Brief Summary
This video explores the concept of using AI to predict when a website visitor is likely to make a purchase. The creator details his journey of building an AI-powered prediction algorithm using data from his web analytics product, DataFast. He explains how he visualized the data, created an API to provide prediction scores, and ultimately used AI to build an even more effective AI prediction model. The video concludes with a vision of a future internet where web pages are personalized based on individual user profiles and behaviors.
- AI can be used to predict user behavior and purchase likelihood on websites.
- Data visualization, while initially appealing, may not be the most practical application.
- Providing data through an API unlocks more useful applications and integrations.
- AI can be used to create even more advanced AI models, improving prediction accuracy.
- The future of the internet may involve personalized web experiences tailored to individual users.
Can AI predict when someone will buy?
The creator introduces the idea of using AI to predict when a website visitor is likely to make a purchase, drawing an analogy to a clothing store where salespeople focus on customers who are most likely to buy. He explains that he owns a web analytics product called DataFast, which tracks visitor behavior on websites. Faced with rising database costs due to the large amount of data collected, he decided to explore whether he could use this data to help his customers, including himself, make more money online by predicting purchase behavior.
Building the prediction algorithm
The creator describes his initial approach to ranking visitors by assigning them a score from 0 to 100, with 0 indicating a cold lead and 100 indicating a hot lead. He leveraged the data collected by DataFast, which tracks the entire customer journey from the first click to the final payment. He analyzed data from various websites tracked by DataFast to generate a conversion profile, which serves as a fingerprint of what a good customer looks like. This profile includes data points such as the visitor's origin, device, and the number of visits before making a purchase. He then used an AI code editor to create an algorithm that compares a visitor's data with the website's conversion profile to generate a conversion score, assigning points based on factors like location (e.g., USA) and device (e.g., MacBook Pro).
Making data sexy
The creator wanted to visualize the visitor data in real-time on a world map, using colored dots to represent visitor scores (blue for cold, red for hot, and gray for average). After initial struggles with the map's appearance, he used Mapbox to create a visually appealing map. He reverse-engineered city names to longitude and latitude coordinates to accurately place visitor avatars on the map. He added features to display the visitor's hot-to-cold range, estimated worth, and the percentage chance of conversion. He shared the map on Twitter, and positive feedback motivated him to add more features, such as a live events log similar to World of Warcraft, showing page views, page exits, and payments in real-time. Users could click on an event to fly across the globe to the visitor. Despite its visual appeal, the creator realized that the real-time map was not providing practical value to users beyond a dopamine hit.
Building an AI using an AI
The creator shifted his focus from data visualization to providing actionable data. He built a simple API that returns a visitor's prediction score, confidence level, and metadata, such as location, visit count, and device. This API unlocked new possibilities, such as controlling when to show a lead magnet on his own website, Codefast. He shows a popup offering an email signup after one minute to visitors unlikely to buy, and he leaves likely buyers alone to make their purchase. He also suggests using the predictions to show discount offers to unlikely buyers or schedule a call for likely converters. He then had the idea to use AI to create an AI prediction model. He fed his website data, including page views, events, and purchases, into an AI, which created a machine learning model. This model considered over 30 visitor signals, from clicks to visit counts to session duration, and the results were significantly better than his previous efforts.
The internet tomorrow?
The creator reflects on the potential of personalized web experiences. He envisions a future where websites customize content, such as headlines, to match individual user preferences and motivations, helping them find what they are looking for. He believes the future is not a generic homepage but a personalized page tailored to each user. He suggests that his project is a small step toward this future.
