What is average handle time (AHT)?

Average handle time (AHT) is the average duration of a customer contact from the moment an agent picks it up to the moment the work on it is finished. It is calculated as total talk time plus total hold time plus total after-call work, divided by the number of contacts handled. A team whose agents spent 500 hours of combined talk, hold, and wrap time on 5,000 calls has an average handle time of six minutes.

AHT is the metric contact center staffing is built on. Every workforce management calculation, every Erlang staffing model, and most cost-per-contact figures take AHT as an input, which is why it appears on nearly every contact center dashboard and in nearly every operational review.

This page covers what AHT is, what it does and does not include, how to calculate it, what the published benchmarks actually say and how much they disagree, how AHT relates to the metrics around it, where it misleads, and what happens to it when AI starts handling part of the volume.

Average handle time in one sentence

Average handle time is the average total working time an agent spends on one customer contact, including the wrap-up work after the customer has gone.

What average handle time includes and excludes

Included in the standard definition:

• Talk time. The live conversation itself.

• Hold time. Any time the customer spends on hold mid-contact, including transfer waits.

• After-call work. Wrap-up, disposition coding, notes, follow-up tasks, and any system updates the agent completes before taking the next contact. Often abbreviated ACW.

Excluded in the standard definition:

• Queue time before the agent answers. That belongs to average speed of answer, not AHT. A caller who waited eleven minutes and spoke for four has an AHT of four minutes.

• IVR and self-service time before the contact reaches a person.

• Work that happens later. If an agent promises a callback and spends ten minutes on it the next morning, most systems do not attribute that to the original contact, which quietly understates the true cost of handling.

That last exclusion matters more than it looks. AHT measures the contact, not the resolution. Two teams with identical AHT can be doing very different amounts of total work if one of them resolves issues inside the contact and the other defers them.

How to calculate average handle time

The formula is consistent across every source that publishes one, including Call Centre Helper:

AHT = (Total talk time + Total hold time + Total after-call work) / Number of contacts handled

A worked example. Over one week, a team of agents records:

• 320 hours of talk time

• 44 hours of hold time

• 61 hours of after-call work

• 4,250 contacts handled

Total handling time is 425 hours, or 25,500 minutes. Divided by 4,250 contacts, AHT is 6 minutes 0 seconds.

Some teams report AHT without after-call work, which produces a lower and less useful number. Wrap is real work that occupies the agent and has to be staffed for, so excluding it makes the staffing model wrong in the direction that hurts, understating how many people are needed.

The concurrency problem

The formula was designed for voice, where an agent handles one contact at a time and handling time and elapsed time are the same thing. On chat and messaging they are not.

An agent running three concurrent chats over 30 minutes has spent 30 minutes of labor but produced 90 minutes of contact duration. Depending on how the platform counts it, that team's chat AHT can look dramatically worse or dramatically better than voice, and neither reading is comparable to the other.

Any team reporting a single blended AHT across voice, chat, email, and messaging is averaging together numbers that do not mean the same thing. Report AHT by channel, and be explicit about whether the chat figure is elapsed conversation time or agent labor time.

What is a good average handle time?

This is where the published guidance gets uncomfortable, and it is worth being straight about it rather than repeating a number. Two credible sources give figures that do not agree.

The first figure comes from Call Centre Helper and the second from SQM Group. They are 65% apart, and both come from organizations that do this work seriously.

The Call Centre Helper figure carries an important caveat. It is drawn from what users typed into a staffing calculator, which makes it a record of the AHT values people are planning around, not a measurement of the handle times those centers actually achieved. It is the same dataset that produces their widely quoted 83.3% occupancy figure, which is likewise an input, in that case the maximum occupancy cap users chose to set. Planning assumptions and observed results are not the same thing, and the distinction is rarely made when these numbers get repeated.

It is also worth noticing what is not available. Of the pages currently ranking for definitional AHT queries, none publishes a benchmark figure with a traceable source. Several reference an industry-by-industry comparison attributed to a Cornell University report that cannot be located in primary form. A figure of around six minutes ten seconds circulates widely with no attribution at all.

The practical answer is that there is no single good AHT, and a target imported from an industry average is close to meaningless. What a contact center should compare against is its own AHT segmented by contact reason and channel, tracked over time, and read alongside first contact resolution and customer satisfaction. A password reset and a billing dispute have no business sharing a target.

