Most of your inbox is the same ten questions. The urgent one is buried underneath them, and the quiet account that was about to leave never wrote in at all.

how it runs

For the first weeks it only suggests. Your agents accept or override, and every override is what tunes the next month.

the workflows

01
Ticket routing
Sorted on what the ticket is actually about and how urgent it reads, so the one that matters is not sitting behind forty password resets.
02
Reply drafting
Answers drafted from your resolved archive and your own documentation, cited, for an agent to accept or override.
03
Escalation rules
Defined by impact rather than by keyword, with a three minute ceiling before a human is pulled in on the severe ones.
04
Churn detection
Usage and contact silence watched together, because the account that stops writing in is the one that is leaving.
05
Renewal reminders
Surfaced with the account's actual history attached, early enough that the conversation is a renewal rather than a save.

how it holds up

01
Read the archive first
Six months of resolved tickets tell us what genuinely repeats. That is a better source than anyone's memory of what gets asked.
02
Suggest only, at first
For the first weeks it drafts and routes as a suggestion. Your agents accept or override, and nothing reaches a customer unread.
03
Only the near-perfect categories go live
The question types where it is reliably right start answering alone. The rest stay as suggestions for as long as they need to.
04
Every override is a lesson
The override list drives the next month of tuning. It is also the honest measure of how well this is actually working.

what it will not do

It deflects about 41% at best, which means the majority of your tickets still reach a person. Anyone promising most of your inbox has not read your inbox.
It will not handle angry. Sentiment above a threshold goes straight to a human, because an automated reply to a furious customer is worse than a slow one.
Six months of resolved tickets is the minimum. A young product has not yet generated the repetition this depends on.
Churn detection surfaces the account, it does not save it. The call is still yours to make and still yours to make well.

you supply

running cost

Model cost per ticket handled, on your own provider account. It scales with volume, not with how many agents you have.

Billed by your own providers, not by us. There is no platform fee and no seat count.

InstallSupport?

Scoped per desk, suggest-only for the first weeks, then release. Tell us about your setup and we will confirm it fits before taking anything.