Future-Proofing WordPress Agencies and E-Commerce: AI and Private Cloud Revolution

Read the summary article here with all the AI tools mentioned in this presentation.

[00:00:00] Todd Robinson: All right, well looks like we are live here. So as people start to come in today we’ve got Tom from Convesio. So I’m Todd from OpenMetal. And we Tom and I have known each other for, it feels like a long, long time. We’ve been in the same industry for so long, and we were lucky enough here at OpenMetal to have a Convesio as a customer.

And we’ve gotten to learn a lot of things together. And definitely in the AI space. I always say to everybody, people don’t really know what it’s for or how to use it yet. But we’re, we’re hoping to share a bit about what we know. About AI and how it can apply to WordPress agencies and Tom with Convesio who is Convesio is a really a brilliant WordPress company that spans all kinds of different parts of their customers.

Troubles and challenges and need to get into new things like AI and so we’re excited to have Tom on here today to kind of guide us through some of the things that he’s seeing that WordPress agencies may find really useful. So yeah, Tom, if you don’t mind intro yourself real quickly and maybe talk a little bit about what got you into Convesio and working on this WordPress stuff.

[00:01:17] Tom Fanelli: Yeah, sure. Thanks, Todd. Appreciate it. And likewise, super excited to be here. You know, Convesio started gosh, I think we officially formed in 2019 and, went to market in 2020 as a, really a WordPress hosting platform and our superpower was scaling. And so we gravitated towards what I would describe as the mid market to enterprise mission, critical sites.

These are typically commerce enabled sites, membership coaches big publishers, and. Since then, we’ve sort of evolved and we’ve really gone and focused on this commerce vertical, right? So that’s and now we have a payments platform and we’ve got an engagement marketing platform, that does cross-sell /up-sell abandoned cart website personalization.

And our mission has really been to service agencies and these end customers via agencies through, you know, our platform and our technology. And, you know, as a sidebar on this, I’ve been in this space a long time. I was thinking, as you were talking, you know, One of the things that I loved, I think I got, I think I built my first website sometime around 1994-95 probably.

And, you know, back then nobody had done anything. So everything you had to figure out, there was no playbook on like, this is how you build this type of website. Right. There was, and it kind of, this revolution in AI feels like those good old days, right? When it’s like the future is uncharted and that is a really awesome place to be.

And for those who are curious and they’re innovators, there is so much to do right now in terms of putting your mental capacity towards figuring out solutions and how they apply to, in our space. Agencies and Commerce. It’s just a really, really exciting time. So I love that. I love that sort of the place that we’re at where this is so new, but it feels so powerful and capable.

Yeah, I think it’s really exciting. Like I said before, I know we don’t know all the things about this or what it’s going to be used for, but to share a little bit of my, my background and why this is part of my passion. So first on the OpenMetal side, we’re an open source focused infrastructure company.

So we basically take in. what would classically be a very difficult to run private cloud hardware automation system and turned it into something that companies generally on a little bit of the larger side would be able to own their own and run their own private cloud. And it’s a fixed cost system.

Companies find themselves in the situation where they need this when they have reached a certain tipping point, either in the public cloud, or they’re trying to get out of the complexity of co-location or their own data centers. But where where it crosses over with AI. So just a little background.

So the group can see how what we’re going to share. So my background in AI is both from the infrastructure side. So we provide GPUs and CPU based inference using open source models and open source kit. So, for example, Llama 3. 2, the latest, the next one up would be running on a GPU or a . CPU here at OpenMetal, and customers are calling through an OpenAI compatible API to then reach those LLMs.

So that’s some of the familiarity that I bring to the table. But I also work on on the InMotion side. So this is another business that I’m involved with. We utilize it for several things. One of them is for actually running a WordPress builder system. And so this is an AI driven create your blocks, but AI driven create your content also.

And that’s inside of one of our builder systems on the in InMotion side. And so, and then also being engaged in helping customers start to either use our stuff directly from an infrastructure standpoint, we’re just talking about what on earth should this product for us be in the future? Or how can we help customers do this?

And I know, Tom, I’ve seen literally some of this stuff. that work that he’s helping his customers with in that same vein is bringing it actually to a useful spot. So customers can now, whether it’s a chat bot running on their site or it’s systems used to ingest content or logs or information about performance to be able to then tell the end user, hey, this is this is what you need to do.

Or here’s some guidance. So that’s the background. So hopefully we’ll get some questions throughout here. But yeah, Tom, I don’t know if you want to lead us out maybe talking a little bit about what you’ve seen now for agencies. And maybe we could talk about how designers might be using it, developer using it whole, whole site managers might be using it.

But yeah, if you don’t mind, take us out and we’ll go from there. Sure. Yeah. So I, I, the way I think about the opportunity for agencies. And I want to go beyond this idea. There’s a million AI tools that are out there. The challenge with that inherently is all your competitors are going to be using those same AI tools.

So how do we really take it up a notch, right? So this is where I think the real innovation and differentiation is for agencies, and you don’t need to code your own AI application for this. There’s all of these tools out there now that are open source, we’ll talk about some of them. But at a really high level, if you’re an agency and you’re thinking about how can I Implement AI in my business.

I think you need to think about this in terms of a few different buckets, right? So one, you can have things you can look at your internal processes that your team has. So this is things just to throw out some ideas like. How do you deal with errors on sites for your clients, right? How do you handle support tickets for your clients?

One of the big, use cases we have is we do a lot of support with clients. And what I’ve found, we support our clients directly in Slack. I’ve found you can copy and paste like an entire slack thread into ChatGPT and poof, you have a really awesome knowledge base article for other people who are going to, you know, deal with this in the future.

