head count because the tasks we've replaced in the AI. Yeah, but you're still looking at it from AI from efficiency perspective. If you thought about it from an AI opportunity perspective, every single one on this, on this webinar, on this roundtable, listening or in the future, on demand, you probably can think of 60% of your work as being repetitive, mundane, but we just have to do it. Imagine what would happen if you got rid of that, say 40% of that.
What else could you do? How else could you think about your role? What are the things that you would get to that you normally wouldn't get to? And that's the AI for opportunity.
That's like, enables companies to grow in a manner that they wouldn't have thought of because a lot of the companies we work with are bogged down at all the stuff. So, but if you're just automating the same stuff, we're all just moving towards building mechanical horses. So that's like, I wanted to preface everything we did today on that. Cool.
So let's start walking through a couple of examples, I think. I want to get in the meat of it as soon as we can here. Yeah. I know we have a couple of things.
We can always go back to a couple of these things just to kind of pause for a second. I know, for instance, for you, there was a couple of things around doing this. Yeah, I was very interested in this part here, which was utilizing Claude within, I think this is utilizing Claude within Excel, right? Yes.
Yes. This is a super interesting use case for me because you would think you'd use Cupilot for this. Yeah. So for example, here, right?
So you have, I got a GTM model from one of our clients. I changed the data around, changed a couple of things around, but the framework's still here. There's multiple tabs. It's a more complex type, essentially spreadsheet.
And this is specifically in Excel. So one of the things you can do, which I love Claude for Excel, this is a plugin, right? Yeah. And to get this, you actually have to pay the max plan, which is unfortunate.
So in Canadian dollars, $140 a month, but I recommend give this a shot for one month, and I'll show you why. So here's an example created in a new tab called P and L summary, Consolidated and Sales Cost, Margins Operating Margin, Implied Break, even a month. Every line should reference its source tab directly. So you hit that.
Now, what's really neat about this is because Claude is tuned for Excel, it just gets right to work and it literally understands what you're looking to do. And it's very hard. Now, I've done this in two capacities. One is with CoPilot.
I did here first. It says, I can add this. I couldn't find the workbook, which is insane because the workbook's literally right here. And then it gives you what you could do to build it, but I have to attach the file, which makes no sense versus Claude.
It's working right in here and it's creating a P and L sheet and so on. The only thing is I have this little thing here. I have to move it over. And so it's now creating a P and L summary tab.
And it's calculating all the things as you saw there. It's now calculating it and taking it from the various tabs. This is truly the ability to work in Excel for those companies that do. This is what to me AI should be helping you with in your calculations, doing analysis and building this out for you.
CoPilot can't do that. I did run it for those who are on Google, Google Sheets. Same thing, same query. And I want live demo.
I want to see what Gemini does. And I don't know if it's the new Gemini 3 or its previous version. I'm not sure, but it is referencing. So that's to me, that's an example.
And for those who don't work in spreadsheets a lot, like even Gemini, I'm still learning currently unavailable to create a new tab and consolidate, which is I was like, I thought you would be able to do that. So we've got CoPilot not being able to do it. Gemini not being able to do it. But damn, like Claude for Excel is fantastic.
And just it keeps working and working and working through, right? So that's just one example. I think even though you pay the max Claude plan, I highly recommend just pay for it a month. Since using this, I don't think I can ever go back.
Because what it can do in Excel and a lot of our clients are working in Excel, we do workshops with finance teams, accounting teams in Excel. And the moment we've started showing them this, it's like this is the massive time saver, right? So just one simple example to start us off. Yeah, this is super interesting to me because remember like conventional thinking, we think, well, I should use the LOM that ties into the same product suite, right?
You're talking about using Gemini for Google Sheets makes perfect sense, right? Or using copilot for Excel makes perfect sense, Microsoft products, Google products. What's so interesting is like, and Claude is really kind of a pretty unique LOM all on to itself when I've used it. And you know, the coding capabilities on it are really incredible.
I'd say compared to chat GPT actually like it more. Just has a couple shortcomings that I can't overcome from our workload purposes. But this goes to the point where like, if you can't do it with like one model, it doesn't mean you can't have it execute with a different LOM provider. And so it's always worth trying and testing that yourself.
I mean, if you can look at what this is doing right now, it's literally building you a tab, building you out of summary, and integrating all the different changes. It's basically building a new SKU into the P&L, which is just kind of phenomenal to kind of see this thing that would take you hours to draft. And the point is not that it's maybe 100% correct, but it gets you 90% of the way there. And then you get to go validated across the other numbers that you see.
