Customer support response time: what the benchmarks are worth, and how to get faster

Somebody asks how your support response time compares to everyone else's, and you go looking for the number. Twenty minutes later you've got a table: average first response 12 hours and 10 minutes, nine in ten customers expecting an answer within the hour, most companies never replying at all. It goes on the slide, and the room accepts it, because a figure quoted to the minute sounds like something that was measured.

It wasn't, or at least not in any way you can check. Those three figures come from the same place, and that place is one vendor's mystery-shopper exercise with no published sampling frame.

So this page won't hand you an industry average to compare yourself against, because there isn't an honest one to hand over. What there is turns out to be more useful: two large peer-reviewed measurements of how email reply times are actually shaped, one enormous dataset where a regulator forced the clock into the open, and a clear split between the fixes that move your typical reply and the fixes that move your worst ones. Those are different fixes, and most advice on this topic mixes them together.

Where the response time numbers come from

Every page ranking for this topic is published by a company selling support software, support AI or outsourced support staff, and each of them leads with a benchmark table. Trace the citations back and they converge on a very small number of sources.

Fair warning about the first three rows: I'm not naming or linking the vendors, because we don't send traffic to competitors here, so you're taking my word for the trail. Here's the claim in checkable form: search any of those figures, follow the citations, and they land on one company's blog post about a test it ran on companies it never lists, measuring a window nobody defines. I'm not saying the number is wrong, only that there's no way to find out, and a figure you can't audit shouldn't be steering your staffing.

None of this is fraud. It's marketing doing what marketing does, and a benchmark table is the most linkable thing you can put on a page about response times. But the precision is the tell: "12 hours and 10 minutes" reads as measurement, and with the method missing, the decimal is decoration. The gap is structural, too. The organizations holding response time data at scale are the mail platforms and the helpdesk vendors, and neither gains anything by publishing a method that would let you compare them. Where an audited benchmark does exist, it exists because somebody was compelled to produce it.

Four figures do survive a check, and each one comes with a scope you have to carry with it:

None of those is your industry benchmark, and anybody who lifts one out of its row and puts it on a slide has just done the thing this section is about. The rest of this article unpacks what each one is good for.

What the research actually measured

Two peer-reviewed studies have measured email reply times at a scale no vendor blog comes close to. Neither one is a support queue, and I'll say what each one is before quoting it, because that caveat is the whole difference between this section and the table above.

The first is Evolution of Conversations in the Age of Email Overload, presented at the World Wide Web conference in 2015, which analyzed "more than 2 million users exchanging 16 billion emails over several months" on a consumer webmail service. Personal mail, a decade ago, not a support address. Its headline timing finding: "more than 90% happen within a day of receiving the message, and the most likely reply time is just two minutes." The distribution behind that is the part worth pinning up.

Mean reply time: 1,157 minutes. Median: 47 minutes. Standard deviation: 19,730.

Read those three together and you can see why an industry average for email is close to meaningless. The mean is almost twenty-five times the median, and the standard deviation is seventeen times the mean, because a small number of replies that took weeks drag the average somewhere no actual message lives. Half of all replies landed inside 47 minutes. The average says nineteen hours. Same data. So if somebody quotes you an average response time and can't tell you the median, they haven't told you anything about what customers experienced.

The same study also looked at what happens under load, and the finding is more double-edged than it's usually quoted as being. Its abstract: "as users receive more email messages in a day, they reply to a smaller fraction of them, using shorter replies. However, their responsiveness remains intact, and they may even reply to emails faster." Busy people do get faster, tentatively. They also answer a smaller share of what arrives, which is the failure mode that never shows up in a response time report, because a message nobody answers has no response time.

The second study is closer to a workplace: Characterizing and Predicting Enterprise Email Reply Behavior, from Microsoft Research at SIGIR 2017, built on the public Avocado corpus of 938,035 emails from 279 accounts at a defunct IT company. One company's internal mail, not a customer queue, and its "external" category is a rough stand-in for the mail a support address gets. Two findings transfer usefully.

Mail from outside takes about twice as long to answer and gets answered far less often: "[M]edian reply time to internal emails is 2.1 times faster than to external emails, 66 vs. 135 minutes," and "[p]eople are 3.4 times more likely to reply to internal emails (7.76%) than external emails (2.26%)." Much of that external mail is bulk rather than people waiting on an answer, so 2.26% isn't a service level. It's a map of where unanswered mail piles up.

