Translating silently is worse than translating openly
A request in a language nobody on shift reads sits until it is noticed. The customer experiences the delay as indifference, which is the opposite of what happened.
Routing to a speaker is always better where one exists. Translation is the fallback, and treating it as the default quietly degrades every conversation that could have been native.
When translation is used, the agent needs to know. An agent who does not realize they are reading a translation matches its register in their reply. Nuance lost in one direction gets amplified in the other.
The original text must travel with the translation for the same reason. A phrase that reads as rude in translation is often ordinary in the source language. The agent can only tell if they see both.
How the ActionFlow runs on the canvas
Language detection happens before routing, not after, and that ordering is the whole design.
A Webhook trigger or Email node delivers the request. A Text Classifier node identifies the language, which is cheaper and more reliable than having a generation model both detect and act.
A Google Sheets node holds the language-to-team mapping, so support leads can change coverage without opening the flow.
A Switch node routes on the detected language. Where a team covers it, the ticket goes there untouched, and the conversation stays native.
Where no team covers it, a Generate Text node translates the request. The ticket is routed to the fallback team with both versions attached and the translation clearly labeled.
A Generate Object node can be added on the reply path to translate outgoing messages back, again keeping both versions on the ticket.
An HTTP Request or Jira node writes the detected language to the record, which eventually tells you which language deserves a dedicated hire.
Nodes this ActionFlow uses
- Webhook (Trigger): delivers the request. Email (Communication) reads the inbox instead.
- Text Classifier (AI Custom): detects the language before any routing decision.
- Google Sheets (Integration): holds the language-to-team mapping for leads to edit.
- Switch (Control): routes to a native team where one exists.
- Generate Text (AI Core): translates only on the fallback path, with the original kept.
- Jira (Integration): writes the detected language onto the record for coverage planning.
- Slack (Communication): notifies the receiving team with both versions attached.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.