So things like that, there are processes that you can really accelerate your business. management by leveraging AI, but then even more, it’s like you can start getting into creating custom workflows. So we, we have, this new thing we’re working on right now. We’ve built a it’s a simple script. It basically just monitors a WordPress site for errors in the WordPress error log, runs when it detects an error- it does a post into our Flowise automation system and then Flowise stitches together LLMs and we use a tool like most people are familiar with Zapier, but N8N, we’ll drop the link in there, is an open source version of Zapier. So we basically have this chain where we can say. You detected an error, send it to AI and analyze it, then post it to ClickUp as a ticket. Or send it to your help desk as a ticket. And so now all of a sudden, and this is feedback that we got from one of our clients I’ll give a shout out to Valet. They’re an awesome WordPress agency. They’re like, you know, we’re AI for AI sake is not really that great, but like. If we can actually have it tee up for us and serve up to us what we need to be doing to continuously improve our client sites, that’s valuable because we have to have someone go look at a site to figure that out. Go look at the logs and go interpret them. Well, this one workflow can automate that entire process, and it’s not just AI, but it makes it so much better than just throwing a log into a task.

[00:09:30] Todd Robinson: Yeah, that’s pretty cool stuff. All right. Yeah, to be honest I’m now taking some notes here to have some of our support flow teams watch this because it is interesting how everybody’s mind works a little bit differently and sees how this can be used.

You know, I think for me and I’ll, I’ll I’ll cross into the the designer, the designer world. And then maybe we can follow up on that is you can definitely see there’s going to be advantages to knowing how to use AI to help you actually create both the site and its layouts and its blocks, but also its content. And again, a lot of people end up a lot of agencies end up supporting their customers all different kinds of ways, right? It might be you’re building a particular plug in for them. If you’re if you’re on that side of the kind of developer business, you might be doing the graphic work. You might be doing the design system. You might be picking the color schemes, et cetera. But I definitely see this this stuff already coming in. But I see that for, I think a lot of times when customers, you know, very end customers are trying to do this themselves, they’re going after and using these tools. Do you see a crossover where designers will gravitate towards systems that are AI assisted, I guess you could say in creation or yeah.

Help me navigate that or understand that myself. If you’ve got, yeah, some perspective to share there, Tom.

[00:10:59] Tom Fanelli: Yeah. I mean, I guess I’m not, I wouldn’t call myself a designer, but you know, being in the agency space, I’m really skilled, more of a technical skill of like, you know, Photoshop and, and, you know, the tools that we’d use Figma.

And I’ve always found, and I think we’ll throw a link to this. I just found uxpilot.ai, which I’ve always found the hardest part is the starting, like when I have to look at a blank canvas, and I think this is why Canva is so popular. And I think the next evolution of this is Canva. It’s like, choose from a million templates, right, to help you get started. I think that whole thing is probably going to go away and we’re going to be like, choose from a million prompts to get you to get you started. And what I find as the, the, the harder thing that is, this is one of the areas I think is the least progressed is product management and product design.

It’s like, yeah, I can create a great image. I can create good copy, but really AI tools are just scratching the very surface level of product management and product design. And I’m not talking about coding. I’m actually talking about a UI interface for an application, you know, a roadmap creation, things like that.

And UX Pilot is an interesting first pass at some, an app that’s trying to do, you know, this notion of you can put in a prompt for what you want for your app and it’ll lay out your UX screens for you. And to me it feels like in five minutes, I can have a starting point that I am able to go quickly on -that eliminated research and looking at competitors and all of this stuff that used to take a lot of high level sort of product management, intuition and thinking.

It’s like now we can get things going much faster. And I think that’s probably the place we’re going to see, we can start and accelerate projects faster with these design tools that, that are out there for AI.

[00:13:18] Todd Robinson: Yeah, I got it. Okay. Yeah, no, I think maybe I’ve got something to add to that. It’s a little bit different part of the space. So with our, our builder system on the IMH side. We also use this builder for our own sites. And one of the challenges we ran into was, hey, when we want to create new blocks you know, you want speed of delivery, you want the, the marketing department to be able to do. So again, these are tools we use ourselves and have out there for hundreds of thousands of users. And so, and this isn’t released yet, but this is what we’re trying to do. So I think this might be applicable to the design agency is we essentially, when we ask the AI to create us blocks, we’re feeding it a suite of blocks, of course, that we have created already because the AI needs this. You know, this set of content to base itself more accurately from.

So if you’re using a general LLM, it doesn’t matter. OpenAI or whatever’s behind there’s versus Llama or whatever that we’re using is you’re often having supplement the request the prompt with the examples, right? So this is your context window. So you’ve got this big context window.

And so what we’ve been trying to do, of course, is to send in a whole suite of the working blocks from our design system. And so what we’re trying to do, and I think this will be applicable to the design agencies is as, of course, you have this suite of designs to choose from, you naturally are based, you have your base, right?

You have your kit, your WordPress kit that when you launch this out, It has the plugins you like. It has your base theme that you like. Well, and I was thinking about that eventually it should extend. So instead of those million prompts, which of course is gonna happen, which is a bummer, but it’s gonna happen that way.

But then you’ll have your own suite of supplemented content that you get to send along with this to say, make me a new block for this or get me started with this kind of block with this kind of slider activity, et cetera. And, but then you know behind the scenes that when you’re making that request, you’re sending along all the blocks that you put into your approved system. Maybe there’s 20 of them and the LLM can then borrow from that and say, I’m going to use those when I create this block for you. So my guess is it’s going to happen like that. I don’t know. Yeah. Do you have a thought on that?