But super interesting use case for me overall, especially just given like, what this would take you to do manually just when you already built most of it out. Well, one thing too. So we'll let that go. Now, on the other side, you can do for me, I think like the Cloud Excel plugin allows you to do the internal work in there.
But it can't really do the content type work. So for example, once all those, you finished your calculations, right? You finished all that analysis, you built your sheet. This is where you take it into something like Cloud, create a one page.
We've all done this, create a one page summary word, cool. But then you can kind of start doing these things, right? You can create a Monte Carlo simulation where you're putting different outcomes to it and applying that so you can create that. But then you can start creating PowerPoint presentations for whatever a pitch deck or whatever you have probably reporting, right?
So you can build your presentations right in. But then with Cloud, you can go even further because then you can build like actual applications. So for example, here I asked it to build an interactive dashboard, right? Where I can start seeing all like, you know, showcasing the trajectories, the pipeline funnels and being able to make it like really bring that essentially present here, this thing to life, right?
So this is, to me, you can use both. So you have the analysis and building and doing all the data work here and then taking it back to Cloud to create that content from it. So that's like really an example of like integrating your work with the large language model in the spreadsheet and using the LOM, how you normally would use it on a web app. Thanks.
Yeah, super interesting, really powerful. And again, surprising that you would use a non-native product within that network. But I think I just said it was how good some of these LOMs are at connecting in with your work suite. You had another example that's a little bit more kind of, this is strategy, but you have another one that's a little bit more heavy online, just looking at your brain visibility and doing some more simple checks.
I think this is also a really interesting case. AIO and GEO is a big topic. A lot of it is people just figuring it out on the fly, quite frankly, there's a couple of new tools out there. But this is a decent way for you to kind of utilize this if you're not an SEO expert or a burgeoning AIO expert to give your executives who are doubtlessly asking you about what our AIO visibility looks like, just a quick synopsis or snapshot of what that look is.
So we built this for an investment company. They have about 50, I think 50 to 60 portfolio companies in their portfolio, right? So they wanted to make a really base simple tool, and we used NADN automation for this, where a couple of things, you probably have an executive who has asked, hey, I searched our brand on chat, GPT or Gemini or cloud, and we're not coming up in whatever order that they believe that you should have. So we built this more as a brand check.
So for example, I took three of your three, I think three companies who could be on this call were on the call or will get a recording one of the other. Yeah. So Zora, I've got degree and then you got IBM, I-B wave, right? And so what this does, and we only put the brand, we put the domain, but essentially we built this, you put the product category, HQ, there's a lot more context you could put in.
So essentially what we did is we're using the API, and that's important because in all these models, when you use like a Gemini or a chat GPT or cloud, because the memory is getting so much better, it's remembering all your previous prompts, your searches, your custom instructions, all that is being considered into when you're doing your search action too. And so if you're searching for a brand or a question or whatnot, that per- it's super hyper personalized. So what we're trying to do here is using the API, it doesn't have any of that, like essentially personalization. So as close as we're going to get to incoming mode as possible.
And so we built this, so chat GPT, grok, perplexity, Gemini, clog, the major ones they wanted to look at is what, how do these models, and we're using open router for those who want to get technical, we're trying to get the latest models, so GPT5, grok4, perplexity, sonar, pro, Gemini, three, pro, clog, I think opus 4.5, it was on it 4.5, but we're trying to see what does the large language model in this case, see your brand. And so we look at it, everything from an overview, it's elevator pitch, your ICP, differentiators, pain points of reasons to choose outcomes. So what essentially we're trying to do here is just showcase for those brands, what, how do are they seen from these models, and what is the messaging getting through, like what are the perceptions and things that they can change, and we recommended each brand, hey, probably run it five times, you get trends because you obviously can't just take one output, but also run your competitors through and see what they see compared to what you, what against you and see those trends as well. So that's sort of what we built here for them.
There are other tools that you can use that showcase all these models, but we wanted to dial it in, because essentially here what we're looking to do is the next step is to build outputs from this, but we wanted a base essentially automation that allows the brands to see, and then whatever they want to do with it next, that's kind of the next step. Yeah, pretty simple workflow too. You want to go back to that real quick on the end and end workflow and just kind of walk through the steps, because this is very much the workflow diagram, but if you want to walk just kind of high level of review of your review. Yeah, so when you put in your brand here, agreed, so we put in just the two, she kicks off that workflow, and then yeah, it takes it to the models that we choose.
And as you see, we're using open router, but there isn't the GPT search is not in open router when we built this, so we had to actually use the module for this, the GPT module, and then yeah, it does the search it, you know, we have, for example, here a pretty comprehensive like prompt structure is pretty big, how we built it. That's the thing that took us the longest time. And then you know, system instruction, user instruction, it's the same for every single one of these, although looking back, we may want it to change that system prompt tailored to each one, it's kind of learning for next steps. And then yeah, it pulls it all together and make sure it's in the output that we want.