AHT and the metrics around it

Of those, the two that most change how AHT should be read are first contact resolution, which catches the repeat contacts a rushed conversation creates, and deflection rate, which changes the mix of contacts that reach an agent in the first place.

Why average handle time matters

Three reasons, and only the first is about cost.

It is the input staffing runs on. Erlang models take contact volume and AHT and return the number of agents needed to hit a service level. A 30-second error in AHT across a large center changes headcount requirements materially. Of every metric on the dashboard, this is the one that has to be accurate, because everything downstream inherits its error.

It is the largest lever on cost per contact. Agent time dominates the cost of a handled contact, so handling time is most of the equation. This is also why AHT attracts more management pressure than any other operational metric, and why it gets misused.

It is a diagnostic when read by segment. AHT on one contact reason drifting upward is a signal: a policy changed, a knowledge article went stale, a system got slower, a product started failing. Blended AHT hides all of that. Segmented AHT is one of the better early-warning systems a contact center has.

Where average handle time misleads

AHT is the metric most likely to produce the behavior it was meant to prevent.

Rushing does not reduce work, it moves it. An agent under pressure to close fast ends contacts before the issue is fully resolved. The customer calls back. The center now handles two contacts instead of one, and the second one is harder because the customer is annoyed. AHT per contact went down and total handling time went up.

It can be gamed without anyone deciding to game it. Transfers, quick escalations, and premature closures all lower an individual's AHT. None of them lower the organization's workload. When AHT is attached to individual performance reviews, these behaviors emerge without anyone making a conscious decision to work around the metric.

Falling AHT can be a warning sign. If AHT drops while repeat contacts, transfer rate, or escalations rise, the number is describing a problem rather than an improvement. AHT should almost never be read on its own, and it should never be the metric a team is optimized against in isolation.

High AHT is sometimes the correct answer. A complex claim, a service recovery, a high-value customer deciding whether to stay: these should take time. Handling them fast is not efficiency, it is a missed opportunity, and in retail and hospitality it is often the conversation that determines whether the customer comes back.

How to reduce average handle time without damaging service

The reductions that hold are the ones that remove work rather than compress it.

Give agents the context up front. A large share of handle time is spent establishing what is going on: who the customer is, what they bought, what happened last time, what has already been tried. When that arrives with the conversation, the time disappears without anyone hurrying.

Attack after-call work first. Wrap is usually the most compressible component and the least visible one. Automated summarization, automatic disposition coding, and writing notes back to the record without retyping remove minutes per contact without touching the customer conversation at all.

Make the answer findable. Time spent searching for a policy or a procedure is pure handle time with no customer value. Knowledge that surfaces the right answer inside the conversation is one of the more reliable ways to bring AHT down.

Cut transfers and repetition. Every transfer adds hold time, and every repeated explanation adds talk time. Both are handle time the customer experiences as friction, which makes this the rare improvement that lowers the number and improves the experience at the same time.

Segment before you set a target. A single company-wide AHT goal pushes teams handling complex contacts toward exactly the rushing that creates repeat contacts. Targets that vary by contact reason do not.

Average handle time when AI handles part of the volume

This is the part most AHT guidance has not caught up with, and it reverses the usual reading of the metric.

When AI resolves the straightforward contacts, those contacts leave the human queue. Password resets, order status, return windows, store hours: the short ones. What remains for agents is the complex, emotional, multi-step work that was always going to take longer.

So AHT goes up. That is the expected result, and it is not a regression. A team whose AHT climbed from six minutes to nine after deploying AI has usually not got slower. Its agents are handling a harder mix, because the easy volume is being resolved elsewhere.

Reading that rise as a performance problem is the most common measurement mistake in AI deployments, and it can lead teams to pressure agents on precisely the contacts where time is most valuable. The comparison that stays meaningful is AHT within a contact reason, before and after. The blended figure across a changed mix is not comparing like with like.

Two things to track alongside it:

• Total handling hours, which should fall even as AHT per contact rises. That is where the capacity gain shows up.

• AHT by contact reason, which shows whether human handling actually got slower or the mix simply got harder.

Where AI assists rather than replaces the agent, the effect runs the other way. Summarization, drafted responses, retrieved knowledge, and automated wrap reduce handling time on the contacts agents still take. Most teams see both effects at once, which is another reason the blended number is hard to interpret and the segmented one is not.