An outbound call center places calls instead of answering them: confirmations, reminders, lead follow-up, renewals, collections, win-back lists and service notifications. What makes one work is rarely the dialer. It is the design of each program behind it, meaning who gets called, when, with what ask, and who works the outcomes afterward.
This guide is for the person who owns that operation. Search the term and you mostly get glossary definitions, agency services pages and job listings, which leaves out the parts you actually have to decide. So: a spec sheet to fill in before the first dial, a way to pick a dialing approach per program, the federal rules that cap how fast you can dial and when, the numbers worth running the program on, and where AI agents fit.
What is an outbound call center?
An outbound call center places calls to customers, prospects or patients rather than waiting for them to call in. The work arrives as lists and triggers instead of a queue, so capacity depends on how many dials the team can make inside legal calling hours, and results are counted per person actually reached rather than per call answered.
It can be your own team, an outsourced provider, automated agents, or some mix of the three. Plenty of centers run both directions at once, with the same supervisors covering an inbound queue in the morning and an outbound list in the afternoon. That is where most of the trouble starts, because the two jobs are managed by different numbers.
How outbound differs from inbound
| Inbound | Outbound | |
|---|---|---|
| How work arrives | Callers queue up on their own schedule | Lists and triggers you choose |
| What sets capacity | An arrival rate you do not control | How many dials fit in the calling window |
| The gating constraint | Answering before people give up | Reaching the right person at all |
| The main quality risk | Hold time | Calling the wrong person, or at a bad time |
| The number it runs on | Service level | Right-party contacts per hour |
The practical consequence is that inbound staffing math does not transfer. For an inbound queue you size the team against an arrival rate with an Erlang model, which is what the call center staffing calculator does. For an outbound program you work backwards from the list: how many records, how long the window is, how many dials it takes to reach one person, and how much time each reached conversation takes.
The programs an outbound call center runs
Most centers run some subset of these, usually having added them one at a time:
- Appointment, delivery and service confirmations.
- Reminders before a visit, a pickup or a service window.
- New lead follow-up, where the value comes from calling while the person still remembers asking.
- List calling to reactivate lapsed customers or re-engage old leads.
- Renewal and payment reminders.
- Collections, which carry their own rules and usually their own team.
- Surveys and post-service checks.
- Service notifications, including outages and recalls.
- Sales and telemarketing, which is the most heavily regulated of the group.
Grouping them as one operation is a mistake, because the rules that reach them differ. A confirmation call to someone who already booked with you sits in a different legal category than a cold sales call, and the pacing limits further down this page apply to telemarketing specifically. That is why the spec sheet starts with why the call exists and what the basis for making it is.
An outbound program spec sheet you can fill in today
Fill in these thirteen lines before you dial anything. Each one corresponds to a decision that, left unmade, turns up later as a complaint, a stalled list or a supervisor retyping notes. The right-hand column shows a filled-in example for a confirmation program.
| Line | What to write | Worked example |
|---|---|---|
| Program name and owner | One named person who answers for results | Pre-visit confirmations, owned by the scheduling supervisor |
| Why the call exists | The business outcome, in one sentence | Catch reschedules two days out instead of discovering them on the day |
| Who gets called | The system of record, the segment, and the trigger | Anyone with a confirmed appointment 48 hours away, from the scheduling system |
| Basis for calling | Where permission comes from for this list | Existing customers confirming a booking they made themselves |
| Suppression rules | Who never gets a call, and why | Already confirmed, opted out, on the do-not-call list, called in the last 24 hours |
| Calling window | Days and hours, in the called party's local time | Monday to Friday, 9:00 to 18:00 local, no weekends |
| Attempt policy | How many attempts, how far apart, and what happens next | Two attempts, 24 hours apart, then a text and a note on the record |
| The ask | The one sentence the call has to land | Are you still able to make Thursday at 2:00? |
| Success outcome | The single result that counts as a win | Confirmed, or moved into an open slot |
| Outcome tags | The complete closed list, agreed before launch | Confirmed, rescheduled, cancelled, no answer, voicemail, wrong number, opted out, transferred |
| Handoff rule | When a person takes over | The caller asks for someone, is upset, or wants an exception to policy |
| Who works the exceptions | A named role and a time of day | Scheduling coordinator, next business morning, from the outcome list |
| Reporting | The numbers reviewed, and how often | Contact rate, reschedules caught, opt-outs, reviewed weekly |
Three lines cause most of the damage when they are left blank. Skip the suppression rules and you will call people who already confirmed. Skip the attempt policy and the list either gets dialed once and dropped or dialed until somebody complains. Skip who works the exceptions and every tagged outcome piles up in a dashboard nobody owns, which is the most common way an outbound program quietly stops producing anything.
