Average handle time (AHT), formula and benchmarks

To work it out, add up the minutes spent on your contacts, including any follow-up after each one, and divide by the number of contacts. If 45 support emails and chats took you 360 minutes last week, your average handle time was 8 minutes.

It’s also called average handling time, the name ICMI’s guide to contact centre metrics uses, or just handle time. A contact is one call, chat or email. The number exists for planning, because it turns a count of questions into hours of work.

For a founder who answers support alone, those hours come out of building. Deacon is an AI customer support agent that answers your customers from your own docs in seconds, so the questions it answers cost you no handle time at all, and it has a free plan.

How to calculate average handle time

The formula comes from phone support. NiCE’s glossary and Genesys’s add up talk time, hold time and after-call work, also called wrap-up, which is the notes and follow-up once the caller has gone. ICMI’s guide leaves hold time out.

An email or a chat has no talk time. Brad Cleveland’s glossary counts the time spent handling it plus any work that follows. For an email, that’s reading it, looking up the customer and writing the reply.

The second line is what the first is for. At last week’s 8 minutes, the 400 questions you expect in the month after a launch will take 3,200 minutes, about 53 hours. That’s more than a working week for one person.

Use the mean, not the median, because only the mean multiplied by the number of contacts gives back the hours worked. Gorgias does the same, averaging handle time while it takes the median of response and resolution times to limit extreme values.

What is a good average handle time?

We found no measured handle time for email, and the six minutes often quoted for calls isn’t a current measurement. Zendesk’s guide to AHT says a good one is around six minutes, and Talkdesk’s calls six minutes a reasonable starting point. Neither gives data for it. The nearest measured figures are old or unusual. Cornell’s 2004 report on 470 US call centres found 6.1 minutes, and Call Centre Helper’s 6 minutes 3 seconds comes from 190,702 entries typed into its online staffing calculator.

Measured figures for calls run longer today. These are the published figures we found with a named sample, read on 23 September 2026.

SQM’s own list of call centre standards puts the standard for customer service calls at about 10 minutes. The chat figures are lengths, not handle times. They count nothing after the chat ends, and an agent can run several chats in the same minutes. For email, compare each week with the ones before it rather than with anyone else’s figure.

How helpdesks time an email or a chat

Helpdesk software, where each customer email becomes a ticket, times handling for you. The six below count different minutes, going by their help pages as we read them on 23 September 2026.

One email shows how far apart they are. A customer asks why they were charged twice. You open it and spend 5 minutes on it, then move on to another customer’s email. An hour later you open it again and write the reply in 3 minutes.

Help Scout counts 3 minutes, the last visit. Gorgias and Freshdesk count 8, the two spells with the email in front of you. Intercom counts every minute the conversation sat open and assigned to you, at least 68, and so does Zendesk’s handle time for email.

Chats raise another question, because one person often runs several at once. Genesys says a forecast handle time that includes the overlap “can appear inflated”, and its staffing sums adjust for how many chats agents run at once. Microsoft’s Dynamics 365 counts only the time a chat is the agent’s open tab.

Why handle time rises when self-service works

US service calls grew from 306 seconds in 2012 to 442 in 2023 in ContactBabel’s surveys, a rise it says is “driven in large part by a rise in self-service taking away the easier and shorter calls”. SQM’s list of standards says handle time “has continued to increase every year for the last ten years”, because self-service handles the less complex calls and because of agent turnover. Chats are going the same way, and ContactBabel’s share taking more than five minutes is its highest on record.

A chatbot does the same to a small team’s numbers. Say 25 of your 45 questions last week were quick how-tos that took 4 minutes each, and the other 20 took 13 minutes each. That’s 360 minutes, an average of 8. If a chatbot answers the 25 how-tos from your docs, you handle only the 20 harder ones. Your average rises to 13 minutes, and your week’s support falls from 360 minutes to 260.

So read the average beside your total hours. A rising average is often the sign that the easy questions have stopped reaching you.

Why a handle time target backfires

ICMI’s guide says handle time “is best used for high level purposes and ongoing tactical planning, not as a strict agent standard.” ContactBabel warns that rewarding agents on handling time “could obviously lead to them dropping difficult calls or not answering customers fully”. It calls the measure “a typical ‘old-fashioned’ metric, which is generally going out of favor” because each call “should take as long as is needed”.

Many contact centres target it anyway. In a Call Centre Helper poll of 194 contact centre professionals in October 2016, 72% used handle time for planning, 47% set agents targets on it, and 25% measured agents on it without telling them.

A longer contact that solves the problem saves the second one. MetricNet says handle times rise as first contact resolution rises, because more of the harder problems get solved on the first contact. For a founder, a 12-minute reply that settles the question beats a 4-minute one that brings the customer back tomorrow.

Average handle time vs resolution time, response time and talk time

They’re all measured in minutes or hours, so they get mixed up, but each one times something different. Here’s the double-charge email again, if it arrived at 9:00 and the refund went through the next morning.

In MetricNet’s words, “contact handle time is not the same as mean time to resolve”.

Reducing handle time with Deacon

Much of a founder’s support time goes on the same few questions, typed out again for each customer who didn’t find the page. Deacon answers those in seconds, from your own docs and in the customer’s language, at any hour. When your docs don’t cover a question, it says it doesn’t know instead of guessing, and asks for the visitor’s email so you can reply.

The questions that reach you take less work to pick up. Each conversation shows every message, the page the visitor asked from and whether they’ve chatted before, with any details your site passes, such as their plan, beside it. When a question is too vague, Deacon asks one question back instead of guessing. If three haven’t settled it, it passes the conversation to you with one sentence summing up what the visitor told it. You reply from the dashboard, and your reply appears in their chat if they still have it open. A Translate button turns any conversation into your own language.

Then you answer each missing question once. Every question your docs didn’t cover has an Answer this button. Write the answer, let Tidy up with AI polish it if you like, and press Save & check. Deacon shows you the reply it now gives and how many other open questions your answer covers. Your answer becomes part of what Deacon knows, so it can answer the next person who asks, and the more you answer, the less it needs you.

It also tells you which questions keep coming. Topics group similar questions, count how many people asked and label most groups as a how-to, a bug report, a feature ask and so on. A how-to near the top is an answer worth writing, and a feature ask is something customers want you to build.

Deacon doesn’t time your replies, so it won’t give you a handle time. Keep timing a week of your own beside your total hours, because your average can rise as Deacon takes the quick questions while the hours fall. Human handoff covers how conversations reach you, and you can try it all on the free plan, which needs no card.