You should be able to achieve what you want by using the Memoization policy for training and also, since it’s the behavior you would expect from you users when speaking to your bot, then then all of the stories in your training data will include the same beginning of the conversation should be learned by the model pretty well to predict the actions you expect to be repdicted. Why memoization policy would work here you can read here Trouble with ‘There is no memorized next action’ message
More on specifying policies: Training and Policies