This is the best approach to reveal most common mistakes.
Is the NLP well trained?
Use the NLP trainer/tester to ensure the right intent was found.
If a score is too low, or a different intent was matched, refine utterances of the intent.
Is there an interaction with appropriate intent condition?
Often there's a missing or incorrect intent/entity condition.
There's a response fallback in previous interaction, which intercepts all user responses
Use conversation history to reveal a context, where your response was processed.
The general condition should be below more complex conditions (2.xx)
With older versions of Wingbot it's important to keep more specific conditions above the less specific.