Will it reply to customers without my approval?

Not until you allow it. Approval is the default for every mailbox. You turn on auto-send per mailbox once you have watched the drafts for a while, and you can set a confidence threshold underneath it so anything borderline still waits.

What happens to emails it cannot classify?

They take the catch-all branch, which is normally 'leave it for a human'. Nothing gets silently dropped — that matters most when the message nobody could categorise turns out to be a security report.

Do I have to move off my helpdesk or change email provider?

No. The agent attaches to a shared address on a domain you already own, and whatever sits behind it stays. Teams usually start with the one mailbox that is drowning and leave the rest alone.

Can it summarise a long thread before a teammate picks it up?

Yes. An escalated thread arrives with the category, the details worth having and a suggested reply, rather than forty screens of quoted history.

How do I know it is working?

Watch how often an approved draft went out with no edits, per category. When that number is high for a category, the category is ready to send itself. When it drops, a snippet has gone stale.

← All use cases

support@

Customer support email automation

Most of what lands in support@ has been answered before. An agent on the mailbox reads each message, answers the repeats from policies you wrote, and hands the rest to a person with the thread already summarised.

What it looks like today

The same six questions, every day, typed out fresh each time.
Somebody has to scroll the whole inbox to find out whether anything in it is urgent.
A customer replies to a colleague who is on holiday, and the thread quietly stops moving.

How the support@ workflow runs

1
A message arrives at support@
No forwarding rules and no migration. The address sits on your own domain and the workflow starts the moment mail hits it.
2
It gets a category
Refund, order status, bug, complaint, noise — you pick the list. Everything downstream branches on this one decision, which is why it is the step worth tuning first.
3
Routine ones get answered
The reply is drafted from the snippets you wrote: your refund window, your hours, your actual policy. Where the snippets run out, the agent stops rather than improvising.
4
The rest go to a person
Escalations arrive with the category, the key details and a starting draft, so nobody opens a wall of quoted text cold.

What changes

First replies go out in minutes at 3am, not whenever the next person opens the inbox.
Your team spends its day on the messages that genuinely need a human.
Every automated answer traces back to something you wrote, and every step the workflow took is on the record.

Where you stay in charge

The agent clears the routine queue. Decisions about your customers stay with your team.

Drafts wait in an approval queue until you say otherwise. Auto-send is switched on per mailbox, behind a confidence threshold you choose — and most teams turn it on one category at a time.
Refunds, complaints and anyone who sounds angry go to a person. The agent's job on those is to make the handover fast, not to have a go.
If the classifier is unsure, that doubt is itself a branch: the message lands with your team instead of guessing its way into a reply.
One switch stops a mailbox sending anything at all.

One morning, concretely

A refund request comes in at 2am. It is read, categorised and drafted against your refund policy before anyone wakes up — and it is still sitting in the approval queue, because refunds are one of the categories you kept for yourself.

Questions people ask

Will it reply to customers without my approval?: Not until you allow it. Approval is the default for every mailbox. You turn on auto-send per mailbox once you have watched the drafts for a while, and you can set a confidence threshold underneath it so anything borderline still waits.
What happens to emails it cannot classify?: They take the catch-all branch, which is normally 'leave it for a human'. Nothing gets silently dropped — that matters most when the message nobody could categorise turns out to be a security report.
Do I have to move off my helpdesk or change email provider?: No. The agent attaches to a shared address on a domain you already own, and whatever sits behind it stays. Teams usually start with the one mailbox that is drowning and leave the rest alone.
Can it summarise a long thread before a teammate picks it up?: Yes. An escalated thread arrives with the category, the details worth having and a suggested reply, rather than forty screens of quoted history.
How do I know it is working?: Watch how often an approved draft went out with no edits, per category. When that number is high for a category, the category is ready to send itself. When it drops, a snippet has gone stale.