And yeah, what you see here is that's it. And it usually takes about two to five minutes per brand, and it just, you know, however many you go through, put five, six, 10, 12, it just keeps going and going. Yeah, super interesting. Yeah, again, this is a really good, really good way to do research.
That's actually not a super difficult workflow to build in an N and N, if you, I mean, if you've built workflows in HubSpot or anywhere else, like that's a pretty classic structure, like branching structure, it's just kind of understanding the prompts and then making sure you're connected into your tools properly. It's a little bit more complicated, but again, a super, super neat and clean example for how you can use that to do LLM. So just to finish that, the concept itself, the framework was pretty easy. The hardest part here of all the things was to actually make sure the output was consistent.
So you always have the open consistent, just so the brand, the, the portfolio companies could see, oh, right? Like, I, like this is what it is for Gemini. This is what it is for Claude. This is what it is for Chai GPT.
So they can, they can understand that, for sure. Yeah. All right. And then we had another one for Gemini that I also thought was going to go through that.
And then I want to go through that because that's that multimodal. This is more about small route content creation. And this is super interesting, because essentially it's doing a lot of the things that you're seeing that maybe you see on Instagram or on LinkedIn where people are taking flat images and making and making motion graphic representations, but really interesting kind of twist on that. And then I also want to get, I have examples as well of kind of what we're doing on the reporting side for, for kind of taking, taking greens of unstructured data and trying to automate those kinds of findings.
So let me share this with the sound on, share the sound, share that. Okay. So, and I kind of inadvertently ran into this, because I was trying a couple of things. So here's the first example, why not just share this video.
Okay. So I want to say, watch the video, identify OSHA safety violations or economic risk, OSHA, why they are risks, then create me a video on the correct technique to lift. And so before a lot, this was very difficult to do, even with Gemini 2.5, right? Most of the time we just take the transcript, this time there's no transcript, there's just the video.
So what neat thing about this is watch the video analyze and then create a this, bend at the knees, keep your back straight, hold the box close to your body and lift using your legs. Bend at the knee. So now you have the capability to use from analyzing one, and then it creates for you the video that is the solution. And what you can do is you can change from one mode to the next.
So I actually had it create a training poster on how to guide a squat lift. But I'm sure all of you can come up with new and different ways to do this. What we did though is for our client, we were working with a construction company and the test was this. So I had Gemini actually watch quote unquote, but at the time lapse, it's two minutes long of eventually building this, right?
And so the prompt was act as a lean construction expert, track the flow of materials from the delivery truck to the hoisin crane, identify the efficiencies, what are the bottlenecks, propose a solution, create an infographic of the proposed solution. And I thought this was pretty amazing. So that was one and then I had to do it again, but something a little different was this. So I know for the construction teams that we work with, this is something that they would load in their video footage from job sites, and then they can use AI to do the analysis.
And then on top of that, it can create them, infographics, whatever it is to showcase this, or they can take it to another level. But it just opens the door to so many different, yeah, so many different elements now. And actually, which I didn't, let me give you one second, this probably is probably better representation of this. So I want to say, I just want to put these four slides.
It allows you to see, think and act. So it's like you take your unstructured visual data, put it into Gemini 3, but I'm sure all models were able to do this and put it out in the structured business outcome, right? So example would be old, the old way you take a picture, and then the person types it, but now it's like, take a picture, you can automate it or whatever, use claw, use chat GPT, and then workflow triggered, you can have videos reasoning across it, right? That's how I got the idea for the safety violation.
But that can mean anything from here's a very interesting concept. One, you can take, you can literally record a demo, right? We all have gone on demos, a SAS demo, you take that demo, you record it, then you run it through, let's say Gemini, and there's a couple of things you do. One is internally, you can say, hey, how are those salespeople?
How is the demo going? Get analysis on that straightforward. The second thing is, oh, this, the demo itself, what elements of it do we need to improve on? Like, when you're showcasing your product to people, you get AI-tanalyzed.
The third way, which is interesting, is I want to record this demo, and I want to build features exactly like this product. And that is kind of the next, really where I see the disruption is, you can go to any product, essentially, vibe code, the features and capabilities. And I think this is where the mistake would be, it's like, oh, people are just going to copy other products and sell it, like, no, no, no, no, that's not the thing. It's like, they want to copy the product and use it for themselves.
