What Is a Good First Response Time? Customer Service Benchmarks

First response time is the support metric customers notice first and teams argue about most, because every benchmark seems to give a different number. Some of the difference is real variation between channels and industries. Much of it is measurement: one study counts calendar hours and another business hours, one reports a mean and another a median, and some quietly count an auto-reply as a response.

This post is for support leads and founders setting a response-time target for a small team. It defines first response time and how it differs from resolution time and handle time, collects published benchmarks by channel with the source, year and sample next to each, explains how to measure your own number so it is comparable, and covers what changes when an AI support agent sends the first reply. PepoChat, which we build, publishes no first response time statistics, so every figure below is third-party.

What is first response time?

First response time is how long a customer waits between sending a request and receiving the first real reply from your team. The important word is "real": an automatic "we got your message" email is not a response.

First response time (FRT), also called first reply time, is the time from the moment a customer's request arrives to the first substantive reply, measured per conversation and then summarised across all of them. Zendesk's article on ticket reply time is a good reference definition: the clock runs from ticket creation to the first public agent comment, it is calculated in both calendar hours and business hours, and most automated or bot actions are left out. Internal notes do not stop the clock because the customer never sees them. When an agent creates a ticket on a customer's behalf, Zendesk stops the clock at the agent's next public comment rather than the first one, so logging a phone call does not count as answering it.

FRT is a per-channel metric. A team that answers chats in 40 seconds and emails in nine hours has two first response times, and averaging them into one number describes neither.

Does first response time affect customer satisfaction?

Yes, in the studies that have tested it, although resolution speed matters as much as the first reply. Two findings are worth knowing, both from social media, where response times are public and easy to study.

A 2017 study in the journal Computers in Human Behavior by Doga Istanbulluoglu used data from consumers who had complained to companies on Facebook or Twitter. Both a quicker first response and a quicker final response led to higher satisfaction with how the complaint was handled, regardless of what the complainant wanted from the company. The participants expected a reply within one to three hours on Twitter and three to six hours on Facebook.

Twitter's own study with Applied Marketing Science, published in December 2015, analysed more than 600,000 customer service tweets sent to major US airlines between March and September 2015. Customers whose tweet was answered in under six minutes said they were willing to pay almost $20 more for that airline in future, against $2.33 when the airline took over an hour. That is a stated willingness to pay from a survey, not observed revenue, and it comes from the platform that benefits from the conclusion. The direction matches the academic work, though: faster first replies leave people better disposed.

How is first response time different from resolution time and handle time?

FRT measures how long the customer waits to hear from you, time to resolution measures how long the problem lasts, and average handle time measures how much agent effort each contact takes. They move independently, and pushing one can hurt another.

Time to resolution is the time from the customer's first message until the issue is marked solved. Average handle time (AHT) is the average working time an agent spends on a contact, including talk or typing time, hold time and the wrap-up after it; it is a cost measure rather than a customer-experience one.

A team that sends a templated holding reply in five minutes and solves the issue in three days has an excellent FRT and a poor customer experience. Track FRT next to time to resolution, never on its own.

What is a good first response time by channel?

Live chat is measured in seconds, phone in about a minute or two, social media in minutes to hours, and email in hours. The published figures disagree even within a channel because they measure different things, so read the "what it measures" column before comparing your numbers to any row.

Email and web forms

The email sources disagree by a factor of two, and the gap is explained by method. Gorgias's April 2026 benchmark uses helpdesk data and medians: an all-industry median first response of 6.3 hours for ecommerce brands at the $10M GMV band, with a 5.5-fold spread between verticals. SuperOffice's benchmark report emailed a simple two-question request to 1,000 companies and found an average reply time of 12 hours 10 minutes; 62% never replied, about 90% sent no acknowledgement, and only 20% answered the questions in full in the first reply. That test is older and measures companies of every size from the outside, like a mystery shopper.

