lovely
lovely6h ago

Understand the User Question

one week ago, my chatbot worked pretty good, the questions that didnt match an intent (NLU) got answered via fallback prompt. when i discovered, that there are settings for the intent, i wanted to TRY OUT the NLU + RAG version, but i didnt know, that i cannot go back from to my old version. right now, the chatbot isnt working really well for questions that are not part of an intent/utterance, pleaseee what is this ... can you leave the settings as they are??? its good that there are many new features and options, but isnt it possible to still keep the ones that were working??? its really frustrating.
3 Replies
Tico | Voiceflow Assistant
Hey! no worries – you can bring your assistant back to the NLU-only + normal fallback behaviour.
Here’s what to check: 1. Open the Intent CMS (Intents tab ➜ “Manage”).
2. Click the ⚙️ Settings button in the top-right.
3. In Intent classification choose “Classify using Natural Language Understanding (NLU)”.
• This is the setting your project used before – it ignores the RAG knowledge-base layer and relies only on the intents you trained.
4. (Optional) lower the confidence threshold a little. Anything below that score will flow to the No-Match / Fallback path you already built. If you also added a Knowledge Base document step and want to disable it, simply remove or disable that step – nothing else in the project changes. After flipping the setting: • Hit Run / Train Assistant to retrain the NLU for development.
• If you use a published version, click Publish so production matches the new config. Your “no-intent” questions will now go back to the classic fallback prompt you had before. Links for reference:
NLU-only setting → https://docs.voiceflow.com (see Enable NLU Intent Classification section) Let me know if anything’s still not matching the way you expect 👍
lovely
lovelyOP6h ago
no, there is no such option! before it looked like this and now it just looks like on the secons picture
No description
No description
lovely
lovelyOP6h ago
@Braden

Did you find this page helpful?