The finding with a lever attached is timing: "the median reply time for morning and afternoon is less than 1 hour, whereas reply time for mails sent in evening and night is more than 7 hours." Those are two thresholds, not a ratio, so a precise multiplier is somebody's invention. The paper doesn't isolate a cause either, and evening mail also draws a much lower reply rate, so it may be different mail rather than the same mail handled worse. What holds either way: when a message lands shapes how long it waits more than anything happening inside your team.

Responded is not resolved

There is one place where response times to consumer complaints get measured at scale, audited, and published with the method attached. It exists because a regulator built the clock, and it's worth saying plainly how unlike your inbox it is: the US Consumer Financial Protection Bureau routes consumer complaints about financial products to the companies concerned through a government portal, on a formal 15-day response deadline, with a formal mechanism for a company to request more time, and with closure categories the company has to pick from. Banks, not a five-person team. What transfers isn't the percentage, it's the shape.

Its Consumer Response Annual Report for 2024 reports that "[c]ompanies provided a timely response to 99.7% of the approximately 2,829,400 complaints sent to them for review in 2024." Read the definition attached to that figure, though, because the two clocks are not the same clock: a response counts as timely if it arrives within 15 calendar days, or within 60 days where the company asked for an extension.

Then the second sentence: "Approximately 13% of complaints were closed within the initial response period of 15 days and 98% were closed within the final response period of 60 days."

So responding on time is close to universal, and finishing inside the first window happens in about one case in eight. Most of these complaints get acknowledged quickly and then run on for weeks under the extension before they close. When a response deadline is watched closely and a resolution deadline is watched loosely, the watched number goes to nearly a hundred and the work moves into the space behind it.

The version that matters for a team of five: if first response time is the only number anyone reports, you'll get fast first responses. You won't necessarily get finished conversations, and the customer who got a same-hour "looking into this" and then waited nine days doesn't remember your team as fast. Report a closure figure next to the response figure, or you're grading yourself on the easier half of the job.

What customers say they expect

The demand side has one study I can point at without embarrassment, and it's narrower than the vendor surveys pretend to be. Istanbulluoglu's 2017 paper in Computers in Human Behavior, Complaint handling on social media, surveyed people who had actually complained to a company on Facebook or Twitter and got a reply. "Participants in the study stated that they expected companies to reply to their complaints within 1-3 hours on Twitter and within 3-6 hours on Facebook."

The method is stated openly, which is exactly why it's usable: 455 people completed the survey, 422 remained after incomplete responses were removed (222 Facebook, 200 Twitter), the sample was purposive rather than random, and response times were self-reported. It's social media, not email, and expectations there run faster than in a mailbox. Take it as a lower bound on impatience rather than a target.

The finding that matters more than the hours is this one: "both a quicker first response and a quicker conclusive response lead to higher satisfaction with complaint handling." Both clocks, not one. And the paper reports that a speedy response raised satisfaction regardless of what the complainant was after, which cuts against the older assumption that speed stops mattering once the complaint is about redress. Getting to yes slowly still costs you.

As for the five-minute rule that keeps turning up in support advice: it comes from The Short Life of Online Sales Leads, a 2011 Harvard Business Review piece by Oldroyd, McElheran and Elkington. I can verify the title, the authors, the date and the subject, which is how fast sales teams call back inbound web leads. I can't verify the five minutes, because the body is paywalled. So: fifteen years old, about outbound phone calls to prospects, and quoted by people who are themselves quoting a summary.

Read your own distribution

Since there's no external benchmark worth measuring against, the comparison that means anything is against yourself. Three numbers, and only one of them is an average.

Median. Sort your first response times and take the middle one. That's what a normal request actually experiences, and it's the number to quote internally.

90th percentile. The response time that nine out of ten requests beat. This is where complaints come from, and teams are usually shocked at how far it sits from the median. That's the heavy tail from the research showing up in your own data.

Count of conversations with no human reply at all. Not a duration, a count. These have no response time, so they drop out of every average you compute, and they're your worst outcomes. Both studies above found unanswered mail rising exactly when volume did, which is when you're least likely to go looking for it.

Pulling those figures and turning them into a target you publish is a separate job, and setting response targets your coverage can actually support covers the instrumentation and the arithmetic. This page is about what the shape means once you have it.

Fixes, sorted by what they actually move

Most advice here is a flat list of tips. But your median and your tail respond to different things, and when you're two people, knowing which is which is the whole game.