There's no intention to sell it to millions of people, or your TAM or an ICP. It's just, I want to use this for our internal use. And so instead of me paying hundreds or thousands of dollars a month, I'll just build something that we will use. And we all know we're paying for tech stacks that we're only using 20 to 30%.
Why can't we just build the 20 to 30%? A really smart person, everybody on this, on this roundtable, take the demo, record the demo, go through Gemini and be like, okay, that's the product we want. Hey, build that product. And take right into Gemini, build that product, level whatever, build that product, you've got an 80% MVP, finish it off with the developer.
You've got something internal that you can use on that product, which is an interesting concept. Yeah, for sure. Definitely. Especially if you're like, just a strat marketer, if you're ahead of, don't have a lot of time in creative ways.
The construction example to me is really interesting because one thing, one thing construction companies or industrial companies in general really struggle with is that, is it some of the manpower or the skill set really to like build these kinds of things on the fly or in short amounts of time, they need consult, or something like that. So this really allows you to be more self-sufficient just as a, as kind of a single individual contributor, or just someone who maybe doesn't have this technical skill set or you dabble in it, but you need to kind of create off of it. And it really just encourage you to do things that you previously wouldn't have been able to do in the time manner that you can. Yeah, yeah.
So there's a lot, it really does open up the ability for visual spatial things that we do at work, right? And there's a lot of things in marketing and sales that you do at work. So now with Gemini 3 and anything going forward, we have that capability. So one of the, again, back to efficiency versus opportunity.
I'm sure we can think of efficiencies, but now it's an op, like, what are the opportunities now? What are the things we have not even thought of that we can do now with this? I don't know, but I'm sure you, you, in your company, this is my point in those companies that you work with and think of all the mundane work that's causing you not to be able to think of things like this. So if we remove that, now is your opportunity to do that.
And that's kind of like the interplay between efficiency and opportunity. For sure. All right. I got a couple examples.
I'm going to walk through these are things that Carl worked on with us and really, really helped flesh out for us. Like things really what we were trying to do and what we try to do with AI is try to take the mundane work and the tedious work and figure out a way to make it faster and more on demand for ourselves. And like, we're really trying to be about that. And probably there's no area where that kind of manifests itself more than in our, than in our data and our ability to process data.
But I think because AI is probabilistic, it takes a lot of testing in my opinion to get there. So we did a lot of stuff which at GPT, like just on a primer for things we're learning with it is just like where we, where we kind of tend to lean is between these different modes here. You know, that we typically use CSV upload for last resort unless we're doing something like building a custom GPT, which we did have built a couple and we will be building a lot more and I'll walk through a couple of those company knowledge, which kind of lets you do a lot of the automated task scheduling stuff on demand, although it will caution almost every single time that chat GPT runs an update or is preparing to run an update, like they're about to, I believe this week, that one setting that thinking extended thinking mode within within chat GPT tends to get a little squirrely. I've actually been gravitating a lot more towards Gemini to do data analysis lately part of that is because we are a Google company.
And so Gemini as a Google product works has worked really well for us has the comfort of exporting to sheets, makes things like graphs and things like that, easier for us to do and then agent mode, which is an uber, uber powerful tool within chat GPT to do nearly anything would caution on that. It's credit wise, pretty costly. And also if you have pretty strict IT protocols in place, you're likely going to find a vulnerability that you have that you may want to let your IT team know about. One of the things we stress a lot with people, at least with the team with the team internally, is utilizing a spec framework and Carl and 0 to 60 in particular really helped us unlock this.
Like this is really where we started to get good results with our, with our AI workflows. Once we started to really think about and utilize a spec framework, basically across the board with almost anything that we did. And mostly that's because you want to put controls in place with the LLMs. And then also because for some things that are maybe lighter task, a lighter spec or a heavier spec might be more appropriate.
And then just going into the, and then, and this is really what allowed us to get a good result to get better results with our data as we were exporting it from our ELT, getting into an unstructured place and understanding what the limitations of it are, and then what it's capable of doing. Like it's not going to replace having a data lake and an SQL server, which we do not, we do not currently have. But this does kind of get that sort of month, like month over month, quarter over quarter reporting as long as the look back isn't far enough, you can really kind of pull that kind of data for yourself with a prompt like this that's well structured. And then I think the other thing when you're looking at data and you want to try to leverage your data with AI is having testing criteria.
Like I cannot stress enough how important this is to have. And so I, before I, before we rolled this out internally, like, and this was on me to do, I had to make sure that it worked. And it worked pretty reliably. Otherwise, the team would lose trust, clients would lose trust because data would be inaccurate.
