We’re currently using AI for two main purposes. 1) By using ‘unsupervised learning’ approach, we’re categorizing websites around the world to a small amount of groups (around 20-30). Afterward we’re doing deep-analysis of few site in each group and labeling the group. by that we are able to auto-learn new sites or better categorize known sites without the need to analysis and can answer the question of “who is like me, who is better than me?” in more easy way (on technology aspects). We’re using the k-mean clustering algorithms using TensorFlow. The categorize process is still not on production. 2) Our second usage is anomaly detection in our big-data database in order to determent if any uncommon event has happen. We have many “human” rules today in the system and we use the anomaly detection to alert us for issues (or good changes) that we weren’t aware. We’re using Microsoft Azure Machine Learning Studio. We have more than 50 virtual servers gathering data in azure cloud, by our unique driver on browser. Our data includes many indicators from multiple (thousands) sites on the world, feeding our data-lake. Our data-lake contains data of the last 3 months more or less and until now we have gathered around 200TB with addition of 1TB per day. and we don’t have any rights issue with our data. The AI helps us to break the need to optimize site after site and build designated rules. With this amount of data, we had to using AI in order to make the system working. While we have some rules written by us, the AI help us to scale and to support many sites, different sites around the world. In our current stage, we’ re testing it ourselves comparing the analysis of the site done by us and the results from our AI algo. And most of the anomaly alerts are still being validated by us