It's 4 p.m., and a contact center lead is reading the dialer report. Agents were logged in all day, yet few real conversations happened. Average outbound calls per hour in a call center depend on dialing mode, while AI capacity depends on concurrent lines.

What humans really dial per hour

Plan on about 9 dials per agent-hour with manual dialing, and 26 to 39 with a dialer. One published benchmark is ContactBabel's US contact center survey. It asked operations how many attempts and live contacts each agent gets per hour.

Dialing mode Dials per agent-hour Live contacts per agent-hour
Predictive 39 14
Preview 26 12
Progressive 27 11
Manual (listed as "none") 9 6

This survey covers US centres only and is from the 2010 to 2011 edition. Use it for the ratios, not as a current average. Predictive dialing gave more than four times the attempts of manual dialing. Check the ratios against your own dialer report before you plan around them.

Talk time and after-call work set the ceiling. Say a manual call takes 30 seconds to dial and ring, 4 minutes to talk and 1 minute of after-call notes. That is 5.5 minutes per call, or about 11 an hour at best. This is a worked example, not a survey figure. A dialer removes the dialing and the ring wait. It cannot shorten the conversation or the notes.

Occupancy caps the hour. Occupancy is the share of logged-in time spent on talk plus after-call work. As a rule of thumb, many centres aim for roughly 75 to 85%, because agents running far above that tend to burn out. Check this range against your own workforce data. In a 60-minute hour, that is 45 to 51 minutes of call work.

Treat every number here as a range. Sector, list quality and call length all move it. A two-minute payment reminder and an eight-minute sales discovery call sit at opposite ends.

Dials are not conversations

Only a fraction of dials turn into live conversations. Think of a funnel with four steps: dialed, answered, live conversation, outcome. Each step loses calls. Voicemail, busy lines and wrong numbers sit between the first two steps. Hang-ups in the first seconds sit between the next two.

The ContactBabel survey shows the loss in its own numbers. Predictive dialing reached 14 live contacts from 39 dials, about one in three. Manual dialing reached 6 from 9, about two in three. A faster mode gives more dials and a thinner funnel. Judge your hour by the live contacts, not the dial count.

Timing and retries move the answer rate, and your list decides by how much. In one test, a consumer lender ran voice calls at different hours. Mornings and evenings answered better than midday, so it moved its start window to the evening. Calling a contact three to four times paid off, because many people picked up on a later attempt. About half of the answered calls became real conversations.

Count carefully. In Insight7's voice campaigns, which are in beta, a call where only the AI spoke is not counted as reached. A carrier's temporary refusal does not use up a contact's retry. Campaigns also set calling hours and retry rules, so you can test windows and attempt counts on your own list.

AI capacity is set by concurrency

Concurrency is the number of calls live at the same moment. A human agent makes one call at a time, so a team's capacity is its headcount times its hours. An AI voice agent can run many calls in parallel. Its capacity is the number of lines open.

Use this formula for dials per hour:

  • Calls per hour = concurrent lines × (3,600 ÷ seconds per attempt).
  • Seconds per attempt = dial and ring time + average talk time, blended across answered and unanswered attempts.
  • Then multiply by your answer rate to get answered calls.

As a worst case, take 10 lines where every attempt is answered: 30 seconds of ringing plus 3 minutes of talk is 210 seconds. That gives about 170 attempts an hour. Real lists have unanswered attempts that end after ringing, so the blended time is shorter and the count is higher. The table below uses the blend. The same logic works in reverse, as a basic queueing rule: lines needed = attempts per second × seconds per attempt.

Several layers cap the line count:

  • Telephony channels or SIP trunks
  • Speech and voice provider rate limits
  • Language model throughput
  • Carrier limits and per-number call caps
  • Your latency budget

Each voice platform sets its own concurrency limit, often tied to the plan, and some price bursts above it differently. Ask each vendor for your concurrency cap in writing.

Human calling capacity grows only by adding people. An AI line is a setting.