Customer expectations are tighter than either. Jeff Toister's March 2020 survey of more than 3,200 people, sponsored by SuperOffice, found that a one-hour reply would meet the expectations of 88% of consumers, and he recommends one hour as the standard. For most small teams that is a stretch goal for business hours, not a 24/7 promise.

Live chat

LiveChat's Customer Service Report, with data last updated in 2024, gives a 35-second average first response across its platform: 55 seconds in retail, 52 in real estate, 35 in software and financial services. Small teams of one to nine people averaged 31 seconds; teams of 50 to 99 took 1 minute 22 seconds. The same report lists an average queue waiting time of 4 minutes 18 seconds, the time customers spend in a queue before reaching an agent. The report does not say whether that average covers every chat or only queued ones, but either way the queue sits outside the 35 seconds: a queued customer can wait several minutes for a reply the dashboard records as 35 seconds. Freshworks' 2025 benchmark, from 2024 data, says its top-performing 20% of companies averaged under 20 seconds across messaging channels. Both are vendors reporting on their own customers.

Social media

Social benchmarks are mostly expectations. Sprout Social's Q3 2026 pulse survey of 2,250 consumers, fielded in August 2026, found 68% expect a reply within 24 hours, 24% within one to two hours and 14% in under an hour. The academic finding, one to three hours on Twitter, sits inside that range. The best measured figure we found is old: the 22-minute median across US airlines in 2015, a sector with dedicated social teams.

Phone

Phone uses average speed of answer (ASA), the average time a caller waits in the queue before a person picks up. ContactBabel's 2026 US guide puts it at 74 seconds, a third above the pre-pandemic average; its 2026 UK guide puts it at 111 seconds, about four times the pre-pandemic level. The familiar 80/20 target is a convention: a February 2023 Verint article on service levels describes it as an arbitrary default with no research behind it. These figures come from contact centres, not from a founder answering the phone between meetings.

How do you measure first response time correctly?

Measure from the customer's first message to the first substantive reply, per channel, in business hours, and report the median and the 90th percentile with auto-replies excluded. Each of those choices changes the number by more than most improvement projects do.

Use business hours, and label the number

Calendar-hour FRT punishes a team for being closed overnight; business-hour FRT hides the wait the customer actually experienced. Zendesk stores both versions for every ticket, and Gorgias's guidance is that the clock follows business hours. Use business hours to judge staffing, calendar hours to understand the customer's view, and always say which one a number is. Comparing your business-hour median to a vendor's calendar-hour average tells you nothing.

Do not count auto-replies

An automatic acknowledgement is useful, since SuperOffice found most companies do not send one, but it is not a response. Zendesk leaves most automated and bot actions out of its first reply time, and Gorgias's guidance excludes autoresponders explicitly. Apply the same rule to templated holding replies: count only a reply that moves the issue forward.

Report the median and the 90th percentile, not only the mean

The mean is dragged up by the few tickets that sat over a weekend. Take 20 tickets: 16 answered in 30 minutes, three in six hours and one in 50 hours. The mean is 3.8 hours, which describes no actual customer. The median is 30 minutes, the typical experience. The 90th percentile is six hours, meaning one customer in ten waited that long or longer. Report the median as the headline and the 90th percentile as the warning light.

Published figures mix these up. LiveChat, SuperOffice and ContactBabel report averages; Gorgias reports medians. For response times the median is usually lower than the mean, because a few very slow replies pull the mean up, so check which one a benchmark uses before deciding you are slow.

Start the clock when the customer writes

Measure from the customer's first message, not from when a ticket was assigned or a chat was accepted. LiveChat's own report shows why: a 35-second first response sits behind an average queue of more than four minutes.

Count the ones you never answered

An unanswered request has no first response time, so it silently drops out of an average and makes the number look better. Track unanswered conversations as a separate count. SuperOffice's 12-hour average describes only the companies that replied; the 62% that never did are not in it at all.

Segment by channel and by hour

Report each channel separately, then split by hour of day and day of week. The overnight and weekend tail is usually where the 90th percentile lives, and it is a staffing or coverage question, not a speed question.