How to set up an outbound call center, step by step
- Pick one program rather than standing up a department. Launch, measure and fix one list before you add a second.
- Fix the list source first. Deduplicate, validate the phone numbers, scrub against your suppression lists, and make sure every record carries a time zone, because calling hours are local to the person you are calling and not to your floor.
- Write down the basis for calling that list, in the spec sheet, in plain words. If nobody can state it, that program is not ready.
- Set a calling window narrower than the legal limit. The rules below are a ceiling, not a target, and nothing good comes of an 8:05 p.m. confirmation call.
- Choose the dialing approach that fits the program, using the table below. This follows from the program, not from what your platform happens to support.
- Write the opening and the ask yourself instead of accepting a template, since that is the part of the call the person decides on. Everything after it can be adjusted once you have heard fifty real calls.
- Agree the outcome tags before launch and keep the list closed. Tags that reps or agents can invent on the fly make the reporting useless within a week.
- Decide which numbers you dial from, and ask your voice provider what caller ID authentication your outbound traffic gets. Under the FCC's rules, originating providers authenticate caller ID using the STIR/SHAKEN framework on IP networks, and how your calls are signed and treated feeds the spam labeling your numbers pick up.
- Staff the exception queue, not only the dialing. Every program produces work that a person has to finish, and that work is where the return actually shows up.
- Review a fixed set of numbers weekly, and listen to calls while you do it. A dashboard will tell you the program changed, but only the recordings explain why.
Which dialing approach fits the program?
The approach sets your throughput and your compliance exposure at the same time. Pick it per program rather than per center.
| Approach | How it dials | Fits | The constraint to plan for |
|---|---|---|---|
| Manual | A person dials each number | Low volume, high value, sensitive accounts | Lowest throughput of any option |
| Preview | The rep sees the record, then chooses to dial | Collections, complex accounts, anything needing preparation | Idle time between calls is the cost |
| Power | A fixed ratio of lines dialed per available rep | Mid volume on a clean list | Abandonment climbs with the ratio, and the ceiling is set by rule |
| Predictive | Pacing software predicts when a rep will free up | High volume, long lists, many reps | The abandonment limit below is what you tune against, so it needs real monitoring |
| AI agents | The agent holds the conversation itself | Repeatable programs with one clear outcome | Works per program, not per center, and the handoff rules do the heavy lifting |
A predictive dialer and AI agents are not alternatives to each other in every case. Centers commonly keep a dialer on the programs where a person has to do the talking and move the repeatable lists to agents, which is the part AI outbound calling covers in detail.