[00:15:28] Tom Fanelli: Yeah. Well, I, I, I, you know, it’s interesting when you said design, I think to myself, like proper designers working in Figma, what you just described, and I think this is a hot application is implementation of the website. To, to me, when you’re in the builder, you’re probably you’re outside the design phase and you’re more in the implementing stage. And as a side Matt, Matt Mullenweg just did a build off, on, against someone, here’s the thing that blew my mind. He’s going into this and he’s like, you know, I, I work in WordPress, but I don’t really build sites often. And he’s like, they gave them this site. It was the first time they saw it. And they’re like, and he’s like, Oh, I’ve got this AI prompt. And he literally screenshot. a segment of the website dropped it into AI and it produced the Gutenberg pattern code for it. And he pasted that in and poof, the image, it took an image that he screenshotted into ChatGPT created a block for it, pasted the block text in, clicked recover block and voila!. It was perfect. That did text translation, images, everything.

[00:16:45] Todd Robinson: This tech though is, I tell you, like, that’s why I said, I think we started the conversation with like, people don’t know exactly what it’s going to be useful and how quickly these things are coming because they’re building on top of other open source systems. So that scraper, I’m not sure which one, the one that’s taking the visual and pulling it in. I’m sure that’s going to be an open source system that is allowing that to happen. If not, there is, there is a competitor to it.

[00:17:11] Tom Fanelli: Yeah, this was just a ChatGPT, a custom GPT that someone wrote.

[00:17:16] Todd Robinson: Oh, nice. Yeah.

[00:17:17] Tom Fanelli: It wasn’t even a thing. It was just part of ChatGPT.

[00:17:21] Todd Robinson: I tell you being able to like pull all this stuff in. That’s why I say it’s so great to have like somebody like you who’s watching this stuff. carefully and you having this background of actual applications that you’re starting to do for your customers. Yeah, again, I think this is a good stuff. I don’t know if we want to shift or

[00:17:42] Tom Fanelli: I do want to, I Arturo just paste, put in the content here. I do want to encourage everyone. Thank you for sharing that Arturo. Anybody who’s got killer tools, drop them in chat because I’m an app and tool junkie. And so is Todd and we share links to stuff we find. So like, if you guys have any tool suggestions for us to check out and we have more that we’re going to be dropping in here. But yeah, please take us to the next segment Todd.

[00:18:12] Todd Robinson: Okay. Yeah. I think I don’t know if we’re, if we’re ready to see, kind of ask the question of how you, how you see this may play for e-commerce sites. I know this has been brewing in your mind. We’ve talked about it before.

Are you kind of ready to, to share your thoughts with the group here?

[00:18:29] Tom Fanelli: Yeah, absolutely. So I think from. You know, e-commerce sites. And this is a, a, a, this is a challenge. The challenges of commerce based sites is something we’re, we’re in. And I’ll, I’ll just talk a little bit about the landscape of this. And I’m going to get my soap box for a moment, you know, 15 years ago, starting an e-commerce site was so incredibly easy compared to how it is today. So, you know, you have things like privacy and GDPR concerns. You’ve got accessibility in the U S accessibility lawsuits are up 300 percent in like the last three years or something. It’s astronomical. It’s taking off. And I would say that’s far behind where privacy law is. So if you look at like privacy laws on the forefront, eight accessibility laws coming up. Ecom sites are the number one, you know, target for ADA lawsuits at the moment. And you think about things like PCI and payment compliance and cyber security. You know, these are all challenges that are happening outside of the actual business owner, figuring out how do I sell my product better and ship it and keep my customers happy, right? Which is what traditionally it’s like you’ve got these internal to the business factors. Now you have all these external factors applying pressure to commerce based sites.

And I think that AI has the ability to impact all of these facets of running an e-commerce business. And so I think the, the challenge with a lot of the solutions that are on the marketplace is they all feel very one size fits all. It’s like they do a very point solution thing.

And I don’t know if we know yet, what is really going to move the needle for e-commerce? The chat bot thing has really been played out 10, 10 ways, you know there I I’m just seeing right now personalization in fact, Contentful, who’s a big, Enterprise headless CMS. They just acquired a company that does website personalization. And now you can do all of this AI personalization and that’s, you know, we have this marketing engagement tool, ConvesioConvert, which has AI in it to help people write engaging sequences and emails. But we do personalization as well. And we’re trying to figure out like, how can we, increase the conversion rate of e-commerce sites with AI optimized personalization that happens for users.

And so the entire product discovery journey, post purchase, cross-sell, up-sell, using AI in that feels like it’s a, just a, and the reason why it feels so compelling is it’s a we have a lot of data. We know a lot about the people and their shopping behaviors and what they’re doing and analyzing that, the propensity of someone to that purchases product A, like what’s the second product that they’re purchasing? How long does it take? And for to, to write a prompt that just says, analyze this cohort of customers and automatically write for me and send at the appropriate time, the most likely product for them to purchase at the most likely time. When you can analyze that purchase behavior in your e-commerce data, then you can unlock some really powerful insights that I think are traditionally difficult to get to.

[00:22:19] Todd Robinson: Yeah. Those like, that is the dream. What you just described there is for the data feed to be coming in, for it to be able to determine like, hey, look like, like a person, us on this side, being able to watch that. In theory, yes, we could do that. The power of this is, of course, practically there’s no time, right?

