Hey, in this video you will see the cheapest AI personalization system available that you can use to personalize your cold emails with AI so that you can have better reply rates and uh you can do that at scale for a fraction of a price of other tools. If you're interested in building this system, you can follow along or if you want the system to be integrated within your business, feel free to reach out and click the first link in the description. With that said, let's get started. The first thing that I want to focus on is how AI personalization systems work on a first principle basis. There are just two component to those. First off, we get some kind of personalization data. That could be data from their website that are from their LinkedIn data from any other source that we have available. And then what we do is we use the LLM, we use chat GPT code or whatever the case might be to create a personalized message based on that data. So this is as simple as possible AI personalization boiled down to the components. The way this works is twofold. There's an integrated approach which combines both of these two things in a single API for example. So it combines all of them within chat GPT within code uh perplexity even and there's a separated approach where you do each of the steps in a separated way. Now the integrated would be the best possible way right because like that is the faster we already have the information available and the AI will just do the web search and then we will extract the information and then out of the information we will create the personalization. That is all good. The problem with that is if we have a look at the pricing over here for Chad GPT uh the pricing for the models is not that bad. But if we search for the web search, we can see the actual price of uh how much it cost to run this kind of personalization through charg like this. So over here you can see uh an example of how much would you pay for 1,000 calls. We can call the 1,00 personalization if we wanted to. So you would pay $35 per 10,000 personalization, which is kind of a big number over here to to pay over here. So there is the actual problem when it comes to the different integrated solution. The other approach is separated and uh over here we just get the data from a website using a web scraper. So there are multiple options to that and probably one of the best one that I found is Tavi. uh Tavi is capable of getting uh search data in a very uh good formatting uh which is perfect for LLMs and what you can do over here is you can also ask for a specific query within uh Tavi. So for example you could potentially ask find me find me all the case studies does this client mention on his website. So this is um how this tool works. There's one tool that I haven't mentioned over here within the separated solution and this tool is called crawl for AI. So crawl for AAI is a web scraper uh that is LLM friendly. So the formatting of this web scraper is perfect for uh chat GPT for code for whatever and it also has the advantage of being open source. So what we can potentially do with this is we can take this version, we can host this on a server and in that way we don't have to pay the cost of running single APIs all the time. Uh so this is the actual tool that we decided to to when it comes to cost we are comparing everything against the best tool out there which is clay.com. Within Clay.com, if you were to pay 50,000 credits, you would pay 800 per month. So, if we do the division of that, we would get approximately 0.016 per credit. So, this is what you would pay with this solution. Instead, what you are paying is you are paying the server cost to host a crawl for AI and that would be approximately 50 bucks per month. And then you are paying open AI API and more specifically you are paying GPT4.1 nano which is 0.003 per call. So if we are looking at uh the amount of email personalization that you can do so the amount of personalized leads that you can have uh for $22 you can personalize the 50,000 leads instead with clay you spend $800. So this is uh one one/4 of that. And then if we were to compare that with one uh 100,000 leads uh email personalization, the cost here would be 354. Instead, if we look at uh that within clay, you see over here that we have 1,500. So this is a real powerful solution. And uh yeah the model that we uh we choose is GPT4.1 nano which is one of the newest model for CHP which is both good in terms of quality and also is very good in terms of cost because it has the the best cost across every single model from Chipity. The the thing about clay is that this is a really powerful tool and if you don't need to run as much volume as we need in this specific case is fairly good with that. But with this solution we were able to save uh four times the amount uh instead of actually paying for clay. So with that said let me show you what is the workflow structure and how everything looks like over here. So first off we check the lead list. We check a specific lead list for personalization opportunities. Um that means we check if they have been personalized before or if uh they have some website present within the list. If that's the case, we extract the domain and then we build the crawl for AI configuration file. And I will show you this part because this is the most important component and what makes crawl for AI an actual really powerful solution for this build over here. Finally, we call crawl for AI and the end results that we will get is uh the information on 10 15 page pages of a website and pass this information directly to OpenAI so that it can create a personalization email and finally update the Google sheet. So let me show you an actual example of the result. So if we go over here, this is a lead list example and over here we have the different personalization that we could get. In some cases where we are not able to get the personalization maybe because the website is blocked or we cannot have access to that. Uh we don't have any kind of personalization like that. In other cases the personalization is pretty specific and gives a very good um messages. Keep in mind those can also be improved based on prompting and based on a lot of things like that. Now let's see um NA10 and how everything works over here. So the flow is exactly the same as the one that I showed you in the diagram. The thing that I want to show you first is the prompt from OpenAI and then this part related to the crawler config. So this is what is needed uh to be put it inside the uh crawl for AI um API call. And so uh what we are doing over here is first off we are filtering the domain based on uh the allowed domain that we want to choose. In this way we only pass on the specific domain that uh we uh have in this uh in this system. And then over here those are two really really powerful components when it comes to crawl for AI. We have a URL pattern finder. So if the system uh scrapes the website and finds different URLs that have this kind of patterns within it will prioritize those over any other kind of uh URL. So this is very useful when it comes to finding specific things within the website itself. And then over here we have a keyword scorer. So the keyword scorer is actually used to find specific keyword and give a specific weight to specific keywords um instead of um others. So if a page has more keywords related to testimonial portfolio case study that will be more important compared to other pages. So this is what we are doing over here. Keep in mind if we were to do different kinds of personalization, we could change this pattern over here and this keywords over here to match the relevant person we would need to do or we could potentially even make those dynamic and make the personalization completely dynamic based on uh the the personalization play or what is asked or whatever the case might be. So that's it. What we have over here is uh we have a solution where if this is a success we uh get an array of the markdown. So this is the LLM version of the website that is way easier to read. And then finally we pass this over to OpenAI and within OpenAI we ask to uh via email copywriter writing a oneline case studies we give them a couple of rules couple of examples couple of success criteria and that is able to give us a very very cool personalized lines and personalized use cases. So that's it. This is the cheapest system that you can use to personalize email at scale. And if you want that integrated within your business, feel free to reach out. With that said, catch you soon in the next video. bye.