The federal rules that cap how you dial
These are the United States federal rules that most directly shape outbound pacing, hours and opt-outs. Each row cites the regulation itself, read on October 9, 2026, rather than a summary of it.
| Rule | What it requires | Which calls it reaches | Source |
|---|---|---|---|
| Abandonment, defined | A call is abandoned if a person answers and the caller does not connect them to a sales representative within two seconds of their completed greeting | Telemarketing calls | 16 CFR 310.4(b)(1)(iv) |
| Abandonment ceiling | No more than three percent of all calls answered by a person, measured over a single campaign if it runs under 30 days, or separately over each successive 30 day period | Telemarketing calls | 16 CFR 310.4(b)(4)(i), 47 CFR 64.1200(a)(7) |
| Ring time | Let an unanswered call ring at least fifteen seconds or four rings before disconnecting it | Telemarketing calls relying on the abandonment safe harbor | 16 CFR 310.4(b)(4)(ii) |
| Recorded fallback | If no representative is available within two seconds of the greeting, promptly play a recorded message giving the seller's name and telephone number | Telemarketing calls relying on the safe harbor | 16 CFR 310.4(b)(4)(iii) |
| Records | Keep records establishing that you met the safe harbor conditions | Telemarketing calls relying on the safe harbor | 16 CFR 310.4(b)(4)(iv) |
| Calling hours | No calls before 8:00 a.m. or after 9:00 p.m., local time at the called person's location | Telephone solicitations to residential subscribers, and calls to a residence without prior consent | 47 CFR 64.1200(c)(1), 16 CFR 310.4(c) |
| Revoking consent | Honor a revocation made by any reasonable method within a reasonable time, and no later than ten business days after you receive it. You may not designate one exclusive way to opt out | Autodialed and artificial or prerecorded voice calls and texts covered by the rule | 47 CFR 64.1200(a)(10) |
| Oral disclosures | Disclose the seller's identity, that the purpose of the call is to sell, and the nature of the goods or services, truthfully, promptly and clearly | Outbound telemarketing calls | 16 CFR 310.4(d) |
Two details in that table are easy to get backwards, and both change how you set pacing. The three percent abandonment figure is measured against calls answered by a person, not against all dials, and it is measured per campaign over a defined window rather than as a daily average across your whole floor. And the words "stop", "quit", "end", "revoke", "opt out", "cancel" and "unsubscribe" are treated as a reasonable means of revoking consent in themselves, which means a reply using any of them counts whether or not it matches your documented process.
Which of these rules reaches a given program depends on what the call is for and where the person is, and states add requirements of their own. This section summarizes public federal regulations and is not legal advice. Have your counsel review the list policy and the script before a campaign runs.
Outbound call center metrics that run the program
Outbound reporting usually measures activity when it should measure reach. Dials are easy to count and tell you almost nothing. These are the ones that change decisions:
| Metric | How to calculate it | What it tells you |
|---|---|---|
| Answer rate | Answered calls divided by dials attempted | Whether your numbers and timing are working at all |
| Contact rate | Right-party contacts divided by dials attempted | The real cost of reaching one person on this list |
| Attempts per contact | Dials attempted divided by right-party contacts | How many tries the list takes, which sets your attempt policy |
| Conversion per contact | Successful outcomes divided by right-party contacts | Whether the script and the ask are working, separate from reach |
| Abandonment rate | Calls abandoned divided by calls answered by a person | Your pacing exposure, calculated the way the rule defines it |
| Right-party contacts per hour | Right-party contacts divided by staffed hours | Throughput, and the only sound basis for a capacity forecast |
| Cost per contact | Fully loaded program cost divided by right-party contacts | The number to compare options with, instead of cost per minute |
| Opt-out rate | Opt-outs divided by right-party contacts | Early warning that the list or the ask is wrong |
The arithmetic matters more than it looks. Take a list where 4,000 dials produce 1,200 answered calls and 800 right-party contacts. The answer rate is 30 percent, the contact rate is 20 percent, and attempts per contact is 5. If 320 of those contacts end in the outcome you wanted, conversion per contact is 40 percent, while conversion per dial is 8 percent. The same program looks broken on one of those numbers and healthy on the other, and only the first one tells you whether the script is working, because the gap between them is reach rather than persuasion. The figures here are placeholders to show the arithmetic. They are not benchmarks and not Telvana results.
