First response time: how to calculate it, with real numbers

Short answer. First response time is the time from a customer's first message to the first reply they get. Work it out per conversation, report the median rather than the average, and count the chats that never got a reply. On our support chat the median went from about 44 hours to 5 seconds when the assistant started replying on its own, but in 6 of 13 chats that first reply was a handover to a person, and those people still waited about 13 hours for me.

First response time is the support number most people quote and fewest people calculate the same way. Here is the plain definition, a way to work it out from any chat export, and our own numbers from before and after the support chat on FanWake started replying on its own.

What it measures

First response time is the time between a customer's first message in a conversation and the first reply they get from you. It is a measure of how long someone waits in silence, so it only counts replies written to that customer. A message that only says "we got your message" is not a reply in that sense.

It says nothing about whether the problem was solved. That is a different number, time to resolution, and a fast first reply that leads nowhere can hide a slow one.

How to calculate it

  1. For each conversation, take the time of the customer's first message.
  2. Take the time of the first reply from your side, by a person or an assistant.
  3. Subtract the first from the second. That is the conversation's first response time.
  4. Across all conversations, report the median and the 90th percentile.
  5. Count the conversations that never got a reply separately. Leaving them out makes the number look better than it is.

Use the median rather than the average, because a few very slow chats move the average a lot. If nine chats get a reply in 2 minutes and one waits 3 days, the average is over 7 hours, while the median is 2 minutes. The 90th percentile is what your slowest ordinary chat looks like.

Most help desks can export conversations as a CSV with a start time and a first reply time. You can paste that file into the first response time calculator, which does these steps in your browser.

Our numbers, before and after

FanWake's support chat ran in two modes in its first weeks. From 31 October to 6 November 2026 the assistant mostly wrote drafts and I sent the replies. From 7 November it replied on its own, answering from the vault or handing the chat to me when it could not.

I sent the repliesThe assistant replied
Dates31 Oct to 6 Nov8 Nov to 15 Nov
Conversations1713
Median first response43.8 h5 s
90th percentile88 h8 s
Replied within an hour4 of 1713 of 13
Never got a reply10

Measured 16 November 2026 from the chat's message log, calendar hours.

A median of almost two days is what support looked like when it depended on me opening the inbox. Most of those customers wrote and then waited, some of them for days.

What the five seconds hide

The second column looks like the problem went away. It did for some people and not for others, and the difference is what the first reply said.

First reply in the 13 chats the assistant handledChats
An answer from the vault4
A handover: a person from the team will reply here6
A reply to a greeting or a thank you3

For the six handovers, my reply came after a median of about 13 hours. That is better than 44 hours, and the customers knew someone would come back to them, which they did not before. But by the definition above, a handover is closer to "we got your message" than to an answer, so the honest version of our number is 5 seconds for an answer in 4 chats out of 13 and about 13 hours for the rest.

Five of those handovers happened because the vault had nothing on the question. Each one is recorded as a gap, so the next customer who asks the same thing gets an answer instead of a handover once I fill it. The first message already takes seconds, so this is where an assistant can still improve first response time: in how many questions it can answer when it replies.

What to report

  • The median first response time, and the 90th percentile next to it.
  • How many conversations never got a reply.
  • If an assistant replies first, how many of those replies were answers, and the time to a person's reply for the rest.
  • Calendar hours, unless you say otherwise.

Questions

What is a good first response time for live chat?

For a live chat with people online, help desk vendors' own benchmarks cluster around 35 to 45 seconds. For a small team that is not online all day, hours are common, and the useful goal is that nobody waits in silence: an honest first reply at once, and a person's reply within the working day.

How do you calculate first response time?

For each conversation, subtract the time of the customer's first message from the time of the first reply from your side. Then take the median and the 90th percentile across conversations, and count the ones with no reply separately.

Should an automatic reply count as a first response?

Not if it only says the message arrived. Count a reply that answers or tells the customer something useful, and if an assistant hands over to a person, measure the time to that person's reply as well.

Should I use business hours or calendar hours?

Calendar hours show what customers live through, so start there. Business hours are fairer for judging a team's effort. If you report business hours, say so next to the number.

Read next

Free tool First response time calculator Drop in a chat or ticket export and see your median first response time, in calendar and working hours, how many customers waited over an hour, and what happens to chats that come in while you are away. Free, in your browser. Handover When should an AI chatbot hand off to a human? The five moments an AI support chatbot should hand a conversation to a person, what the handover message should say, and how often it happened in a live support chat's first week on its own. Buyer's guide AI customer support for SaaS: a checklist for small teams What to look for in an AI support chat when your team is small and answers its own customers: sourced answers, honest handovers, learning from every chat, knowing what to build, and working with your coding agent.