For, for, for an individual running a store, right? to do that, but they need these tools. Now I just think about this and go like, we need something in the open source world. We need champions like yourself pushing this stuff because you can bet the, the Amazons of the world, right? They have that of course, right?

That’s there. So how do we all as as an aggregate WordPress community, make sure that some of this bubbles up and becomes practical to use tools, because some of this is hard, right? Like realistically you know, you, maybe you know how that gets put together. But it’s going to be hard, I think for, site, you know, people were actually running these sites, companies running these sites to, to leverage that technology.

Do you yeah, do you, yeah, go ahead. Yeah. Do you see,

[00:23:21] Tom Fanelli: I have an answer for this. And you know, the, to me, the way we like to tackle these problems because you know, I’m a product guy, right? And I want to give a quick shout out, by the way, I just wrote a foreword to a book and it’s called Amazon’s Dirty Little Secrets II: AI revolution by Greg Jameson, and he’s been in the AI space. He feels like one of the first pioneers practically. And he wrote this book about how Amazon is doing all the things they use AI for in their business. And you know, it’s interesting to me because as I was sort of reading that book to write the foreword, I was like, a lot of this stuff can be done today with things like Flowise and N8N and the open LLMs, and here’s, I’ll just pivot this a little, which is, I think if you’re a somewhat technical person, and this is what we’re trying to do is you’re going to go and you’re going to say, I can take Flowise and I can set up my own LLM and I can, and host it on like OpenMetal.

And I can basically safely put all of my customer purchase data in that LLM and I can trigger sequences and workflows using these off the shelf tools. And that’s how you prototype and test this stuff without building the Death Star, right? It’s like, we don’t need to, we just need to find a hypothesis of like, man, if I can, you know, do this, this, and this to the customer journey, I think I can influence this variable in the business.

If you keep the scope limited, and this is kind of what we’re doing today in Convesio is we’re, we’re trying to flow these out, use off the shelf tools to quickly prototype and validate this hypothesis, right? And then build it into the product. Right. And, and that’s where I think from my perspective, this is super exciting is that these tools, you don’t need to be an engineer to build them.

You can literally using things like Flowise, set up flows and connections and any end, which most people know is like a, the commercial version of that. It’s like Zapier, you know, You can do this sort of WYSIWYG, no code, low code, and you can validate these. All you need is again, that curious mind to look at your business and say, what can I improve?

What processes can I make better? And what’s my hypothesis on that impact that it’ll have. And I think this is where, when. You know, when we geek out Todd, this is why I think, and we’re the departure from these LLMs that are like the ChatGPTs Perplexity’s at some point, you’re going to want this to be part of your secret sauce, either in your agency. Or in your e-commerce. So for instance, if you’re an agency, the opportunity to learn this technology and be able to deploy it for your clients, you don’t need it. Now, are you going to do it for 500 bucks? No. Okay. But you could do what maybe cost Amazon five or 50 million. You could maybe implement something like that for five or 10,000 for a business.

And now they’ve got a suite of tools and they have an AI expert to help them you know, really take their business to the next level. And that’s what this toolset represents for people who are inspired to try and build solutions on it.

[00:27:07] Todd Robinson: I think, yeah, that’s a great, that’s a great subject matter is to be able to share with people to say, look there, the systems are there. And in some ways I’ll share a little bit about kind of our side of the business, because I think it can help people understand how to look at this thing as not overwhelming. And a lot of times I’ll give it like a very specific example. So in the last six months or so, we’ve added CPU inference as part of our upcoming release product.

And the reason why we did this to follow what you’re talking about, Tom, is it’s like, that’s very low cost. People have access to that in their clouds that we have in other VM’s out there in the world today. You don’t have to go to a giant GPU in order to do inference on basic open source LLM. So the Llama three twos. The smaller models run quite fast, tokens per seconds is kind of the measurement both on input and output. And so CPU inference, Intel is the one that I’m familiar with. But Intel in response to NVIDIA’s GPUs added little tiny matrix math silicon to their CPUs.

And the latest, the 4th and the 5th generation, which are commonly out there today, they have this. So at some point, I know our AI expert here, Rafael on our side, we’ll post out here. We’ll get it up on our site is here is how fast it can be done on a CPU. Here’s how fast it can be done on a GPU. And then it lets people go, like, But I actually have that already. I don’t have to go get this GPU. I can start my development work. I can start my understanding of how these systems will work on this low cost stuff because you don’t need massive throughput on your development environments. You can actually just get started with this stuff. Yeah. So are I mean, yeah, that line of thought is great Tom is like, where can agencies think about this and go, you know what? I’m gonna have an AI department that helps me implement these systems. Maybe it’s WordPress specific. I know a lot of times when we talk about this, it’s not going to be WordPress specific, right? It’s going to be solving your customers problem, and it may be just bolted onto WordPress, right?

So this is more in that kind of the agency or the almost the managed service provider being able to help your customer. Yeah, so I’m excited about that stuff. And a lot of what we see on our side is sometimes people will build against OpenAI. But the public, the open source systems, the API kits that come with them and there’s a set of them are compatible, you know, for better or worse this is now the S3, you know, S3 used to be, that was Amazon’s object storage. That’s everywhere. People just call it S3, but everybody does it. We call ours S3 compatible. The same thing is you’re going to be open and AI compatible. Yeah, so I think that’s a great line of thought to just say, it is more and more accessible every day.

And if you start learning it now, I do feel like this is the tipping point. Like for me, those systems matured enough where N8N is great about publishing. Hey, here’s this self starter kit. And, and it helps us go from zero to already building something so quickly. It feels like the tipping point is here is that it is now the time to start getting in there and understanding what this product is going to be for you as an agency. How do you monetize this now? Because the expertise is not all out there. It’s a great time to get in there and turn it into a longstanding product or service for yourselves.