And we'd ultimately be kind of beyond the hook for those kinds of things. So this testing criteria, I pulled this straight from what they do in preclinical trials for, for pharmaceuticals, essentially. So this is creating a first thing I did with almost any other structured data that I worked with is create a control version manually to understand how accurate the data analysis should be from the LM. So I pulled it, did all of it myself manually.
And then I would have, you know, you have five, six dozen kind of metrics that you can pull from. And then I would basically pick 10 metrics to consistently test across all the different LM versions. So I would pick two prompts, I have you one in a light one. So I pick frame of one or three and then also two and then write the same prompt.
And so I would utilize prompt one, ask for the analysis and summary off of the infrastructure data. And then I would utilize prompt two. And then I would mark it as a pass fail if it matched, if it matched my control version to 90% accuracy or greater. So I would pick 10 metrics, it had to get nine out of 10, right?
And then if it didn't, it was just instantly marked to fail. And then the other thing you would have to do is test against a bunch of different modes. So if you go to chat, GPT, even right now, you know, just all these different modes that exist that, you know, people aren't even necessarily fully aware of or utilizing it. So you have extended thinking, you have standard thinking, you have agent mode, which should be somewhere.
Where's agent mode? Sorry to chat. See if I can find it. The agent mode just go away.
No way. There it is. We have agent mode. There it is.
Super powerful. And then you also have like company knowledge as well, which is another one. And these things all work a little differently. And you hate a little differently company knowledge.
You have to add sources, you can have Google Drive, you can have HubSpot, you can have whatever it is that you're using as kind of your primary tool and it'll go dig in and utilize this thing. So we had a test against all of those and also gem and I was thinking mode. And then if it passes, plan the rollout, we would retest. I retest constantly, especially when when all of them have run updates or anything like that, if a new model does happen, I kind of retest almost everything.
And then another thing is personalization settings. I'll share this with you all after this event. You do not have to use all of these, but if you really want to control the level of the output that you get, having personalization settings that you utilize within your profile here are super important. You go to personalization here and you can utilize personalization with custom instructions.
This is stuff you would use like here. And to make sure you're getting a pretty consistent output and you're getting the kinds of checks on the prompt that you can. So when it comes to the things that we were doing overall, I would say like the first thing we're going to do is work with our data a little bit better. So one of the first things I wanted to endeavor on was looking at our high-intent funnel analysis in a better manner than what we were currently doing.
We still have the right one here. So this is a keyword analysis. I'm going to find the right one here. No, I have a lot of these open.
No, no, this is it. So we're just using a heavy spec and I'm starting to open the Google sheet. I was using company knowledge and extended thinking mode when I did this particular prompt. But you give it kind of the instruction of where to open it.
You give it a scope of analysis. You give it the way you want it to handle your data and your metrics. You give it your output requirements. You give it corrections and validations.
And then you let it run. And so what's interesting about it is it'll find the right sheet that you wanted to test against. Super effective. And then I'll start to give you the numbers against the channeling systems average.
And every single time I was validating something like this, I would have the manual work that I would do and I would check all of these or pick 10 metrics across my high funnel metrics, across my conversion rates, across my unit economics. And I would check against all of those things. The one thing I would find to be a little bit of a shortcoming with this, especially if you are a marketer who works and understands the dynamics of your company, AI does not know that, is it's going to sometimes give you some actual recommendations that you may not necessarily agree with. And that's because it's only looking at the data that you give it.
It's not. And it's only applying kind of broad knowledge that it has. So these actual recommendations are the things that I would kind of take a beat on and validate. But if you're looking to kind of get really quick runs, look at how your thing is doing against the trailing average, what your delta and things like that are, this is super useful.
And then when this works, you can basically create a schedule for yourself to do this on repeat. So this is just kind of like a high level version of how we're doing KPI analysis kind of across the board and looking to kind of roll this out and make this faster for our team to get these insights and then utilize it for presentations, actions, strategic conversations that we're having with clients really, really useful in that regard. We've caused any questions on that. Thank you.
Yes, Donna. Thank you guys. I do have a zillion tabs. I just wanted to make sure I could jump across where I needed to.
So this is like very high level. What about when you want to get really more tactical with it? Like, let's say you want to look at keyword data. And it's fairly new, we've been doing this for about a month and a half, two months now.
It's been great. It's been a huge time saver. It's been really fast time to insight for us. But it's been, yeah, it's about two months we've been using this kind of full on at the moment.
And we're still rolling it out more. Like I'm canary testing. We canary test this with our clients. So like we have one client, we make sure it works with it.
And then we start to roll it out a little bit more. So it's not just like every single client gets it. I want to make sure it works with the team. One thing I do don't want to discount either is what you can do with an Ingemini.
