Skip to content

docs: add AI fallback use case combining Label the incoming message and Open a ticket #645

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
SangeetaMishr wants to merge 6 commits into
base: main
Choose a base branch
from
+39 −1
Open

docs: add AI fallback use case combining Label the incoming message and Open a ticket #645

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
519731e
docs: add use case combining Label the incoming message and Open a ti…
sangeeta1911 Jul 4, 2026
323dfb2
docs: update Last Updated date and drop backticks in new use case sec…
sangeeta1911 Jul 4, 2026
8d95830
docs: fix MDX build error from unescaped angle brackets
sangeeta1911 Jul 5, 2026
585ab80
docs: address review feedback on AI fallback use case
sangeeta1911 Jul 22, 2026
0ef4e0d
Update 16. Open a ticket with a human agent.md
SangeetaMishr Jul 22, 2026
49e0cc1
fix: close br tag to fix MDX build failure
sangeeta1911 Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 39 additions & 1 deletion ...uct Features/03. Flows/2. Flow Actions/16. Open a ticket with a human agent.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
<tr>
<td><b>5 minutes read</b></td>
<td style={{ paddingLeft: '40px' }}><b>Level: Advanced</b></td>
<td style={{ paddingLeft: '40px' }}><b>Last Updated: October 2025</b></td>
<td style={{ paddingLeft: '40px' }}><b>Last Updated: July 2026</b></td>
</tr>
</table>
</h3>
Expand Down Expand Up @@ -175,6 +175,44 @@ Once a ticket is created, users can make edits by utilizing the `Add Remark` opt
<img width="567" alt="image" src="https://github.com/glific/docs/assets/143380171/554f2688-426b-4019-9552-b351ebbe0f6c"/>


## Use Case: Escalating unanswered AI queries to a human agent

If you're using an AI Assistant to answer contact questions (see [File Search Using OpenAI Assistants](https://glific.github.io/docs/docs/Integrations/Filesearch%20Using%20OpenAI%20Assistants/)), some questions will fall outside what it can answer. Label the incoming message and Open a ticket with a human agent can be combined to automatically flag and route these to your team.

1. Capture the contact's question with a Wait for Response node.

2. Call your AI Assistant with a Call a Webhook node, giving it a result name such as `gptresponse`. Like any Call a Webhook node, this has separate `Success` and `Failure` exits.

3. In your AI Assistant's Instructions, ask it to always reply with one clear, consistent phrase when it can't answer, for example "I'm not able to help with that yet." This gives the flow something reliable to check for.

4. On the `Success` exit of the Call a Webhook node - before the assistant's reply is sent back to the contact - add a [Split by a result in the flow](https://glific.github.io/docs/docs/Product%20Features/Flows/Flow%20Actions/Split%20By/Result%20in%20the%20Flow/) node on `@results.gptresponse.message` (the assistant's response - `gptresponse` here is the result name you gave the Call a Webhook node in step 2, so adjust it if you used a different name). Create a branch matching that fallback phrase, or use the Other branch to catch anything that isn't one of the assistant's expected answers.

@coderabbitai coderabbitai Bot Jul 22, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Do not present the Other branch as an equivalent fallback.

If only the fallback phrase is configured, Other receives every response that does not exactly match it—including valid AI answers. Clarify that the exact fallback-value branch should be used for escalation; use Other only when all non-matching responses are intentionally considered unhandled.

🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/4`. Product Features/03. Flows/2. Flow Actions/16.  Open a ticket with a
human agent.md at line 188, Update the guidance around the Split by a result
node on `@results.gptresponse.message` to require an exact branch matching the
configured fallback phrase for escalation. Clarify that the Other branch must
not be presented as equivalent, and should be used only when every non-matching
response is intentionally treated as unhandled.

@tanuprasad530 tanuprasad530 Jul 29, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This wont work as is, as today we can apply split by on a result variable or a contact variable. Now @results.gptresponse.message is not a "direct" result variable. Meaning in the Split by result in this flow node, you would not find @result.gptresponse.messge. You will only find @result.gptresponse and applying the split on that will not work and the response from the AI assistant is stored and can be referenced only using @result.gptresponse.message.

There are two ways to do it:

  1. Simple one: Store @result.gptresponse.message inside another result variable using Save result for this flow. Let's call this result as answer. Now, after this in the dropdown of Split by result in this flow you will find answer and the same validations can be applied on this.
  2. Slightly complicated one: is using Split by expression to check if @results.gptresponse.message contains the fallback string.

I'd suggest you add alternative 1 (the simple one) to solve this.


5. On that branch, add a [Label the incoming message](https://glific.github.io/docs/docs/Product%20Features/Flows/Flow%20Actions/Label%20the%20incoming%20message) node and enter a label such as AI_unhandled. No other Wait for Response node runs between the contact's original question and this point, so the label still lands on that original question rather than on the AI's reply.





6. Right after the label node, add an Open a ticket with a human agent node, following the steps above. Use `@results.<result name from step 1>` as the Ticket Body so the ticket shows the contact's original question, and assign it to the relevant staff member or queue.

<img width="565" height="355" alt="image" src="https://github.com/user-attachments/assets/5be8f7ac-5fea-40c2-af2d-e2394a774c0f" />


<img width="544" height="398" alt="image" src="https://github.com/user-attachments/assets/fdde70e2-dbff-4ee7-a094-7ef59cb1b576" />

<br />


<img width="463" height="463" alt="image" src="https://github.com/user-attachments/assets/966f9482-d9b0-4ace-a627-49c937ec69d0" />

@tanuprasad530 tanuprasad530 Jul 29, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In this screenshot, I can see the flow starts with this node making the reference slightly misleading.
Kindly check and make the necessary changes.




7. On the branches that don't match the fallback phrase, continue as normal and send the assistant's reply (`@results.gptresponse.message`) to the contact.

You'll also want to handle the `Failure` exit of the Call a Webhook node separately - for example, by sending an error message or routing straight to a ticket - since that covers the webhook call itself failing, which is a different case from the assistant successfully responding with "I don't know."

This way, every query the AI can't resolve is labeled AI_unhandled and immediately handed off as a ticket, so you can report on how often the assistant falls back and make sure nothing gets missed.

## Best Practices

- Always assign tickets to the right staff to ensure faster resolution.
Expand Down
Loading