The median movers. Arrival-hour coverage is the cheapest win most teams have never costed. If a third of your mail lands after 17:00 and nobody opens the inbox until 09:30, you've built a sixteen-hour floor under those conversations that no amount of hustle during the day can lift. Shift one person's start by an hour, or ring-fence the first thirty minutes of the day for the queue and nothing else, and set priority rules that hold up on a busy Monday so that half hour touches the right things. Then there's the writing: your third-most-common question doesn't deserve ten minutes of fresh composition every time, and a library of reusable replies turns it into ninety seconds of editing. Deflection is the same idea one step earlier. If one question generates a fifth of your volume, answering it properly somewhere public doesn't speed your queue up, it shrinks it.

The tail movers. Queue order is the lever nobody pulls. A queue sorted by thread start date shows you old conversations that were answered days ago; a queue sorted by the oldest customer message still waiting on a reply shows you the ones actually rotting. Those are rarely the hard cases. They're the ones that got opened on a busy afternoon, mentally filed under "later," and never reopened by anybody. Whether you can sort that way at all depends on what your queue knows: a mailbox knows dates, and a shared inbox tool like TriageFlow knows which conversations have an owner and which are still unclaimed, which is the field that ordering actually needs. The round-trip fix costs nothing and pays the same day. If answering needs the order number, the account email and a screenshot, ask for all three in your first reply, because asking for one, waiting a day, then asking for the next turns a two-touch case into a week. And any handoff that leaves your team needs a name and a date attached, or it becomes the three-week thread nobody was lying about.

The two that don't work. Hiring adds hands, not sequence: if the queue has no order, a third person just misses the same threads in parallel, so fix the order first and then decide whether you're short-staffed. Auto-acknowledgements are worse, because they look like progress. They take your reported first response time to a few seconds and change nothing about when a human showed up.

When these run out and the numbers still won't move, the question stops being process and becomes tooling. Whether your team is big enough to carry help desk software is worth answering on its own terms before you buy anything.

What speed costs

Nobody selling response time dashboards writes this part down, so here it is. Chasing a sub-hour target with two people has a price, and sometimes it's higher than the prize.

The first cost is focus. A one-hour promise means somebody's watching the inbox continuously, which means nobody's doing the work that removes the causes of the tickets. The second is answer quality, and the research points straight at it: in the consumer study, replies sent from phones had a median of 28 minutes against 62 minutes from a desktop, and the same paper found people under load replying faster and shorter. Fast and short is fine for "your order shipped Tuesday." It's bad for a refund policy question, where a half answer generates the clarifying round trip that lengthens exactly the tail you were trying to fix.

So set your speed by message type rather than buying one number for everything. And when every lever above has been pulled and the tail still runs long, the constraint is headcount, not method, at which point asking for more urgency is just asking people to absorb it personally. How to scale a support team works through what changes at each size, and the support team size calculator turns your volume into a figure you can put in front of whoever signs off on hiring. Until then, three people answering properly within a business day will keep more customers than the same three answering in twenty minutes with something you have to write back about.

Frequently asked questions

What is a good customer support response time?

There's no published industry figure that survives a check, so the useful answer is relative to your own data: measure your median and your 90th percentile, then improve the 90th first, because that's the one customers complain about. If you need an outside sanity check, the research on consumer email in 2015 put half of all replies inside 47 minutes and more than 90% inside a day, which makes a target in business hours rather than business days realistic for most small teams. Setting and publishing a target is a separate exercise from measuring one.

What is the average customer support response time?

Nobody credible knows. The circulating figure of roughly 12 hours comes from one vendor's mystery-shopper test with no published sampling frame, and every page repeating it is citing a page that cites it. Even with perfect data the average would be the wrong statistic: in a dataset of 16 billion emails the mean reply time was 1,157 minutes against a median of 47, because a handful of very slow replies drag the mean somewhere no real message sits.

Does an auto-reply count as a response?

Not in any measurement worth keeping, because it takes your reported first response time to a few seconds without changing when a person actually showed up. They're still worth sending if they do real work: name the address the customer reached, say what happens next and roughly when, and say what to do if it's urgent. What they must not do is quietly stop your clock.

Why is our response time good but resolution still slow?

Because those two clocks come apart much further than teams expect, and there's audited proof: in the CFPB's 2024 complaint data, companies met the response deadline on 99.7% of about 2.8 million cases while closing only around 13% inside the initial 15-day period, with most of the rest closing later under an extension. One number is a promise about attention and the other is a promise about outcome, and the first is much cheaper to keep. Report only the first and you'll look excellent while most of your cases are still open.

How can a small team respond faster without hiring?

Start with the two changes that need no extra capacity: cover the hour your mail actually arrives rather than the hour your day starts, and sort the queue by the oldest customer message that still has no reply after it. After that, stop splitting one case across three round trips by asking for everything you need up front, and write down the answers to your three most repeated questions so they cost a minute instead of ten.