I’ve created a monster…

0:00 Imagine you own a small clothing shop. Someone walks in and you know instantly if they're going to buy. If they're not very likely to buy, you let them browse on their own. But if they're very likely to buy, you send over your best salesperson. You guide them. You show them your best stuff. Because when you know who is likely to purchase, you make smarter decision. You focus on the right customers and you make more money. Now, I don't own a clothing shop, but I build websites, a lot of them, like too many. And every day, random people visit them. Sometimes they buy something but most of the time they don't. This is when I had this dumb idea. Can AI predict when a visitor is going to buy my product? And if so, maybe AI can tell me when I will get rich. It was a typical Monday morning. I was switching between asking an AI to build my next startup and browsing Twitter. Then I got slapped in

0:46 the face. My database build past $1,000 per month. And it's not because I'm storing videos or anything fancy. It's just because I have a lot of data. I built this product called Dataf Fast. It's a web analytics. You add a script to your site and it tracks your visitors, where they come from, what they click, and how long they stay. I've got almost 200 paying customers. And he makes around 2,000 in monthly recurring revenue. Each of these customers track at least one website. And if you imagine, each sites gets a,000 visitors per day. That's hundred of thousands of page views getting recorded every single day. I built fast, so I didn't really care until the bill showed up. But here is something I've learned after 7 years building tiny internet startups. Big problems usually mean there is an opportunity somewhere. So instead of trying to save database cost, what if I

1:33 could use all these data to help my customers, including myself, make more money online. So I went back to my desk and I wanted a simple way to rank visitors like giving them a score. Zero means that they're cold, they're not likely to buy. 50 is the average baseline. And 100 mean they're hot, they're very likely to purchase. Luckily, I had everything I needed because my product DataFast does not just track pages. It tracks the entire journey from the first click until they make a payment. So, first I had to analyze every website tracked with datafast. And for each of them, I generated a conversion profile. It's like a fingerprint of what a good customer looks like. I look at data like where do most paying customers come from? What device are they using? How many times do they visit the site before making a purchase? Then I wrote an algorithm. Well, an AI wrote the

2:20 algorithm for me. It takes one visitor data and compare it with a website conversion profile and boom, it spits out a conversion score. Something like if the visitor is coming from the USA, they might get an extra 10 points. If they're using a MyBook Pro, they will get 20 points and if the first time they visit the website, they might lose five points. Next, I wanted to test this data to see how good my predictions are. So I asked my AI code editor to extract data and create a prompt for another AI to test and visualize how good the data is and surprisingly it was much better than I thought it would be. I wanted to feel what it meant to have visitors across the planet hot cold moving in real time. So I viodated a world map and the result looked terrible like my drawing when I was 5 years old. For about 5 minutes I

3:05 genuinely thought I'd give up. Maybe this whole thing is just a stupid idea. Then I found Mapbox and within a few lines of code, the map looked great. Now the fun part, placing visitors on the map. I had to reverse engineer the world from city name to longitude and latitude to slap a tiny avatar exactly where the visitors are. Then I added prediction score. Each visitor would get a colored dot, blue for cold, red for hot, and gray for average. It started to feel alive, but not enough. So, I added more features showing the visitors hot to cold range, estimate how much they're worth, and display a percentage chance that they will convert. I showed it on Twitter and people loved it, which gave me, of course, the permission to waste more time on it. I added live events log like World of Warcraft, showing page views and page exit payments in real

3:50 time. Clicking on an event would fly across the globe straight to the visitor. Then, someone on Twitter said, "You should make the map full screen and auto rotate." So, I did that, too. Another thing people kept asking is a way to see the globe without having to sign up for datafast. So I added this new feature where datafas users can now make their dashboard public and share it with a link. So if you want to see the globe in action, there will be a link below this video. This real time map feature make me feel like I was running some sort of startup command center while sitting in a coffee shop in Korea. It was beautiful. It was exciting. It was completely useless. It didn't do anything for users apart giving them a nice little dopamine hit. I needed to go further. So I spent a few days designing this big Apple keynote style visual showing patterns, trends, deeper insight about who is likely to buy. I launched on Twitter and nobody cared. Then I

4:37 realized maybe instead of showing people the data, I should give them the data. So I went back to my laptop and built an API, a really simple one. You send a visitor's ID and you get back their prediction score, confidence level, and some metadata about who they are, like where they come from, how many times they visited, and what device they're using. It sounds simple, but it actually unlock lots of possibilities. For example, on my own site, Codefast, I use it to control when to show a lead magnet. If a visitor is unlikely to buy after 1 minute, I pop up a little offer asking for their email. And I have collected hundreds of emails in the past couple day using that method. Of course, if they're likely to buy, I leave them alone and let them purchase. And these predictions could be used for anything. Show a discount offer to visitors unlikely to buy or show a button to

5:22 schedule a call for visitors who are likely to convert and build more trust. Then I had this crazy idea. I asked an AI to create an AI. Literally, I gave it my own website data, thousands of page views, events, and purchase. It created a prediction model using machine learning. It waited over 30 visitor signals from clicks to visit counts to session duration. And when I tested it, the results were crazy good. way better than anything I had built previously. It was the first time I saw it clearly. We are just scratching the surface. Today, most websites show the same web page to everyone. But what if the headline would be customized to your preference? What if the page would rewrite itself to match your motivations? Not in a creepy way, in a way that helps you find what you're looking for. The future is not a homepage. It's your page. And maybe,

6:09 just maybe, this stupid little project is one small step toward that future. [Music]