I believe there are two options for your situation:
Create more intents and train them to improve your agent's accuracy for the intents. This is what you are supposed to do if you have a lot of training data. If not, try the option 2.
Create a custom Fallback Intent with an input/output context.
Option 1. is obvious like sid8491 is saying.
Option 2. is kind of a hack, but I think it can solve your situation by providing the correct input context and solved mine too.
One example is when you want to ask users to write feedback. Since users can write any feedback (i.e. it's hard to train the feedback intent with data beforehand), this is where a custom fallback intent comes in. Since you expect feedback from your user, you set an input context for your custom feedback fallback intent and users response (feedback) will match your feedback intent.
You can create a custom Fallback Intent by clicking the three dots right next to "CREATE INTENT" in Intent tab.
Hope it helps.