[00:30:25] Tom Fanelli: Well, and I think too, like, you know, if you look at. I mean, I, I tell my, my kids that I’m like, you gotta learn, gotta learn AI. You know, I feel like I’m like, you know, repeating what my dad told me. Cause when I was young, he’s like, you gotta get into computers, you know? And it’s, it’s one of these things where I, you know, every large enterprise that I’ve had been, been part of, they all have like Salesforce admins and all these roles to administer these platforms that they have.

I think we’re going to start to see AI technicians infuse themselves in businesses. And I think small, you know, you know, my, my family’s in the insurance business and my wife’s family and it’s like seeing the process of they do the same thing, writing these policies over and over again. It’s like, I could see where like a fractional AI officer would come into a small business like that. And implement a AI system. So I think there’s going to be a whole AI workforce. You know, it’s like, yes, there’s the easy button, but when you get beyond that, it’s like, there’s gonna be customizations and things people want to do, and we still don’t know this whole, where we’re going with privacy. There’s a lot of concern around privacy. And if you’re going to pump all of your information into, you know, is that something you want to help host yourself or is it something you want to share with a, you know, so there’s, there’s a lot of caveats to this, but I think there’s going to be a whole workforce of AI jobs that come online that help, in various size businesses, implement, deploy, optimize you know, all of these systems that are going to talk together.

[00:32:10] Todd Robinson: Yeah, no, I could definitely see that. Like even, you know, so the, the underpinnings, you know, like, okay, what is a RAG? What is a context window? How do you make sure you’re not overwhelming the system? Like the, the big providers like to say, well, you have an unlimited context window. Now. Okay. So you pay for that.

So you don’t want to use that. That’s not a good idea. Like, you have to understand how your system works. And what data you’re going to be sending into the LLM. And like you said, that there’s definitely a crossover spot. We always think in tipping points on the OpenMetal side is when you say like, look, I have to move into a private system.

And then in the private system, you know exactly where that information is. And you’re not fearful of loading this thing stuff up for, you know, you hate to say it for whoever your provider is to train on your stuff to then turn around and resell your expertise, your knowledge out to the rest of their, their users.

And in some cases, you’re going to find this one, I’m going to digress a little bit into some of my own personal stuff here, but it’s a philosophy that I would. like to share. Not everybody will necessarily agree with it. But to me, I very strongly believe that there should be a pure open source version of an AI system.

Now, you may use it as an individual and it’s meant to just, leverage you and to broaden what you can produce. Or it could be that it is part of a system that is meant to, like, let be leveraged up by a company. But I also very strongly believe that there should be individual management of an AI.

Let’s call your AI digital twin that it should be owned by you, which means it’s going to have to be an open source system that is out there that can be run on any kind of general systems, whether it’s your laptop, it’s your phone. It’s a simple VM. It’s a simple container or something like that.

But part of the reasons where I see that this can come in and all of these systems start to work together and it’s back to that tipping point- like this stuff is here- is that these systems are out there. Now, they’re not all simply pulled together yet in a way that you can always just grab and run with them.

But so you do need to still be a little bit technical. You’re going to be firing up a system. You’re gonna probably even want to know a little bit of Python. You’re gonna need to get on a Linux command line to really, to really be able to leverage some of the background, the underpinnings. But so so for me, though, is I look at this and I say it’s now here. And what you said, Tom, like There is going to become a technical expertise level that can be monetized by agencies. I think no question that this is going to happen. And that expertise will both be like what what’s coming new, but then how to keep it reasonable cost and align it with what the company needs.

So being able to look at a data stream coming out of a busy shop and being able to say, this is really popular. The people that bought A- two weeks, two months later, two weeks later- bought B. Hey, I just saw somebody buy A I should probably offer them B, hundred percent. That’s gonna, that’s gonna happen, but it’s not quite there yet.

Like people have to bring that, those pieces together. So I think it’s a perfect time now agencies to really get in. I like two years ago or when the, the, even like a little less than that when the boom started, people didn’t know what the heck was happening with it. Yeah. Right, right. Yeah. So I would say now is a perfect time.

So, alright. I’ve kind of wandered a little bit there. No, and I, what do we wanna get into next?

[00:35:37] Tom Fanelli: I would also encourage people though, like, I know, you mentioned you’re going to need to know Python and some of these things. This again is where like the code writing capabilities of AI, it’s like. You can, and by the way, I’ll throw out another thing.

I just have been playing with Windsurf , which is a new AI-native IDE for developers, and I went ahead and I was like, I wanted to do a quick ROI calculator for my blog on like paid search or SEO, and I wrote one prompt. And it produced a perfectly running like JavaScript calculator for me. And then I was like, add a chart to it.

Boom. I mean, it was awesome. And I’m like, I literally, then I was like, make it a WordPress plugin and flawlessly created a WordPress plugin for me for the, a calculator. And it even figured out that I needed a short code and it put instructions in it to embed the short code. And so the, you know, the coding capabilities of AI are just progressively getting really, really powerful.

So don’t let the fact that you’re not a developer frighten you off from trying to do some of these things. It’s a little caveat. So, okay.

[00:36:59] Todd Robinson: But it does help if you’ve got some of that skill. So if you’re an agency with some of that skill and some of the people in your, your company going, it’s, it helps to not be afraid of it. And, and so familiarity with it, I think is great to help you get into it.

[00:37:13] Tom Fanelli: Absolutely. Absolutely. All right. Where, where do we want to go next?