How can you improve first response time?

Match staffing to when requests arrive, shorten the queue before the reply, set honest expectations when nobody is available, and remove the reasons people write in at all. Faster typing is rarely the fix.

1. Staff to the arrival curve. Count requests by hour and weekday for four weeks and move start times or breaks to match. Most small teams find a morning backlog from overnight messages and a quiet mid-afternoon.
2. Triage before you reply. Tag urgent topics (a failed payment, an outage, a delivery that is due today) and answer those first; everything else is first in, first out.
3. Use saved replies for the top ten questions, edited for each customer, so the first reply is also a useful one.
4. Close channels you cannot staff. Live chat shown as "online" with nobody behind it produces the worst numbers you will ever record. Switch it to email follow-up outside your hours. Live chat vs chatbot covers when each makes sense.
5. Publish your hours and your reply time. "We reply within one business day" turns a four-hour email reply into a pleasant surprise rather than a complaint.
6. Remove the reasons people write in. Better order emails, clearer pricing pages and in-product help reduce volume, which improves every response metric at once; 15 ways to reduce support tickets lists the tactics that work.
7. Let an AI support agent send the first reply. This changes the metric itself, which is the subject of the next section.

How does an AI support agent change first response time?

An AI support agent replies within seconds at any hour, so first response time stops being the bottleneck. What matters next is whether the AI actually resolved the request, and how long people wait for a person when it did not.

An AI reply counts as a first response only if it is substantive by the same rule you apply to people: it answers the question from your own content, or completes the task. An AI agent that replies instantly with a wrong answer has a perfect FRT and a worse customer experience than a correct reply four hours later. Report AI first replies and human first replies separately, and add the metrics that sit behind them: resolution rate, handoff rate, time to first human reply after handoff, and time to resolution for escalated conversations. AI chatbot metrics that matter defines each one with a formula.

The handoff wait becomes the whole waiting experience. If the AI answers in two seconds and the escalation sits until Monday, the customer remembers Monday. That is why the handoff needs as much design as the AI; human handoff for AI chatbots covers how escalation should work.

Here is how this works in PepoChat, including what it does not measure. Replies stream into the widget word by word, so the visitor sees an answer start forming within a second or two; that is a description of how the product works, not a benchmark. The agent escalates to the team inbox when the visitor asks for a person, when it finds nothing relevant in your content and the visitor accepts the offer of help, or when a tool or the model fails, and every workspace member gets an email at that moment, at most one per conversation every six hours. A Team availability setting decides what the agent promises at handoff: that a teammate will pick up shortly, your business hours with an away message and the real time you are back, or, for teams that never watch the inbox live, an email follow-up, with the agent asking for the visitor's email address before it hands over. The human handoff documentation has the details.

Two gaps to plan for. Visitors are not emailed when your team replies in the chat, so if a handoff will wait hours, answer by email to the address the agent collected. And on the Free plan, once the 200 monthly AI replies are used, new conversations go straight to the inbox with an automatic notice; by the rule above, that notice is not a response, so your human first response time applies again for those conversations.

PepoChat's Analytics page shows satisfaction votes and topics, not response times. The team inbox shows each conversation's status, and its CSV export has a created-at and a last-message-at time per conversation but not a time per message, so time to first human reply is something you measure by sampling: for a set of escalated conversations, note when the escalation email arrived and when someone first replied. PepoChat publishes no first response time statistics, and we would distrust any vendor quoting one number for every business.

What to do next

Pull four weeks of conversations, compute your median and 90th-percentile first response time per channel in business hours, and count the requests nobody answered. Compare each channel to the row in the benchmark table that measures the same thing, then write down one target per channel and publish it where customers can see it. If the overnight tail is your problem, 24/7 support without a night shift covers the options. To see how an AI first reply and a clean handoff change those numbers on your own traffic, start a free PepoChat workspace; every feature is on the free plan.