Answer
The trained model is wrapped inside a service that accepts requests, runs the model on the input, and returns the prediction as a response, often in real time.
Real-world example A mobile app sends a photo to a model serving API and receives back a prediction of what object is in the photo.