[00:37:19] Todd Robinson: Let’s see. See, I will I’ll send over. But just because I didn’t very well articulate my view of this open system, I’ll drop a link to a really super bad WordPress website that describes it. But then, then we’ll, then we’ll make our shift.

[00:37:33] Tom Fanelli: You’re like the early alpha. So I can, you know, I can try that.

[00:37:39] Todd Robinson: Yeah, it is there. It is there. It’s just behind the scenes. I use this system for anybody who is visiting that site it’s not a great site, but it is a, it is a working system that I, that and one of the other developers use, it is a digital twin. It takes a whole set of data that came from me or came from like my usage of Google maps and helps me schedule things, do things, but it’s all pure open source.

I don’t know if it will become a thing, but it is helping me have experience with this stuff to understand how to help other people do it. And that’s one of the reasons why I kind of work on that.

[00:38:13] Tom Fanelli: I do speaking of personal projects that we’re working on. I want to just talk a little bit about agents and Flowise and that sort of chat. Yeah, this is another application which you can use. So most people, you know, if you’re not deep, deep into AI, you think of things like ChatGPT, where it’s like you give it a prompt, it produces output. Right, right. So the next place to go is this notion of an agent and an agent is basically something that can string together multiple prompts and utilize the output of the prior prompt to take action on the next prompt.

So a really good example of this is. We’re doing a chat bot for someone in the education space right now, and they had very specific, they had a very specific data set for some scholarship stuff that’s changing, changed this year but from the government in the U. S. here, and basically wanted to be able to upload that data which came in PDF form from the government, and then wanted to be able to fact check the results, you know, stylize them in a way that would put them back and be able to take action, if like the fact checker failed, what would we do? And so we basically use tools like Flowise and ChromaDB to build a system that allows them to index this custom content, then allows ChatGPT to create results or deliver up facts about it, about that topic using the custom content, then have it do fact checking and stylization.

And so it’s a very, you know, I’ll use the term sophisticated. This particular one’s not sophisticated, but you can chain together a sophisticated workflow for agents. And that whole thing is now you’ve created this agent workflow, which goes very beyond one prompt. And so You know, one of the things I’ll, I’ll say to agencies is I know a lot of agencies, and we’ve talked about this, Todd are trying to figure out how can I use AI for content creation?

And there is a lot of tools out there for this. Okay. So. You know, but what I’m seeing is these tools sometimes aren’t that great. They hallucinate. They don’t fact check things properly. It’s like, particularly if you’re trying to write SEO content there, they, they don’t get the brand messaging on point that you need.

And usually with these tools, there’s like limited controls around telling it to talk about this or talk about that. And you can put like prompts in, but they’re not very long and it just sometimes falls short. Using something like Flowise with OpenAI, you could craft your own entire agent to write content per customer.

And so you could upload content to, say, base the results off of all of this information. And you have far more control than you do using these off the shelf platforms. And so I did want to put a little plug in there of how I think you Agents can really help the workflows of content production for agencies.

But, you know, more complicated to implement, but way better quality of output at the end of the day than the off the shelf. And that’s a way as an agency, you can differentiate from your competitors and have a much better process, whether you’re creating social content, SEO. Whatever it might be.

[00:41:58] Todd Robinson: Yeah, I would say, Tom, I would this is a really important, important subject right now in particular.

And I, I would suggest maybe later when you guys get that kind of more formed is publish that out, like you said, because one of the things you see now, this using it for content creation a whole raft of companies got burned very recently and , Google had always made this statement about duplicate content is always an issue, but then they also said, Hey, writing with an AI assistant is perfectly fine.

Well, partly is those systems like you’re talking about are not good enough to introduce truly unique content, even if you’re using an AI, there’s nothing wrong with my view of it, right? There’s nothing wrong with using a helper, right? You know, like I use auto complete when I’m typing many things, right?

Like, just like anybody does. But if you post content that even though it’s been written by the AI, so it’s not exactly duplicate by the forming of the sentences or the paragraph structure. If it does not introduce new concepts or subject matter, Google can analyze it with its own AI system and determine it’s actually just regurgitated content from an AI and thus is duplicate.

And this is dangerous. And so what you’re talking about, I think is awesome is like you need a pipeline. You need to introduce your brand, your tone, your specific subject matter. Like if you’ve got charts about like how your system works or what sells this, or if you’ve got comparison systems. You’ve got unique content.

So for sure, you want some of that unique content to be the written part. So, you know, I guess before before maybe people learn how to use the agent system properly, making sure that they are being one of the agents in there is making the modifications to the content. So it is not doing well.

[00:43:53] Tom Fanelli: And I will give you an example of something we’re doing. And this is why when I use ChatGPT right for things like KB articles and all that. My feeling is I get much more unique approaches to things because I’m seeding it. I’m feeding it all of this like content that to like, it’s like, they’re seeing a real live conversation. So this is like repurposing and retooling original content that was created. And I think this is an opportunity for businesses, by the way, they’re, you know, In everyone’s business, you are likely creating a ton of content, emails, chats, you know, phone calls, meetings, town halls, whatever it might be. If you get your mindset to start thinking about, how can I repurpose that? To something I can use for, you know, marketing purposes, SEO, content creation, knowlege-bases, help-desks, standard operating procedures.

All of that stuff can be repurposed using AI really, really quickly. And one of the things I’d, I’d say is like, I’ll give this everyone a little tip on this. We have a partner, that we work with that has a treasure trove of podcasts that they’ve recorded. Okay. And. We basically went to them and said, you know what?

