A response-time target is the maximum time a held action may wait before a person picks it up and decides it. Targets are set by type: a held payment might need a decision within two hours, a held letter by the end of the day. They stop human review from quietly becoming a delay the customer pays for.
Why do held actions need targets?
Because a held action does nothing on its own. When an agent holds a refund for a named person, the refund waits until that person decides. Without a target, held items drift to the bottom of busy people's lists, and customers wait. Worse, a growing backlog creates pressure to loosen the agent's bounds so fewer things are held, which removes the very control the hold was there to provide.
Targets also make review capacity visible. If held payments routinely miss their two-hour target, the team has too few certified reviewers for the volume the agent creates. That is a staffing and training problem, and the target is what reveals it.
How are targets set?
Work through the held action types one by one and ask three questions:
- What does waiting cost? A customer locked out of their account loses more from an hour's delay than a supplier whose invoice is due next week.
- What does a good decision need? Some held actions require a call-back or a document check; the target must allow for it.
- Who is available? A target that assumes reviewers who are not on shift is a wish, not a target.
Then set two times: pick-up (someone has started) and decision (it is done). Both are measured from the moment the verdict is filed. The response-time target template for held actions gives a format, and how to set response-time targets for an AI agent review queue walks through the method.
An example
Take a hypothetical payments firm. Its agent holds three main types of action: payments above the customer's limit, new payees, and reopening accounts locked after failed logins. The operations manager sets targets: locked accounts picked up within fifteen minutes and decided within an hour, because customers cannot use the service; payments above limit decided within two hours during business hours; new payees decided by end of day, since a call-back is often needed. A month of data shows locked accounts meeting target and new payees missing it every Monday. The fix is a second reviewer on Monday mornings, not a looser rule.
How is it different from time to competence?
The names sound related but measure very different things. A response-time target measures how quickly work is decided. Time to competence measures how long it takes a person to become able to do a job well, such as a new reviewer reaching the point where they can decide held actions alone. The two connect: a team with a long time to competence will struggle to meet response targets when people leave.
It also differs from incident response time. AI incident response covers how quickly a team reacts when something goes wrong with an agent, which is an emergency measure, not a daily service standard.
Related terms
Held action, review queue, reviewer rota and cover arrangement all connect to response-time targets. The guide to AI agent skills by role shows who sets and who meets them.
Questions people ask
What happens when a held action misses its target?
It should be escalated automatically to the next person on the escalation path. The miss is recorded, and repeated misses are a sign the queue needs more certified reviewers.
Should targets be the same out of hours?
Usually not. Set separate out-of-hours targets for the types that cannot wait, and accept that others will be decided the next working day.