DialogFlow CX : No match event forwards to Default Welcome Intent

[CDConvert]

Hello everyone,

I built this flow in Dialogflow CX (image below). As you can see, I set event handlers that forward to the same page in case of a "no match" event (I didn’t specify it for all of them because it’s not necessary in my context).

For the "Demande nom" ("Last name asking") page, I created an entity type defined with a regex that only accepts letters and a few specific characters commonly found in French last names.

For some reason, the event handlers work for some inputs like "John123" or "??", but for others like "é(", the chat is forwarded back to the beginning of the flow (Default Welcome Intent).

According to the "original response" JSON, it looks like "é(" matched one of the training phrases of the Default Welcome Intent with 80% confidence (the training phrases are "bonjour", "salut", "hi", and "hello"—so not related at all to "é(").

Interestingly, this issue only happens in French. In English, everything works fine.

Have you ever encountered this issue? Do you know how to avoid this situation?

Thanks a lot in advance for your help! ‌‌🙂

[a_aleinikov · Accepted solution]

H @CDConvert ,
Yes, this can happen — Dialogflow CX sometimes matches unexpected inputs to the Default Welcome Intent if no-match handling isn’t tightly configured.

Recommendations:

• Add a global no-match handler at the flow level, not just on pages.

• Lower the ML threshold for the Welcome Intent or disable ML matching on it if possible.

• Add a catch-all route or fallback intent on the “Demande nom” page to prevent jumping back.

This issue is more common in multilingual bots — adding explicit fallback handling usually fixes it.

[nikacalupas]

Hi CDConvert,

Welcome to the Google Cloud Community!

The issue likely stems from the Default Welcome Intent being too broadly trained, causing it to mistakenly match invalid inputs like "é(" with high confidence, even though they don't resemble any greetings.

Here are some approaches that you may try to solve your issue: 

• Narrow the training phrases to only include the most basic greetings (e.g., "bonjour", "salut", "hi", "hello").
• Add negative training phrases (e.g., "é(", "John123", etc.) to the Default Welcome Intent. This will help the intent recognize what should not trigger it, reducing false matches.
• Create a dedicated "Invalid Input" page for handling incorrect inputs.
• Route NO_MATCH events directly to this page.
• Use fulfillment code on this page to apply a more robust regex to validate inputs and filter out unwanted data.
• Double-check the regex for your "Last name asking" entity to ensure it’s precise and accurately restricts input to valid last names (letters and specific French characters).

Was this helpful? If so, please accept this answer as “Solution”. If you need additional assistance, reply here within 2 business days and I’ll be happy to help.

[CDConvert]

Hi, I lowered the ML treshold and I don't have this problem anymore. Thank you !