Yes, and raising priority is among the safest things it does. An agent can move tickets with real urgency signals, such as fraud in progress, a safety risk, a locked account before a deadline, or signs of vulnerability, to the front of the queue. Lowering priority is riskier; a person sets the rules and samples the results.
Why is raising safer than lowering?
The costs are lopsided. If the agent raises a ticket that turns out to be routine, a person spends a minute on it sooner than planned. If it lowers a ticket that was urgent, a customer losing money to a scammer, or reporting a gas smell, waits behind hundreds of password questions.
So design the agent to raise freely and lower cautiously. Missed urgency should count far more heavily than false alarms when the agent is tested.
What counts as urgent?
The business decides, in writing, and the list should be about harm and deadlines, not about how loudly the customer writes. Typical categories:
- Money at risk now: a scam in progress, unrecognised transactions, a payment that must go today.
- Safety: anything suggesting danger to health or property, for businesses where that is possible.
- Locked out before a deadline: payroll, a tax payment, a flight.
- Vulnerability or distress.
- Legal or regulatory deadlines, such as a complaint approaching its response date.
- Repeat contact about the same unresolved problem.
"URGENT!!!" in a subject line is a customer's view, worth reading, but not a category. History matters too: an agent that can read a customer's previous tickets can see that this is the third message about the same fault.
What may the agent do, and what stays with a person?
| Action | The agent | A person |
|---|---|---|
| Raise priority on a listed signal | Raises it and notes the signal | None |
| Raise priority on an unlisted but worrying signal | Raises it and flags for review | Reviews the rule |
| Lower priority on a ticket marked urgent by the customer | Proposes, with reason | Samples or approves |
| Safety signal | Sends approved safety instructions at once and raises | Responds |
| Change the definition of urgent | Never | Owns the list |
| Re-order the whole queue | Applies the rules | Overrides when needed |
Prioritising sits alongside labelling and routing, covered in whether an AI agent can tag and route incoming tickets. The method for finding urgent tickets in a large queue is in how to detect urgent tickets in a busy queue with an AI agent.
What does this look like on a Monday morning?
Take an energy supplier with 2,000 tickets waiting after a weekend. The agent reads every ticket in minutes.
One, sent at 6 am, mentions a smell of gas near the meter. That is a safety signal. The agent replies at once with the approved emergency instructions and number, and puts the ticket at the top with a safety flag. Another, headed "URGENT", complains about a marketing email. The agent leaves it in the normal queue and notes why. A third is from a customer whose prepayment meter has no credit and who mentions a newborn at home; the agent raises it under the vulnerability category. A team leader reviews the agent's raises and its decisions not to raise during the first weeks, then samples them.
How do you stop everything becoming urgent?
Watch the size of the urgent queue. If it keeps growing, the definitions are too broad or the agent is reading them too loosely, and urgent stops meaning anything. Review raised tickets regularly with the team, tighten the rules, and retest. If the agent also drafts replies, prioritisation decides what gets answered first, not what is said; see how to let an AI agent reply to tickets in your helpdesk.
Which controls make it safe?
- A written list of urgency categories owned by a named person.
- Raising allowed at depth 3; lowering customer-marked urgency held or sampled.
- Safety and crisis signals linked to approved immediate replies.
- Testing weighted towards missed urgency.
- Each priority change filed with the signal that caused it.
The broader stance on knowing when to stop is in the guide to AI customer support agents.
Questions people ask
Can customers game the agent by writing urgent messages?
They can try. Basing urgency on signals of harm and deadlines, rather than tone, makes that harder.
Should the agent tell customers their ticket was prioritised?
It can say when they can expect a reply. Telling some customers they are high priority invites everyone to ask for the same.
What if two urgent categories conflict?
Safety and money at risk usually come first. The business sets the order, and the agent applies it.