AHT Meaning: Average Handle Time Formula and How to Improve

AHT stands for average handle time: the average time an agent spends on one customer interaction, counting talk time, hold time and after-call work. In a call center, it is the metric that turns call volume into staffing, because every forecast multiplies expected calls by expected handle time.

AHT starts when the agent picks up the call and ends when the agent finishes the wrap-up work for that call. Time the caller spent waiting in queue is not part of it. That belongs to speed of answer.

How do you calculate average handle time?

Add total talk time, total hold time and total after-call work for a period, then divide by the number of calls handled in that period. Five9 publishes the same formula in its contact center FAQ.

Average handle time formula: AHT = (Total talk time + Total hold time + Total after-call work) ÷ Calls handled

1,000 calls with 4,500 min of talk, 500 min of hold and 1,000 min of after-call work = 6,000 min. 6,000 min ÷ 1,000 calls = an AHT of 6 minutes per call.

Count only calls an agent actually handled, and keep wrap-up time in.

A few rules keep the number honest:

• Count handled calls only. Abandoned calls never reached an agent, so they belong in abandonment rate, not AHT.
• Keep after-call work in. Notes, CRM updates and follow-up emails are part of the cost of the call. An AHT without wrap-up flatters the team.
• Look at the median and the 90th percentile too. A handful of 40-minute escalations can drag the average up while most calls are short.
• Split by call reason. A billing question and a claims call have different natural lengths. One blended number hides where the time goes.

What is a good average handle time?

There is no single good AHT, because it depends on what callers need. A password reset and an insurance claim should not share a target. The most useful public reference is SQM Group's annual benchmark of North American call centers: in its 2024 results, the average AHT across participating centers was 697 seconds (about 11.6 minutes), including talk and wrap-up time, up 18% from the year before.

Treat that as context, not a target. The better benchmark is your own baseline by call reason, tracked week over week, so you can see whether a change to process, training or routing moved it.

Set AHT targets by call reason, not one number for the whole floor. A billing question and a claims call have different natural lengths. Track each call reason against its own baseline week over week.

Why can pushing AHT down backfire?

AHT measures effort, not outcome. If agents rush, callers hang up without an answer and call back, and the second call costs more than the minutes you saved. That is why AHT should always sit next to first call resolution benchmarks and a repeat-call rate on the same report.

SQM's 2024 data shows the link from the other direction. Calls lasting 1 to 3 minutes had a 3% higher FCR rate than calls of 5 to 10 minutes, and calls around 15 minutes had a 5% lower FCR rate than 5 to 10 minute calls. Long calls are usually long because the problem is hard, so the fix is to remove the reasons for complexity, not to cut agents off.

How does AHT change when an AI voice agent handles part of the volume?

Your human AHT will probably go up, and that is usually good news. Teams that put an AI agent in front of inbound customer service usually start with the short, repetitive calls: order status, appointment changes, balance checks. The calls left for people are the longer, harder ones, so the average for the human team rises even when every agent is working exactly as before.

Human AHT after an AI agent takes the simple calls: Human AHT = Human handle minutes ÷ Calls handled by people

Before: (3,000 simple calls × 3 min) + (7,000 complex calls × 8 min) = 65,000 min ÷ 10,000 calls = 6.5 min AHT. After the AI agent handles the 3,000 simple calls: 56,000 min ÷ 7,000 calls = 8.0 min AHT, while human minutes fall by 14%.

Report total human handle minutes next to AHT, because that is the number that drives staffing cost.

So report three numbers instead of one after launch:

• AI handle time: the average and p90 duration of calls the voice agent handles.
• Human AHT: talk, hold and wrap-up for calls people handle, split by whether the call came through an AI transfer.
• Total human handle minutes: the number that drives staffing cost. It should fall even if human AHT rises.

The second effect runs the other way. When the AI agent collects details and verifies the caller before a transfer, the human on the other end starts with context instead of a blank screen. In the Matic Insurance case study, the average time to collect quote data dropped from 9 minutes to about 6 after the team had the AI agent gather the details first and hand only eligible leads to licensed agents.

How do you reduce average handle time without hurting resolution?

Most handle time is lost to searching, repeating and re-keying, not to talking. These changes cut it without pressuring agents:

• Route on the first try by confirming the caller's intent before any transfer.
• Verify identity and collect account details before a person joins the call.
• Pass a short summary to the receiving agent so nobody asks the caller to repeat themselves.
• Auto-write call summaries and fields to the CRM to shrink after-call work.
• Put the answers to your most common questions in one knowledge base agents and AI share.
• Review your longest calls each week and fix the most common cause.

On transferred calls, the handoff itself matters most. A warm transfer that briefs the human agent before the caller joins removes the "can you tell me again why you are calling" minute that every transferred caller dreads. In Retell AI, a warm transfer can play a private whisper message to the receiving agent and a three-way introduction once both sides are connected.

How do you measure handle time for a voice agent in Retell?

In Retell, open the Analytics tab and add a chart with the Call Duration metric. Duration supports avg, median, p90, sum, min and max, so you can plot the average and the long tail on one chart. Break it down by agent version to see whether a prompt change made calls longer, as described in the analytics dashboard documentation.

To get AHT by call reason, add a selector field such as issue_category under Post call extraction in the agent editor. Retell fills it from the transcript after each call, and any post-call analysis field can be used as a breakdown on a chart. Filter by the call_transfer disconnection reason to separate calls the agent finished from calls it handed off. If you run AI QA, its Transfer Wait Time metric shows how long callers waited for the human leg to connect, which is time your customer feels even though it sits outside human AHT.

The same charts belong on a wider scorecard. The voice agent KPI dashboard template shows where handle time fits next to resolution, containment and cost.

The fastest way to see what AHT looks like with an AI agent in front of your team is to measure it on your own calls. Retell runs pilots on real call traffic before any contract, so you can compare AI handle time, human AHT and total handle minutes against your current baseline. Talk to the Retell team about a pilot.