you need to retrain the model as new data will be available, otherwise, accuracy in prediction will drop.
How often it depends on several factors.
Consider there are two different approaches
- offline AKA batch learning: you retrain the model with all the data available, something could be expensive (so you do once a day or a week, etc)
- online AKA live learning: you train the model sequentially, with just the new data (mini batches), so you don’t need to keep all the data and processing it is much faster
Hope it helps