Pick four of these, publish them weekly, and leave them alone for a quarter. A metric set that changes every month cannot show a trend. An inbound queue runs on a different set entirely, and the inbound metric set, with the failure mode attached to each number covers that side if you run both directions.
Where AI agents fit in an outbound call center
An AI agent is worth considering for a program where the call follows the same path nearly every time, has one clear outcome, and works fine ending in a transfer to a person. Confirmations and reminders usually qualify, and so does the first contact on a new lead, which is most of what outbound campaigns are. Negotiation, saves and anything resting on an existing relationship do not.
On the Telvana platform, an outbound program runs one of two ways. You upload a contact list and the agent works it inside the calling hours you set, or your own systems place each call through the API when a record changes, so the call goes out while the reason for it is still current. The agent asks your questions, answers from your knowledge base, and books, reschedules or cancels in the scheduling and CRM tools you connect. When a call meets a rule you set, it transfers to a number you choose. Every call is recorded, transcribed, summarized and tagged with its outcome.
The limits are as useful to plan around as the capabilities. Nothing retries a number on its own, so an unreached call lands in your outcome list for a person to decide about. Voicemail is identified from the transcript after the call rather than detected while dialing. Transfers are plain transfers, so whoever picks up should open the call record. The voice agent does not send texts, because texting is a separate channel with its own agent and its own campaigns. Which calls to automate first goes through the selection in more detail, and what AI agents handle in a call center covers the inbound side and the three different products sold under the same name.
Teams running more than one program should give each one its own workspace, with its own agents, numbers and reporting. OakTech Systems runs calling programs for its clients that way on Telvana, with a separate workspace per client program and more than 20 AI agents in production. Across five production customers, Telvana has handled more than 195,000 calls in the last 12 months.
If you are weighing agents against hiring or against a provider, call center outsourcing against AI agents sets out how each is priced and includes a scorecard for deciding one call type at a time.
Frequently asked questions
What does an outbound call center do?
It places calls rather than answering them. The usual programs are appointment and delivery confirmations, reminders, new lead follow-up, list calling to reactivate customers, renewals and payment reminders, collections, surveys, service notifications, and sales calls. The work arrives as lists and triggers rather than as a queue.
What is the difference between an inbound and an outbound call center?
An inbound center answers calls that arrive on the caller's schedule, so it is staffed against an arrival rate and measured on service level and hold time. An outbound center chooses who it calls and when, so it is planned from the list and the calling window and measured on right-party contacts, contact rate and cost per contact.
How many calls should an outbound agent make per hour?
There is no general figure worth quoting, because it depends on the program. Work it out from your own numbers instead: divide the staffed minutes in an hour by the sum of average talk time, after-call work and dialing and waiting time for that program. Then check it against your actual right-party contacts per hour, which is the number that matters.
Do you need consent to make outbound calls?
It depends on what the call is for, how it is placed and where the person is. Federal rules set consent requirements for autodialed calls and for calls using an artificial or prerecorded voice, plus calling hour limits and do-not-call obligations, and states add their own. The list is yours, so the basis for calling it is yours too. Have counsel review your list policy.
What is a good abandonment rate for an outbound campaign?
For telemarketing calls, federal rules cap abandonment at three percent of calls answered by a person, measured over a single campaign under 30 days or over each successive 30 day period (16 CFR 310.4(b)(4)(i)). Treat that as the ceiling rather than the target, and set your pacing well under it, since every abandoned call is a person who picked up and got nothing.
What does it cost to run an outbound program with AI agents?
Telvana starts at $399 a month based on your call volume, plus a one-time setup fee covering the work of building, testing and launching the agent. The pricing page has the detail. To judge it, compare against the fully loaded cost of the staff hours that work the same list today, using cost per right-party contact rather than cost per minute.
Start with the list your team never finishes
Bring the outbound program that keeps slipping, along with its list and its script. On a demo we will show you how Telvana would work it, what it would hand back to your team, and where it would stop and transfer. Book a demo.