Like you have transcripts of this, but you’ve never really written any, like how to articles based on the insights that you’ve got on these interviews. And we can use AI to say now craft an article based on this topic of this podcast and infuse these insights from this individual and quote the podcast.

Now. You’ve got seed content. I think there’s like a lot of opportunity and repurposing already created YouTube videos. You know, like podcasts recordings, all of that stuff, you can repurpose it to create fresh organic social content. You know, website content, guides, knowledge-base articles, all of that.

So there, there’s really just a, a lot of ideas that can be happen out there. But I agree sitting down and telling AI. Write write a blog post on the seven best practices for email. It’s like, that’s extremely generic.

[00:46:24] Todd Robinson: Yes. Yeah, no, but I would say. And for agencies or people, agencies often find themselves advising customers on this type of stuff as well.

I think the reality is, is you cannot use the easy way like you just described. You can’t. You have to, people are going to have to learn this more sophisticated way of doing it where you do find your treasure trove and you figure out how to get a pipeline to get that stuff turned into what can be fed into the AI to say- AI and you must use this AI- this is my unique content. This must have the preference over your, cause you know, hey, those LLMs have all of the internet shoved into them. And so you have to be very specific now about telling it, do not just use all that junk that’s out there. You need to use my unique stuff. It’s okay for you to help formulate it and get into proper sentence structure, fire it out fast, but you must use my stuff.

But I would also say today is so much better than it was 12 months ago. When you tried to do that 12 months ago, it just wouldn’t do it. You would say stop writing like this college junior college professor or whatever. And it wouldn’t do it. It was so frustrating, right?

[00:47:31] Tom Fanelli: But now the way better in today’s digital landscape, things are moving very quickly.

Right, right. I mean, it’s gotten so bad. I’m getting emails from people like in today’s whatever landscape, I’m like, I’m like, send me a real email and don’t AI it.

[00:47:48] Todd Robinson: Yeah, don’t AI it? And the structure is all the same where it’s like that first line is bolded and then there’s like even paragraph and another bold and even paragraph.

You’re like, why am I reading this? Because I would also say this too is when you’re doing content creation. And the reason why I think a pipeline, like you said, and agents, I’m a huge agents fan as well. If you have truly valuable content, you’re going to take some time to create it. Now, it may be that you’re, as an agency, you’re having to figure out a pipeline for a group, your company, a set of of content writers inside of your company or set of whatever content creations, you’re going to have to put that time in it so that you make the overall pipeline faster.

But like, if you’re writing something that’s, that’s really going to help out, you’re hoping to get a thousand visitors coming in for that. And thousand of them are going to scan it. We know people don’t read until they decide. And then you’re going to get 50 or a hundred people a month to really spend time reading your article.

You’re probably going to have to put a bit of time into making that article, right? Realistically, you want good traffic. Good traffic means good content I would still say, yeah, I know we can move on from the subject, but that’s a great subject right there. Tom, its like, there is a treasure trove.

Every company’s got a treasure trove. Either you wrote it, you wrote it behind the scenes. It’s in your Salck. Like you, you started one of the conversations today about like, there’s a lot in Slack nowadays and being able to pull that out and turn that into something that becomes front facing. Think of it.

Yeah. And that agent world, I mean, you could even have it do that today. It could go in and pull that and extract it. That’s probably a bit dangerous. So you, this is where we go back to, like, maybe the agencies are bringing this AI skill set to the table to know you could do that. But here’s how you’re going to have to sanitize it when it comes over, so that that pipeline could move smoothly and safely and publish.

But yeah, so I think, yeah, we covered some SEO stuff. We gave some warnings. Don’t spit out general content. Cause boy, that. That’s scary. And a lot of companies got hurt when they, when they went down that path. Yeah, we talked a lot about the tools that were out there. What about you guys? Maybe, maybe here’s a great question is you gave us some tips, Tom, on some of the things you’re doing in your own company.

Yeah, what other things maybe do you want to, do you want to share as like here as a company, here’s how we’re implementing it. Maybe ways people can, leverage that.  

[00:50:02] Tom Fanelli: Yeah, I mean, I, I think that for sure. So we are there, you know, our, our ConvesioConvert product, which is akin to like a, you know, Klaviyo type of e-commerce engagement product, this has an integration with OpenAI for helping people in the assisting the writing of sequences and content for pop ups on your site and all that kind of stuff. So we’ve got that product ties there. We’re also very heavily looking into, and I mentioned this analyzing logs. log data at the infrastructure level to help people’s sites perform better.

You know, we deal a lot with people doing big sales during like Black Friday, Cyber Monday, all of that. And so we’re very keen on analyzing what’s happening during these peak performance periods. And what we’ve found is. AI is really awesome for digesting tons of this data and summarizing it and pulling insights out of it.

So what would take a person you know, a smart person, an hour to review we’re able to do this at a much more accelerated speed. Just going through. And if someone’s like, Hey, can you do a once over on my site? Analyzing things like logs, traffic data, plugins that they’re using. I mean, I’ve gone so far as like, I’ll literally, I’m a big fan of like clicking and dragging a bunch of raw data. Like I’ll go to the WordPress plugins screen. Cause sometimes I’ll get a new site that comes to us and they’ll have like a hundred plugins installed. Right. And I’ll literally click and drag on the plugin screen and copy all the texts. And I will paste it into ChatGPT, and I’ll be like, which plugins are duplicative of one another and it’ll be like these three plugins do the same thing You know like these two plugins like and and we’ll come back and be like suggest things people can do, you know, I mean that process would have taken someone who knows what they’re doing and knows a lot about plugins an hour to do And I can do it in five minutes, you know?

