AI ticket triage that leaves the hard calls to people
Every email to your support mailbox is read, given one of your categories, a priority and a queue, then turned into a ticket. Anything sensitive or unclear is held for a person, and every decision the AI makes is written down.
What the AI decides for each email
It reads the sender, subject and body, along with the names of any attachments. From that it fills in what a person on the desk would otherwise work out by reading the email themselves.
Category
One of your own categories, so tickets are grouped the way your team already thinks about work.
Priority
From P1 critical to P4 low, judged on urgency and impact. Where the customer has a contract, its tier is taken into account.
Queue
A suggested queue from your list. Your routing rules are checked first and win wherever one applies.
Acknowledgement
A short draft reply confirming the request has been logged, ready to send automatically or held, depending on how you set it up.
When it hands over to a person
Some emails should not be handled by a model alone. These go to a review queue instead of straight through:
- A complaint
- Legal or contractual wording
- A sales enquiry
- Sensitive wording from the customer
- Anything the AI is genuinely unsure about
The acknowledgement for a held ticket waits too, so nothing reaches the customer on an uncertain call. A person signs the ticket off and it carries on as normal.
Your rules come first
Routing rules you set, such as sending one customer's tickets to a named engineer, are checked before the AI's suggestion. If that engineer is on leave, the ticket goes to their cover instead.
Where no rule applies, the ticket goes to the queue the AI suggested. The ticket's history shows the rule that applied, or the queue the AI chose.
Every decision on the record
The AI's full output is stored with each ticket, every time, with no path that skips it. When someone changes a category or priority, that change is recorded next to what the AI said.
So when a ticket went to the wrong place, you can see exactly why, and when the AI and your team disagree, you can see how often.
Common questions
- Which AI model does it use?
- OpenAI's GPT-4o models, through the OpenAI API. OpenAI is listed as a sub-processor in our privacy policy, with processing in the USA under standard contractual clauses.
- How accurate is it?
- We do not publish an accuracy figure, because it depends on your categories and your mail. Every ticket shows what the AI decided and any change a person made, so you can judge it on your own emails during a trial.
- Does it learn from our corrections?
- When someone changes a ticket's category or priority, the change is recorded against the AI's original decision. Those records are kept so they can be used to improve triage in future. Today they do not change how it triages.
- Can we use our own categories and queues?
- Yes. The AI chooses from the categories and queues you set up in the admin area, so a new category is available to it as soon as you add it.
- Does it read attachments?
- It sees the names of attached files, which often say a lot, but not what is inside them. The sender, subject and body of the email are what it classifies on.
- Will it reply to customers without anyone checking?
- Only if you switch automatic acknowledgements on, and never for a ticket held for review. Those wait until a person has looked at them.
- Can it close tickets or raise jobs on its own?
- No. It classifies and routes. Closing a ticket, replying beyond the acknowledgement and raising a Simpro job are all done by people.
Try it on your own emails
The best test of triage is your own mailbox. Connecting it takes under an hour.