Like here's Gemini doing the exact same thing. This is the exact same prompt. And Gemini kind of gives me the exact same data. Right?
Here it is. And it's a little bit better for us because I get the export to sheets and it gets to be craft and stuff. But Gemini does just a good job of doing this. It's chat GBT does provide a new connectors in place.
And for us, it's much easier because we are a Google workspace company. Makes it really, makes it really nice and solid. One of the other things like doing something that's really tactical about keyword analysis that happens a lot when you're running Google ads, right? So we have an example like this, when we're running for a client where I want to look at the last three months of keyword analysis, this particular client has custom has offsite conversion tracking and they have pixel based coverage tracking.
So there was a lot of instructions around the conversions here that we had to give them. Yes, all of it's all contained within our workspace. It's not it's it's it's not public. So essentially it's kind of running through through this.
It's kind of giving you like here's the continue funding keywords. Here's the good efficiency keywords. Here's the okay efficiency keywords. And then here's the ones with poor efficiency.
So we just can do three months lookbacks on keyword data really, really fast. And that allows us to to again have more actionable conversations with clients, we kind of get this feedback back really quickly. We can then focus on like, well, what are the ones that are a little bit efficient? I want to focus on landing pages or the closer at the ag group, we can scale spend there.
We can cut ones that are not profitable for us or not producing things that are good enough efficiency. And that in and of itself is very powerful. This is drastically cutting our time down to do analyses like this, things that would normally take quite a bit of time when you're kind of pulling across match types and terms and look back periods, this stuff cuts a lot of time for us. If you look in chat, GBT, it gives a very similar kind of output for us.
So just again, these are the two main models that we utilize, but it doesn't really sell a job. I would say I probably like the Gemini output here a little bit more, but very similar output here overall with the exact same prompt. And then just kind of going into like other kinds of things, we try to do like strategically with clients. So another thing we're trying to do is build more tooling for ourselves.
Also, so we have this one right here, which is like a creative messaging alignment one. And we do this a lot for creative when we're doing copy with clients. And so we kind of built this, we built this tool here ourselves with with China Mills, who's our VP of creative and phenomenal job with it. And so what we do here is we'll essentially pull, we'll pull kind of campaigns, we'll pull the internal, the initial draft, and then we'll pull the final draft.
And then we have a set of stack and instructions here within this. And then we basically ask it to kind of split off, sorry, we have to ask it to split off the tone of voice and the differences overall that we see, and between the two versions to kind of recognize what the client expects versus what we initially deliver. So we can kind of have that as a knowledge base going forward. So if we look at like that here, like we kind of get a key takeaway, like anchor every concept in real user, real environments, leave it clear outcomes, name the product, motion and keep the copy tight, and just recognize some keywords that way.
They always want to keep in just based on the comments that they leave in the final version that we end up creating. And then one of the last things that we did, something that I just recently built, and I'm essentially looking at taking almost our entire onboarding process and essentially AI enabling the entire thing, so things from our vault and those kinds of tools. I just want to make these as like public as market facing and kind of AI friendly as possible. They just invite themselves a lot to these kinds of things.
So one of the other things would be like a, would be like a leak quality analyst, something that I really just recently built, because it was actually an interesting phenomenon for me. Like I had a client who had a month where they needed to get the lead reviewed. It was not the greatest, it was not the best month for them in November. And so I was doing this analysis manually, and immediately made me think why couldn't I do this with AI and get the exact same output.
So it's exactly what I did. This is all scrub data overall. So I did this with the CSV upload, and this is when you make these custom GPTs, they are CSV uploads, although I believe connectors are going into place with these. So I think they'll be an update on that from chat GPT.
So that's the thing is like, when you build certain tools. Yeah. Yes. But the thing is, we don't know, I don't know when those will happen, but it's within the next weeks or months.
And I think the big rumor is tomorrow there's a pretty significant update. I think chat GPT 5.2 was coming out or something like that. Yeah. And you notice the two remember, like, I remember what several weeks ago, could they drop company knowledge?
Oh my goodness. Yes. Just the existential dread I felt when that happened because I was going, so all these, all this entire testing framework that I had, I was almost done with it. And then chat GPT dropped a huge product update, and then none of it worked anymore.
And I was just like, oh my god. So that let me do two things. One is it let me look at Gemini a little bit more closely and see if it was going to be a little bit more easy. And it was, and then the thing it did is it made me have to run through all the settings of chat GPT to figure out how to get it to work again.
