Plan requirement
| Subscription | Any plan |
| Access | Admin |
An AI agent answering email. Higher risk than messaging, and worth doing only where the questions are genuinely repetitive.
Connect it
- Open the AI agent's channel settings.
- Enable email and choose which addresses it covers.
- Scope which tickets it should answer.
- Decide what happens when it cannot.
- Test by sending real messages from outside.
Why email is harder
No clarification. The customer wrote what they wrote and has gone. If the agent misreads it, the reply is wrong and arrives as your company's considered answer, in writing, in their inbox.
On messaging a misunderstanding costs one exchange. By email it costs the interaction.
Scope it narrowly
Use triage or tags so the agent only answers the kinds of email you are confident about. Everything else goes to a person, untouched.
Broad email automation is where deployments get switched off.
Long emails, several questions
Customers write three questions in one message. Decide what should happen when only one is covered: answering one and ignoring two reads as not having read it.
Handing over is usually the right choice there.
Say it is automated
In the email itself, with a clear route to a person. An automated reply presenting itself as a colleague is the version people resent.
Watch replies to replies
When the customer answers the automated email, that continuation needs to reach a person unless you are confident it does not. This is where loops start.
Read the first week in full
Every email the agent sent. It is a finite number at the start and it is the only way to catch the answers that were confidently wrong.
Comments
0 comments
Article is closed for comments.