Average Handle Time: How to Calculate, Diagnose, and Improve AHT

What is average handle time?

Average handle time (AHT) is the average amount of agent time required to handle an interaction, from acceptance through the end of after-call work. For a voice call, it usually includes talk time, hold time, and after-call work (ACW), also called wrap-up time.

AHT is an operating metric. It helps teams forecast workload, plan staffing, and find process friction. It does not show whether the customer's issue was resolved, whether the answer was correct, or whether the customer had to contact the company again.

Queue wait before an agent accepts the call is a separate measure, commonly average speed of answer (ASA) or average queue time. This boundary matters. Google Cloud's reporting definition starts handle time when the agent accepts the interaction and ends it after wrap-up. Twilio's queue metrics likewise separate AHT from the time a task waits for acceptance.

The average handle time formula for voice calls

Use totals from the same channel, reporting window, and population of handled calls:

AHT = (total talk time + total hold time + total after-call work) / handled calls

The three time components are:

• Talk time: time the agent and customer are connected, excluding hold and wrap-up.
• Hold time: time the agent places the connected caller on hold.
• After-call work: time used to enter notes, choose a disposition, update systems, schedule follow-up, or complete other work tied to the call.

The denominator is handled calls, not calls offered to the queue. Abandoned calls never reached an agent and belong in queue and abandonment reporting.

Treat this formula as a measurement contract rather than a universal software specification. A platform may include outbound dialing, consultation, transfer, or before-call work. Genesys, for example, includes dialing and contacting time for outbound calls in its AHT definition. Document your included events, then keep the definition stable across dashboards and experiments.

Worked AHT example

Assume one support queue handled 120 voice calls in a day:

The calculation is:

(600 + 60 + 180) / 120 = 7 minutes

The queue's AHT is 7 minutes. Talk contributes 5 minutes per handled call, hold contributes 30 seconds, and after-call work contributes 90 seconds.

Calculate from the total numerator and total handled count. Do not average team or interval averages unless every subgroup has the same number of handled calls. A two-call queue and a 200-call queue should not carry equal weight in a combined result.

Why AHT matters

AHT is useful in four specific ways.

1. Workload planning. Interaction volume multiplied by handle time estimates the work the team must absorb. A small AHT change can materially affect staffing requirements at high volume.
2. Friction detection. Rising hold time may point to slow tools or hard-to-find knowledge. Rising wrap-up may expose duplicate data entry. Rising talk time may reflect a change in contact complexity.
3. Change evaluation. A new routing rule, agent desktop, policy, knowledge source, or AI assistant should change a specific AHT component. Component data helps show whether the change worked as intended.
4. Cost analysis. Handle time contributes to cost per resolved interaction, although it is only one input. Repeats, transfers, rework, platform cost, and staffing model also matter.

These uses depend on consistent segmentation. A blended AHT can improve because the contact mix became easier while the underlying workflow stayed unchanged.

There is no universal good AHT benchmark

A good AHT is long enough to complete the intended outcome correctly and short enough to avoid preventable work. Its value depends on channel, intent, complexity, authentication, customer population, transfer rules, and what the reporting platform includes.

Published figures illustrate the problem. Call Centre Helper labels 6 minutes 3 seconds as an industry standard, but the figure comes from 190,702 values entered into an Erlang staffing calculator. Those planning inputs are not a representative sample of measured calls. SQM Group reports 697 seconds for centers participating in its 2024 benchmark, and defines that figure as talk plus wrap-up. The page does not give a separate AHT sample denominator or say whether hold is included.

Neither number is an agent quota. Use external benchmarks only when the channel, contact type, population, time components, and sample method match your operation.

A safer comparison has three layers:

1. Compare the same queue and intent with its own historical baseline.
2. Compare like-for-like cohorts, such as billing calls with billing calls and authenticated support with authenticated support.
3. Read AHT beside resolution and quality outcomes.

Expect the baseline to move when self-service or voice AI removes short, routine calls. The human queue may then contain a larger share of complex cases, so its AHT can rise even while total workload and customer effort fall.

AHT does not transfer cleanly across channels

The voice formula is easy to understand because one agent usually handles one connected call at a time. Digital channels break that assumption.

• Live chat: an agent may handle several chats concurrently. Elapsed session time overstates labor if it includes time waiting for customer replies. Active handling time and concurrency-adjusted workload may be more useful.
• Email: a ticket can remain open for hours while the agent spends only minutes reading and responding. Track active work separately from first-response and resolution elapsed time.
• Asynchronous messaging: a thread may pause, reopen, and change agents. Define when a handling segment starts and stops, and whether reopened work creates a new interaction.
• Automated voice: the duration of an AI-handled call is an experience and infrastructure measure. It is not human agent handle time. If the agent transfers the call, track automated time, human handle time, and end-to-end resolution time separately.

Never blend these channels into one AHT unless their time definitions represent comparable work.

Diagnose high AHT by component

Begin with the arithmetic. Split AHT into talk, hold, and after-call work, then segment each component by intent, queue, agent tenure, transfer outcome, customer type, and workflow version. Review the distribution as well as the mean. Median and 90th-percentile results can distinguish a broad process problem from a small set of extreme calls.

