Plan requirement
| Subscription | Any plan |
| Access | Admin |
An AI agent can work in several languages, and it can only answer as well as your content in each of them.
How it works
The agent detects the language the customer writes in and replies in it. Your fixed replies are translated by you; generated answers come from articles in that language.
The catch
Generated answers need articles in that language. Where a language is switched on but the knowledge base is only half translated, the agent has little to work from and falls back to saying it cannot help.
Enabling a language does not create coverage; it exposes how much you have.
Check per language before switching on
Take your ten most common questions and confirm each is answered by an article in that language. That check takes twenty minutes and predicts the outcome accurately.
Fixed replies need translating too
Greeting, system replies, handover text. An agent that greets in Dutch and apologises in English reads as broken, and the system replies are exactly the ones nobody remembers to translate.
Detection is not perfect
Short messages, mixed languages and technical terms confuse it. Let the customer switch language explicitly, and make sure a wrong detection is recoverable rather than final.
Handover must match
Passing a Dutch conversation to a group that answers only in English solves nothing. Route by language, or restrict the agent to languages you can actually staff.
Fewer languages, done properly
Two languages the agent handles well beat five where three produce apologies. Add a language when its content is ready, not when the setting is available.
Comments
0 comments
Article is closed for comments.