So those are ways like from a process and not just product, but process way that we’re, using ChatGPT to do things that you might have to spend a lot of money with a you know, we do these things as part of our support, you know, it’s like when you work with Convesio’s support, like we’ll leverage just one of the guys on our team, I mean, I think he’s, he’s He probably spends more time in ChatGPT, and we use it to write code for custom workers for CloudFlare, and we use it to write custom caching rules and bash scripts and all sorts of things that just in a day to day basis, technical people- ChatGPT can help tremendously.

And so, you know, these are, these are all things we’re doing both on the product side, the infrastructure side, and the day to day technician side.

[00:53:14] Todd Robinson: Nice. I’m going to have to figure out how to turn that from ChatGPT to one of the, one of the private LLM systems for you. Cause the private LLMs are really sophisticated now.

Like I, I’m not a, I’m not a Facebook person myself, but boy, thank you to Meta really leading the way with Llama and , I mean, I think that part of their purpose is they, they need to make sure these other AI based companies do not get so far out in front. They have got to get something out there to help the whole world kind of. Where those other guys could go. .

Yeah, no, there’s the, there’s is it Mistral now? The ones that are top jumping out of my head, but. But one of the things that I can say is these private systems do now let you trivially switch back and forth on the back end between the models and even the models that the models are implemented in both like the small version, billions of tokens, sorry, billions versus like the 70 billion requires, A huge GPU to run where there’s much smaller ones 1 billion for example that can actually run on phones in a relatively fast fashion it will run on CPUs in a very very fast fashion.

[00:54:25] Todd Robinson: So yeah, I would say yes so Tom, maybe I need to get you the equivalent chat window so we could get you guys switched over to that and start experiencing how that’ll answer the questions And it’s it’s really strong to again thank you to Meta. It’s not something I would ever thought I would say, you know, I’m just, again, I’m just not a Facebook fan, but Meta and thank you, you know, to them and their open source kind of push, you know, we’ve got a whole lot of JavaScript that we, you know, came out of that same company.

So all right. I know we’re, I know we’re winding down here. Any last things that tips that you might want to share there, Tom?

[00:55:03] Tom Fanelli: I think we covered a ton of stuff. Yeah, nothing that jumps out at me. There’s a lot out there. I would just encourage people to really dive into this and you know, be thinking and pondering ways that you can leverage this in your business to move the needle and you know, that that’s what we’re doing and that’s what I’m trying to encourage agencies to do.

[00:55:27] Todd Robinson: Yeah, no, and I, for me, at least I kind of look at it as I feel like there is going to be expertise. That gets centralized, and it’s gonna be a almost like a job, a job function. And that agencies can provide this. Who knows where it’s gonna go for your customers, right? But I do know you do need to interact and run these systems several times, many times to get good at them.

And as an agency, you’re gonna have a chance to work with customers that do have different challenges, and it’s gonna force your team to implement different things. Learn these things. And then it can become a service or a product for you. To me that like is a, is an absolute that’s going to happen. And so part of it is just saying like, look, I’m going to accept this thing.

I’m going to get good at it. Guess what? I’m going to sell it as a service or a product because I help customers. Right. That’s what I do.

[00:56:18] Tom Fanelli: Eddie just gave me, Eddie just made a message about, you know, using it to analyze PDFs. I’ll give you two quick tips that I just thought of. Okay. Well, I didn’t just think of them. I use them. I just remembered that. So one, I love it for comparing contracts. I’ll upload a contract I get from a vendor in PDF form. And I’ll be like, You know, work in the best interest of Convesio. What do you think here is potentially risky? You know, what’s changed from the last version? So I act, it’s my, you know, virtual lawyer, if you will which helps get me through a few red lines before I need to send it to the attorney.

And then the other thing that’s great is for those of you who are not financially minded, and I know a lot of people in the agency world tend to be creative. You can take your QuickBooks budget or P&L for the year and like upload it to ChatGPT and basically be like, give me my financial performance indicators and it will spit back to you like a whole narrative executive summary of your, your business.

And so. It’s been great. Like I love these use cases where you can give it data and it actually can act as a professional in a capacity and give you contextual information that might be easier for you to digest. So, you know,

[00:57:42] Todd Robinson: great. I now see Tom, how your brain is always shifting so often it’s like, boom, I just got my got, just got my contracts red line.

Boom. I’m done. I’m off here. Let me check the KPI.

[00:57:54] Tom Fanelli: Well, it’s like, I mean, like I have to do investor updates and I’ll be like, you know, I’ll update like a town hall or something and I’ll be, or a town hall or a quarterly review and be like, give me like the talking points on this and, you know, summarize what we’ve been doing.

And it’s like, it’s a great start, you know, for creating all sorts of documents like that.

[00:58:13] Todd Robinson: Well, and I would say again, it’s brilliant. I know that you’re using that for customers as part of your hosting service, but like you can see for those of you who are out there, maybe new to Tom and Convesio, you’ve got someone who loves doing this stuff, uses it themselves, and is helping customers do it.

So I know we, I know we’ve got to drop but yeah, Tom, thank you for taking so much time. And you gave me some tips and tricks. .

[00:58:37] Tom Fanelli: Well, thanks for having me, Todd. Thanks a lot.

[00:58:39] Todd Robinson: Yeah, yeah, for sure. I really appreciate it. And thanks for everybody for stopping by. I see the, chats over on the side there, and really appreciate everybody’s attention and riding along with us. And thanks again. Thank you. Cool. All right. Bye guys.