How to reduce average handle time without sacrificing quality

Average handle time (AHT) drives contact center costs more than almost any other metric. Every second saved, multiplied by thousands of calls, means fewer agents for the same service level. But the easiest approach – telling agents to talk faster – usually backfires: customers call back, complaints go up, and total workload grows. Let's look at how to cut time by removing waste, not by cutting the help customers get.

What makes up handle time

Handle time is the sum of three parts. Before you change anything, measure each one separately: the waste is rarely where you expect it. Start with after-call work and hold time – that's usually where most of the waste is, and it's the easiest to remove.

• Talk time – the conversation with the customer itself.
• Hold time – the agent puts the customer on hold to look something up or check with a colleague.
• After-call work (ACW) – notes, forms, tickets, and status updates after the call ends.

Hidden time lurks inside all three: switching between systems, asking the customer to repeat a number, waiting for a screen to load.

There's no single good handle time for everyone: it depends on the type of inquiry. Compare your queues against their own history, not against other companies or industries.

What 30 seconds is worth

Handle time feeds directly into the Erlang C formula. Take a line with 120 calls an hour, an average handle time of 240 seconds, and an 80/20 target, open 9:00 AM-9:00 PM seven days a week. It needs 11 agents on the line, 16 people per shift at 30% shrinkage, and 45 employees on staff. At $50 per agent hour and 160 hours a month (you can plug in your own numbers in the AI agent savings calculator), the team costs $360,000 a month, and each of the 43,200 calls costs about $8.33.

• 15 seconds less (225 seconds): still 11 agents, but service level rises from 81% to 87%.
• 30 seconds less (210 seconds): workload drops from 8 to 7 Erlangs, 10 agents on the line are enough, and 15 people per shift.
• Headcount drops from 45 to 42 employees. Three positions are $24,000 a month, or $288,000 a year. Each call gets about 56 cents cheaper.

Plug your own numbers into the Erlang C calculator and the shift calculator. But these savings are real only if quality and first contact resolution (FCR) don't drop. If customers call back, each call is shorter but the total workload is bigger.

Why fast calls that don't resolve the issue cost more. Line workload is the number of calls multiplied by handle time. Cut handle time to 210 seconds, and the workload is 7 Erlangs instead of 8 – 10 agents are enough. But if rushing causes one customer in twenty to call back, calls rise to 126 an hour, the workload to 7.35 Erlangs, and you're back to 11 agents and 45 employees. The savings are gone, and the customer had to call twice.

Cut after-call work

This is the easiest win: most after-call work is manual data entry the system could do instead.

• Replace free-form notes with short fields and inquiry topic codes.
• Fill in the date, time, number, queue, and customer automatically from the call.
• Let AI write a short call summary from the recording, and have the agent just confirm it.
• Create tickets and tasks from the call record in one click.
• Compare after-call work across agents: a wide spread points to training gaps or clunky forms.

Build a knowledge base people actually use

Hold time grows when agents don't know the answer and have to search for it or ask colleagues. A knowledge base helps only if the answer takes seconds to find.

• Write short answers to real questions, not long policy documents.
• Group articles by inquiry topic – the way agents think during a call.
• Keep a single source of truth: outdated copies in chats and files lead to wrong answers.
• Track which searches come up empty – that's a ready-made list of missing articles.
• Assign owners who update articles when products, prices, and policies change.

Add an AI assistant for agents

An AI assistant for agents (agent assist) listens to the call or reads the chat and suggests the next step in real time. No more searching: the right article, script, or answer appears on screen while the customer is still talking.

• Suggested answers and knowledge base articles based on what the customer says.
• Reminders about required phrases and checks, such as identity verification.
• Next-step guidance in complex procedures.
• A call summary at the end – which also shortens after-call work.

The assistant suggests only what's in the knowledge base, so it works best with a clean, up-to-date one. You don't need a huge knowledge base to start: begin with the most common inquiry topics and expand gradually.

Pop the customer record automatically

A lot of talk time goes to identification: asking for a name, account or order number, then typing it into the CRM. A customer record that opens automatically removes that step.

• Identify the customer by phone number and open their CRM record – for example, in Bitrix24 or Kommo (amoCRM) – when the call comes in.
• Show recent orders, open tickets, and past inquiries across all channels.
• Pass along data collected by the IVR or an AI agent so the customer doesn't have to repeat it.
• Let agents work from one screen instead of switching between three systems.

Route calls to the right agent

Transfers are expensive: the second agent starts from scratch, and the customer explains everything again. Skills-based routing (the call goes straight to an agent who knows the topic) and a clear IVR send the call to someone who can resolve it the first time. And if a transfer is still needed, smart transfer shows which colleague with the right skills is available. Fewer transfers mean both shorter handle time and more issues resolved on first contact. A typical benchmark for first contact resolution is 70-80%.

What not to do

• Don't give agents time targets without quality monitoring – they'll start cutting calls short.
• Don't reward short calls if the customer calls back the next day.
• Don't skip required steps – identity verification or legally required disclosures.
• Don't compare handle time across queues with different types of inquiries.

How to measure the results

1. Record your baseline: talk time, hold time, and after-call work by queue and agent for a month.
2. Change one thing at a time, such as the after-call form, and compare with the following month.
3. Alongside handle time, track first contact resolution, repeat calls, and quality scores.
4. Use speech analytics to check that shorter calls still follow the script and resolve the issue.
5. Recalculate staffing with the new handle time in the Erlang C calculator and shift calculator, and convert the difference into positions and dollars – that shows whether the change paid off.

Learn from your best agents, not your fastest. Listen to calls from agents who combine short handle time with high quality scores and few repeat contacts, and compare them with the median. They usually share habits: how they open the customer record, which knowledge base articles they keep handy, how they wrap up the call. Turn those habits into training materials and a standard agent workspace layout.

In CONTACT.WORK, the knowledge base, the AI assistant for agents, and CRM integrations all live in one agent workspace, and reports break handle time down into talk time, hold time, and after-call work. You can see which part of the call each change affected. After a call, AI writes a short summary from the recording – the agent just reviews it and saves it to the record.