- The first model we can consider for our use case can be a simple neural network that works on classification. We take an input vector and classify users as churned or not. - For event based inputs we will convert them to vectors simply by taking into account how often the event occurred in the time period or as a boolean value of whether it occurred in the first day or not. - Later on in phase two we can improve the model and use a transformer and take time into account. So rather than simply if an event occurred or not we can input on each day what all events occurred and create a series of neural networks over time. We'll optimise this by applying a transformer to make the model more efficient and predict the events that might occur the next day and the following days ahead.