And that's why I say like, if there's a new model update, retest and adjust the setting recommendations to get the output that you need because it will change. Like it used to be on company knowledge, you could not pick any of these and for every now, I can't now I cannot pick anything. But normally when you do this now, and you pick up a knowledge, you have to pick one of these things. I'm going to, I'm trying to on choose how spot I cannot.
I can pick Google and on choose how spot and then I can't choose Google. So like now you have to use one of these before you can uncheck them both and actually made it a lot more reliable and locating the file that you would ask it to in your spec. So like when these updates happen, you need to kind of play around with the settings and figure out what changed and how you can get that output again. And that is, Colin, I've talked about this when we've had our discussions like, this is just going to be the state of the world for the next several years.
It's almost like if you are a hardcore SEO person and Google runs a core algorithm update, and then everyone's asking you like, what's different in our SEO, that's essentially what it's going to be like to work in AI for everybody. And so if you're not, if that change of pace bothers you, like it's not going to change in the near future, which is really interesting. Because like, I think today they dropped the app, the Adobe Acrobat app. So I was like, Oh, does that mean I don't have to pay for Adobe Acrobat?
And I can just go into chat GPT and edit my PDFs, which I'm actually trying to test right now, because they're like, Oh, there's they have Adobe Acrobat Adobe Express. And what is the other Adobe Adobe Photoshop? So like, what can you do in chat GPT that you can and can't do in the actual application itself? And back to the whole, if I'm only using Adobe 30%, then I'll just go into chat GPT and do all my Photoshop edits there.
And I won't have to pay. It's just interesting, what's happening, like literally, as we speak, and the changes that are just coming pretty quick. Yeah, 100%. All right, we have a few minutes left.
And I want to open it up for questions, because I have to imagine, first off, I got Carl's just an AI native and such an expert in this. I know there are people here who are thinking about doing stuff with AI, playing off AI, needing AI issues, wanting to even talk through it with clients. And I want to open it up for anyone who has any questions for them. I mean, Carl's a pretty expensive resource that I have a month or an hour here with us.
Christopher, I see you have your hand out, man. Let's take him off mute and welcome to the event, man. Thanks very much. My question is regarding sort of protected environments, so things like law firms that are highly regulated or defense and security companies.
If they want to implement AI, you know, I have a lot of people that are very resistant to it, sort of in the cyber security space. What sort of systems can you recommend to use in those sorts of spaces? And what's the easiest sort of or rank for them to not be panicking about it? Yeah, no, no, good question, because there's a lot of, as we're dealing with different type of clients and there's different security and privacy, I think, like, I would call them policies in place.
But I think too, there's perception, there's myths, and then what you can actually do. And I think it's super unique to each and every single company, more so because they each have their own policies, depending on the industry, and also the risk tolerance of their leadership team and the company itself, right? Because, for example, some municipalities or local governments we work with, hey, we need data residency, but we also have international clients that are like, we, it's not even data residency, we need data sovereignty. So it's like two different things like, oh, okay, but then on top of that, it's like, but we want all this stuff.
So it's like, there are some trade offs that you have if you have data sovereignty, data residency, but we want everything that can be done in chat GPT. It's well, like, well, we'll have to bring it back. But also it depends on the data. What kind of data are you using to accomplish whatever task, because that could impact the outcome too.
If your data is, let's say, red data has a lot of private information, then the solution would be a little bit different. If it's a lot of green data, then that gives you a little bit more flexibility. So going back to your question, Chris, is that the ideal way to start with this is do some sort of audit discovery to understand what are those issues, those workflows and processes, because you need to understand that whole piece before just having a one, like, you know, a lot of companies were like, we're going to implement co-pilot and that's it. What's like, you don't even know what are the major processes issues.
Yeah, your frontline staff is dealing with, yet you've already recommended a tool that I would say for 50% of them. This does not help me. This is my use case, right? So those are the things ideally you look at.
And two, we've got to start, stop thinking of AI as focus on the tools, but as systems, because we all know, just from this webinar alone, different tools have different strengths and limitations. And it isn't one tool rules them all. There are so many use cases, like an accounting or finance team would find Claude for Excel amazing, but an operations team wouldn't have any like need for that because they're using something different. So we've got to start thinking about it more as systems.
And then how do we build our system around these capabilities, knowing these tools could switch on us pretty quickly. So essentially, yeah, do an audit, make sure you have your discovery, make sure you lay out your process and workflows and identify, Hey, what solution could potentially work for what for this department, for this team, even for this individual and start treating AI not as finding the best specific tool, but think of it as systems specifically for business use. And it's sort of like, Hey, I have a job, I need a hammer for this job. I have another job, but I need a planer for this job.