Twenty agents match about 25 AI lines

Twenty agents on a dialer and 25 AI lines land close together in raw dials. The gap opens as lines are added. The table uses the predictive-mode figures from the ContactBabel survey: 39 dials and 14 live contacts per agent-hour. All AI figures are illustrative.

Metric (per hour) Human agent Team of 20 AI, 5 lines AI, 25 lines AI, 100 lines
Dials 39 780 190 950 3,800
Answered calls (live contacts for humans) 14 280 68 342 1,368
Working hours per day About 8, with breaks About 8 Campaign calling window Campaign calling window Campaign calling window
Ramp-up time Hiring and training Hiring and training Campaign setup and test Campaign setup and test Campaign setup and test
Cost driver Paid hours Headcount Lines and plan Lines and plan Lines and plan

Assumptions, all illustrative:

  • Answer rate is 36%, the survey's 14 live contacts from 39 dials.
  • An unanswered attempt rings for 30 seconds.
  • An answered call takes 30 seconds of ringing plus 3 minutes of talk. The blended attempt is 0.36 × 210 + 0.64 × 30, about 95 seconds.
  • Optional step: to estimate real conversations, apply your own ratio of conversations to answered calls.

Over an 8-hour window: 20 agents make about 6,240 dials and 2,240 answered calls. 25 AI lines make about 7,600 dials and 2,700 answered calls. That is 1.2 times the dials. Calling windows and retry rules still apply.

What moves the number (25 lines, base 950 dials an hour):

  • Talk time: at 5 minutes, an answered attempt is 330 seconds. The blend is 0.36 × 330 + 0.64 × 30 = 138 seconds. Dials drop to about 650.
  • Answer rate: at 20%, the blend is 0.2 × 210 + 0.8 × 30 = 66 seconds. Dials rise to about 1,360, but answered calls fall to about 270.
  • Ring time: at 20 seconds, an answered attempt is 200 seconds and an unanswered one is 20. The blend is about 85 seconds, which gives about 1,060 dials.
  • Lines and retries: more lines raise dials in a straight line. Retries add attempts but not new people.

Answer rate and conversation quality still decide the outcome.

Related reading: How to A/B Test AI Voice Agent Scripts, Voices and Offers on the Same Customer Segment, AI Outbound Campaign Report Template for Sharing Results With Leadership and 12 AI Voice Agent KPIs Growth and Customer Success Teams Should Track.

Questions people also ask

How many outbound calls per hour is normal for a call center agent?

Plan on about 9 manual dials per agent-hour. Predictive dialing reaches 39, per ContactBabel's survey. Live contacts run from 6 to 14 an hour.

What is a good outbound calls per hour target for a sales team?

Set the target on live conversations, not dials. At 75 to 85% occupancy, an agent has 45 to 51 call minutes an hour. Eight-minute discovery calls then give about six conversations an hour.

How many calls can an AI voice agent make at once?

As many as your plan and provider allow. Limits vary by vendor and often depend on the subscription tier. Some vendors let traffic burst above the cap at a higher rate. Get your cap in writing.

How many calls per hour can an AI voice agent handle?

Multiply lines by attempts per line. At a 95-second blended attempt, one line makes about 38 attempts an hour, so 25 lines make about 950. That is illustrative. About a third would be answered.

What is the difference between a power dialer and a predictive dialer?

A power dialer calls when an agent is free. A predictive dialer forecasts when agents free up and dials ahead. The survey has no power mode, so progressive is the closest match: 27 dials an hour against 39 for predictive.

What is a good answer rate for outbound calls?

List quality decides it. In the survey, predictive dialing turned about one in three dials into live contacts, and manual dialing about two in three. Benchmark against your own list.

Do AI voice agents need consent to make outbound calls?

In the US, the FCC's 2024 ruling applies TCPA consent rules to AI voices. In the EU, Article 50 requires the AI to say it is AI. Check the rules in each region you call. Insight7's beta campaigns include consent marking.

Measure the hour in conversations

Compare humans and AI on the same list, window and scorecard. Human capacity is hours, and AI capacity is lines. Rule of thumb: a dial is a try, and only a conversation is a result.

Book a demo of Insight7