When talk time is high

Look for:

• contacts that are genuinely more complex;
• repeated authentication or information collection;
• unclear policies and long explanations;
• weak knowledge retrieval;
• slow tool responses while the agent keeps the customer engaged;
• poor conversation structure or coaching gaps;
• failed self-service that sends partially completed work to an agent.

Sample both long and short calls. Very short calls can signal disconnections, premature transfers, or unresolved contacts.

When hold time is high

Look for:

• agents searching across disconnected systems;
• approvals that require a supervisor;
• calls routed to the wrong skill or queue;
• slow downstream tools;
• warm transfers and consultation time;
• missing context from an interactive voice response (IVR) or automated agent.

Hold time is often process latency made visible to the caller. Fixing the system or routing path is usually more durable than telling the agent to return from hold faster.

When after-call work is high

Look for:

• duplicate entry across the customer relationship management (CRM) system and ticketing tools;
• free-form notes that could use a structured template;
• unclear disposition codes;
• follow-up work that belongs in an automated workflow;
• system latency after the call ends;
• tasks being attributed to ACW even though they support several interactions.

When all three components rise

Check contact mix, a new product or policy, staff tenure, tool incidents, upstream automation failures, and reporting changes. A measurement-definition change can create an apparent AHT regression with no change in customer or agent behavior.

Guard quality before you reduce AHT

AHT should sit inside a metric tree. At minimum, pair it with:

• first-contact resolution and repeat-contact rate, using a defined observation window;
• quality assurance, including accuracy, policy adherence, and required disclosures;
• customer satisfaction or effort, segmented by intent rather than read only as a center-wide average;
• transfer and escalation rate, including avoidable transfers;
• error, correction, and rework rate for completed actions; and
• abandonment and queue time, reported separately from handle time.

For automated or assisted voice workflows, add verified task success, correct handoff, tool failure, unauthorized action, and end-to-end customer time. A lower human AHT is a bad result if the automated path adds a failed five-minute attempt before every transfer.

Set guardrails before an experiment. For example: reduce wrap-up time while holding first-contact resolution and QA score steady, with no increase in repeat contact. This makes the improvement testable and keeps speed from becoming the only objective.

How to improve AHT without rushing the customer

Prioritize work that removes friction from the process.

Fix routing and preserve context

Route by intent, account state, language, entitlement, and required skill where the available data supports it. Pass verified context across transfers so the customer does not repeat the same information. Track transfer reason and whether the receiving team could act on the context.

Put approved knowledge in the agent's workflow

Shorten the path from question to supported answer. Search results should be current, scoped to the relevant policy, and easy to cite or inspect. If agents still open several systems to assemble one answer, measure the retrieval steps and tool latency before adding more training.

Remove duplicate work

Prepopulate known fields, connect systems of record, simplify authentication, and eliminate duplicate notes. Use structured dispositions where they produce useful downstream data. Avoid adding mandatory fields merely because they are easy to report.

Train and coach against real call patterns

Use representative conversations from each intent and complexity band. Coach diagnosis, tool use, explanation, control of the call, and escalation judgment. Evaluate both extreme long calls and suspiciously short calls.

Automate after-call work with review controls

Summaries, dispositions, and follow-up drafts can reduce wrap-up time. Keep the source transcript and tool results available for inspection. Require human confirmation for sensitive records or consequential actions, and measure correction rates rather than assuming generated notes are accurate.

Give voice AI a bounded role

Voice AI can improve AHT when it removes a defined source of work. Useful patterns include:

• resolving routine, low-risk intents end to end;
• collecting intent and approved context before a human handoff;
• retrieving relevant knowledge for an agent during the call;
• completing structured tool actions with confirmation; and
• drafting a call summary and disposition for review.

Design each path around a verified outcome and a clean transfer. Track automated completion, repeat contact, transfer reason, total customer time, and the human AHT after transfer. This fuller view prevents a team from claiming an AHT win by moving time into an unmeasured automated step. Our guide to AI in customer experience shows how to connect those operating measures to task success and customer outcomes.

We built Dasha for technical teams running production voice AI. We provide a managed runtime, REST APIs, and a web application for telephony, integrations, testing, monitoring, and call execution. Your application still owns identity, authorization, business rules, data, and success criteria. That boundary lets teams use voice AI for a specific handling-time problem while keeping the operational decision under their control.

This model fits technical teams that want managed live voice while retaining control of tools and policy. Teams seeking a generic no-code contact-center suite or an entirely self-hosted open-source stack should choose an approach aligned with those requirements.

Release changes in measured stages

Start with one intent, queue, or agent cohort. Record the baseline and metric definitions, release the change to a limited group, and compare components and guardrails. Inspect failed and extreme interactions before expanding. Roll back when resolution, policy adherence, or customer effort crosses the agreed threshold.

The goal is efficient resolution. AHT helps locate the work that slows it down. If you are building a bounded voice workflow and want a managed runtime for the live conversation, start a technical evaluation with Dasha.