So it's like two different things, but you wouldn't take the hammer to the job with a planer doesn't make any sense. So that's how we got to start treating this more as what is the best use in this scenario and be able to pivot really quickly. Right. That's great point.
I mean, I have a little bit to add to this, just because I have a friend of mine who works pretty high level at one of the major cyber security firms does a lot of incident response for large companies. And his advice to me for those kinds of concerns is if you can move it, if companies can withstand a movement on prem or a version of on prem, like that's the best way to lock your stuff down and make sure that you're not risking getting any of your getting of your data kind of stolen or having any vulnerabilities exposed. So like, I kind of think that AI as a reach is critical mass is going to bring a version of on prem back and Vogue a little bit more, especially for certain companies. And I wouldn't be surprised that that becomes more of the norm, especially as you go up to try to sell AI and enterprise.
You do have to do you have challenges with on prem itself, right? Because of things like, do you have for the use cases you want to do? Do you have the GPU's to actually run your things on prem, right? Because like whatever model you're using, yeah, for simple classification, cool.
But if you want to run your own like video generation model using an open source system, you have to have the compute to back it up. So these are the challenges too. And you're on the hook for the maintenance. You are like, can are you can you maintain?
Do you have the staff? Do you have the time? Do you have the resources? Like these are all the things that you have to actually consider as well.
So like, that's the kind of stuff that's a trade off that we're having is because there's so many systems, so many tools, so many things, there's a lot to consider whenever you deploy on premise cloud and all that kind of stuff. It's very interesting. It's super, super all over the place. It's also going to create a lot of nice and industry like there's there's going to be hybrid instances, there's probably going to be services that prop up that basically serve for some of these companies who want to be on prem and they don't maybe want the manpower but a managed service that could do that.
It's going to kind of create its own economy kind of like it is right now. Tracey did a point about cloud and GPT like I Gemini wasn't even on my radar. Honestly, it's all about two months ago. And I've been like, so the Gemini three update was probably one of the biggest and best updates that have come through from an AI model this year.
I mean, it's almost the one that's going to handle the same old man to say that they're in a code red, which at GPT, I mean Google, Google has a lot of the other kind of workspace settings. And the thing with all of these different models is like the quote unquote leader is always going to be this horse race for a long time. It's all going to be based on perception. And it's all going to be based on like how good some of these core updates that they make are.
But the Gemini three update was, as my opinion, been nothing short of stellar. And it has a lot of people on the refined team very excited to use it. I mean, getting our design team because in it comes Vanna, banana and veo and like there's just that all that'll suite a product opens a lot of possibilities up for us. But that's that exactly is why like Gemini three would for marketing teams, content generation teams would be amazing, but may not be as useful to an administration team who spends all their time trying to search emails in outlook and in teams, which I rail on a co pilot on a regular basis.
But for that specific use case, co pilot itself is so good because of the Microsoft graph, right? And hence tool to fit situation not, Hey, everyone's going to get Gemini three and those admin teams be like, that's fantastic. But I need this and this doesn't help me in AI's useless now, right? So it's that kind of stuff we're seeing.
Yeah, I mean, every company at some point in time, they're going to use multiple of these core core AI tools across their entire business. And it's a matter of like which function in department does it fit best in for which output and that is going to increasingly become the norm. I know a lot of people are plugged into chat GPT to do a lot of stuff and things like clay. But when you're talking about just operational output, there's just going to be a suite of those tools that get utilized as a result.
So it's important to back out and zoom out with that perspective as you're wanting to pitch a little bit some of the tools you're looking to do and maybe understand what other teams are looking at as well as you kind of navigate that with your companies. Cool. Man Carl, I really appreciate it your time. Thank you so much for coming on.
Thank everyone here who listened and a couple of some of the questions that dropped in. Donna, I see your question. I don't want to necessarily bring it up here, but I'll ping Carl on it and drop that to you and drop that to you directly. Donna mentioned Nate Jones, who I know Carl is a new normal span of and we definitely put some Nate Jones inspiration for parts of the slides.
So if you don't know, Nate Jones is he is worth definitely worth watching and checking out. He is one of the authorities on AI and is very excessively speaking about it. So I would definitely recommend. All right, guys, this is our last event of the year.
Really appreciate it. We're going to be back next next year. So I don't talk to you all before then. I want to wish everyone here and who maybe listens to the podcast when it goes back out.
Super happy holiday, great respite with your family. Thanks to all the people who have listened and attended throughout the year. We're looking forward to getting going back in 2026. And yeah, man, I just wish you all nothing but success and prosperity going into the end of this year and in the next man appreciate you all so much.
So have a great rest of your day, great rest of your week and I hope you all have a one up holiday, man. Thank you.