Plan requirement
| Subscription | Suite Professional or higher, Explore Professional or higher |
| Access | Agent |
The number most quoted and most often misleading. Build it as a median, in business hours, and it starts being useful.
Build it
- Dataset: Support - Tickets.
- Metric: First reply time - Business hours (min), or the calendar-hours First reply time (min) if you answer around the clock.
- Aggregator: MED.
- Attribute: by week.
- Filter: excluding spam, and excluding tickets with no reply yet.
The metric comes in minutes, hours and days, and the unit is part of the name rather than a setting. Pick the one you want to read: a median in minutes with four digits is a number nobody takes in.
Median, not average
A handful of tickets answered after two weeks pull the average far above what most customers experienced. The median says what a typical customer got, which is the question being asked.
Show both if you like; the gap between them is itself informative.
Business hours, not calendar
A ticket arriving at six on Friday and answered at nine on Monday is fourteen hours calendar and one business hour. Only one of those describes the service.
Whichever you use, say which on the report.
Exclude the unanswered
Tickets with no first reply yet have no first reply time, and including them either skews or silently drops them. Filter deliberately so you know which is happening.
Split by channel
Messaging expectations are minutes; email is hours. A combined figure describes neither, and it moves whenever the channel mix moves.
Whether a bot answer counts depends on the channel
In a messaging or voice conversation handled entirely by an AI agent, no SLA applies at all, and an automated response does not stop the clock. A team that adds an AI agent there and expects this number to fall will find it unchanged, and the metric is right.
On the email channel it is the other way round. Where an AI agent is connected to email and sends the first public reply, that reply counts as the first reply: the first reply time target shows as achieved and the metric reads zero minutes or close to it. So an email AI agent transforms this number overnight without anybody replying faster, which is worth saying out loud before somebody presents it as an improvement.
Two related details: a public comment made from another account through ticket sharing does not count towards your first reply time either, and on live chat the reply time is not tracked at all unless reply time SLAs are switched on, which they are not by default.
The phone case that inflates it
An agent takes a call, deals with it and solves the ticket without writing a public comment. That ticket has no first reply time at all. If the customer reopens it a week later and the agent then replies, the clock runs from the original creation, and a call handled well in four minutes reports as a week.
So a team on Talk should expect a tail of very long first reply times that describe nothing. Look for them before you explain the number to anybody.
Watch the tail separately
The median can be excellent while a hundred tickets a month wait a day. Those are the customers who complain, and a percentile report is what finds them.
There is a shortcut for that: the attribute First reply time brackets groups tickets into No replies, 0-1 hrs, 1-8 hrs, 8-24 hrs and >24 hrs. Count tickets by it and the tail is a bar rather than a calculation, with the unanswered ones in their own group instead of quietly missing.
Set it against a target
A time with no target is a number people nod at. With a target, it is a question: are we meeting it, and if not, where.
Comments
0 comments
Article is closed for comments.