First response time is the easiest support metric to report and the easiest to misuse. A dashboard says the average is under an hour, leadership is pleased, and meanwhile customers are getting an automatic "we received your message" and then waiting a day for a person. The number looks good because the definition lets it.
This post is about getting the number to mean something: what counts as a first response, how to set targets that differ by channel and urgency, how to measure honestly, and what actually moves it on a team of a few dozen agents.

Decide what counts as a response
Start with the definition, because every later decision depends on it. There are three common ones:
- Any reply, including an automated acknowledgment. Easy to hit, tells you nothing.
- First human reply. The first message a person wrote, even if it only asks a question.
- First useful reply. A message that moves the case: an answer, a specific question, or a clear next step with a time.
Most teams do best tracking the second as their headline number and sampling for the third. An acknowledgment is worth sending, since it tells the customer you have their message, but keep it out of the metric. If you track it, call it something else, such as time to acknowledge.
Write the definition on one page and make sure every report uses it. Two teams arguing about performance are often just using different clocks.
Set targets by channel and by urgency
One target for everything is wrong in both directions. A caller on hold and a customer who emailed on Friday afternoon have different expectations, and your staffing differs too.
Think in tiers:
- Live channels (phone, chat). The target is measured in seconds or a few minutes. If you cannot staff it at some hours, say so on the page and offer a message option.
- Asynchronous channels (email, forms). The target is measured in hours. Decide whether the clock runs on business hours or around the clock, and tell customers which.
- Urgent categories. A customer blocked from the core thing they bought, or a suspected outage, gets a shorter target than a general question, on any channel.
For the urgent tier you need a way to spot it quickly, which is what a good triage category scheme is for. Without it, the shorter target applies to nothing in practice.
Choose numbers from your own history, not from a published benchmark. Pull a few weeks of data, look at the median and at the slowest tenth, and set the target at something the team already hits most of the time. Tighten it once you see how it holds. A target that is missed every day teaches people to ignore it.
Measure the median and the tail
An average hides the cases that hurt. Ten quick replies and one that took two days produce a comfortable mean. Report these instead:
- the median, which shows the typical experience
- the share of contacts that got a first response inside the target
- the slowest group, for example the oldest ten contacts waiting right now
The last one is the most actionable. A queue view sorted by waiting time, checked a few times a day by a lead, catches problems that the weekly report cannot.
Be careful with the clock too. Decide how to treat time spent waiting on the customer, weekends, public holidays and merged contacts. Write the rules down. When the rules are vague, the number gets argued about instead of acted on.
Find out why replies are slow
Slow first response has a limited number of causes, and each has a different fix.
- Not enough people at the busy hours. Look at arrivals by hour and day against who is working. Moving a shift by an hour often does more than adding a head.
- Messages sit unseen. Shared inboxes without an owner are the classic case. Assign by rule, so every contact has a name on it within minutes.
- Agents hold contacts they cannot answer. They read it, cannot solve it, and leave it for later. A clear escalation route, as in an escalation policy, shortens this.
- Answers are hard to find. If looking up the policy takes five minutes, the first reply waits five minutes. A better knowledge base is a speed fix.
- Repeat questions eat the queue. Simple, frequent questions crowd out the ones that need people.
Check which one applies before changing anything. Hiring into a routing problem makes the problem more expensive.
Do not buy speed with empty replies
Pressure on the metric produces predictable gaming: one-line replies that ask for information the customer already gave, or a quick "looking into it" sent to stop the clock. The number improves and the experience does not.
Guard against it in two ways. Sample first replies each week and ask whether the customer would need to write back before anything happens. And pair the response metric with one about outcomes, such as how often a contact needs a second reply to learn the answer. If the first improves while the second gets worse, you are rushing.
What a good fast reply looks like depends on the case, but it usually has one of three shapes: the answer, one specific question that unlocks the answer, or a plain statement of who will act and when. Something like:
Example textThanks for the details. I can see the order is still in the warehouse queue. I've asked the fulfillment team to confirm the ship date, and I'll write back by 3 p.m. Pacific either way.
It names the state, the owner and a time. That is a response; the automatic acknowledgment is not.
Cover the hours nobody is working
Many teams miss their target not at the busy hours but at night and on weekends, where no clock-hour staffing exists. The options are a shifted or follow-the-sun schedule, an on-call rotation for urgent categories only, or an automated first line that answers what it can and takes a message for the rest. After-hours coverage compares these. Whichever you choose, tell customers what to expect, because an honest "you'll hear from us Monday morning" often beats a silent queue.
Where RyteDesk fits
If part of your first-response gap is routine questions at odd hours, an AI assistant can take that first contact. RyteDesk is in early access, built for mid-market support teams. It answers on chat, email and phone from the knowledge and policies you provide, tells callers it is an AI, and hands off to your team or takes a message when it cannot help. You can read every conversation, and test it in a preview that sends nothing to customers. Join the waitlist to try it; it is free to join.