An auto dialer is software that places outbound calls from a list so your agents don't dial by hand. It loads the list, drops numbers it shouldn't call, holds each record until it's inside the allowed calling hours, dials, listens to what picks up, and passes live people to an agent while it handles busy signals, no answers and voicemail itself. Then it saves the outcome and decides when to try the number again.
What changes from one dialer setup to the next is pacing, meaning how many numbers it dials for each agent who is free. That one setting decides how long your agents wait between calls and how many people answer to silence. The table and the arithmetic below cover both, for an in-house team setting up or auditing a campaign.
How does an auto dialer work, step by step?
Every record goes down the same path: load, filter, check the calling window, pace, dial, analyze the answer, connect, record, disposition and recycle. Each step has a setting that someone on your team owns. When a campaign misbehaves, check these settings before you blame the dialer.
| Step | What the dialer does | The setting you control | What goes wrong when it's off |
|---|---|---|---|
| 1. Load the list | Imports records with numbers, names and the fields the agent needs | Field mapping, deduplication, which number to try first | Duplicates call the same person twice; a record with no time zone breaks step 3 |
| 2. Filter | Removes numbers that shouldn't be called | Your internal do-not-call list, the national registry, consent fields, records already resolved | Someone who asked not to be called hears from you again |
| 3. Calling window | Holds each record until it's inside the allowed hours | Days and hours per campaign, where the time zone comes from, holidays | A 7:40 a.m. call because the window followed your floor's clock instead of the customer's |
| 4. Pace | Decides how many numbers to dial right now | Dialing mode and, for power or predictive, the abandonment target | Too slow and agents sit idle; too fast and people pick up to silence |
| 5. Dial and ring | Places the calls and waits for an answer | Ring timeout, outbound caller ID | A short timeout hangs up on people still reaching for the phone |
| 6. Analyze the answer | Sorts busy, no answer, disconnected, fax, machine and live person | Answering machine detection on or off, and its timing | A person gets mistaken for a machine and dropped, or detection takes so long the person hangs up |
| 7. Connect | Bridges a live answer to a free agent with the record on screen | Which agents or queue take the campaign | No agent is free and the call becomes an abandoned call |
| 8. Talk and record | The agent handles the call while the dialer records it | Recording, script, screen pop | The agent asks questions the record already answers |
| 9. Disposition | Saves the outcome: system results for calls that never connected, the agent's wrap-up code for the rest | The code list, and which codes close a record | Vague codes, so nobody can tell which records to call back |
| 10. Recycle | Puts unfinished records back in line | Attempts per record, time between attempts, which outcomes come back | One number rung five times in a morning, or a reachable one never tried again |
Federal rules sit on steps 2 and 3. For telephone solicitations to residential numbers, the FCC's do-not-call safe harbor expects you to scrub against "a version of the national do-not-call registry obtained from the administrator of the registry no more than 31 days prior to the date any call is made" (47 CFR 64.1200(c)(2)(i)(D)). The calling window is set by where the person you're calling is: no solicitations "before the hour of 8 a.m. or after 9 p.m. (local time at the called party's location)" (47 CFR 64.1200(c)(1)). The FCC also runs a reassigned numbers database that answers whether a number has been permanently disconnected since the date you got consent, and a caller who checked it and got a wrong answer has a defense under paragraph (m) of the same rule. The other federal rules on pacing, ring time and opt-outs are in the table in our outbound call center guide.
Step 9 is the one teams tend to underbuild. A disposition code list that separates "not reached" from "reached, call back later" from "done" is what lets step 10 do its job.
What are the four auto dialer pacing modes?
Preview, progressive, power and predictive differ in how many numbers the dialer calls for each agent, and when. Preview waits for the agent to choose. Progressive dials one number per free agent. Power dials several per free agent. Predictive dials before agents are free, using a forecast of when they will be.
| Mode | Lines dialed | What triggers the dial | Abandoned calls | Agent idle time |
|---|---|---|---|---|
| Preview | One, placed by the agent after reading the record | The agent | None | Highest |
| Progressive | One per free agent | An agent becoming free | Rare | High |
| Power | Several per free agent, adjusted to an abandonment target | An agent becoming free | Some | Lower |
| Predictive | Ahead of agents finishing their current calls | A forecast of when agents free up | Some, managed by the forecast | Lowest |
Genesys Cloud's documentation describes all four plainly. Preview "presents information about the contact to the agent before dialing," and the agent places the call. In progressive mode "the system automatically dials one contact for each available agent." Power "dials multiple phone numbers for every idle agent," speeding up while abandonment risk is low and slowing once "it reaches the configured call abandonment threshold." Predictive "can launch the call before the agent is available, so that by the time a customer answers the call, an agent is just becoming available" (Dialing modes, Genesys Cloud Resource Center, read November 17, 2026).
The same page has two details to carry into your setup. Progressive dialing still abandons calls sometimes: if a blended agent counted as available takes an inbound call before the outbound one connects, "that call becomes abandoned," and Genesys offers a precise mode that reserves agents before dialing for that reason. And it recommends at least 10 to 15 agents on a campaign in power mode and at least 15 in predictive, because the pacing math needs a large enough sample. Counting those agents on a floor shared between campaigns is its own exercise, worked through in predictive dialer vs AI agents.
Why does dialing more lines per agent abandon more calls?
Dialing several numbers for one free agent is a bet that exactly one person answers. Sometimes nobody does, and the agent waits for the next round. Sometimes two or three people answer, and everyone after the first needs a different agent within two seconds. If none is free, the call counts as abandoned.
The federal definition for telemarketing is narrow. A call is abandoned when the person answers and isn't connected to a sales representative "within two (2) seconds of the person's completed greeting" (16 CFR 310.4(b)(1)(iv)), and the safe harbor caps abandonment at three percent of calls answered by a person, measured per campaign (16 CFR 310.4(b)(4)(i)).
You can estimate your exposure from your own live-answer rate, which is the share of last month's dials that reached a person (your dispositions have it). The table assumes each number answers independently at that rate, and that one agent is free with nobody else about to finish, which is roughly what a small team sees.
| Lines dialed per free agent | Chance nobody answers, at a 20% live-answer rate | Answered calls left with no agent, at 10% | At 20% | At 30% |
|---|---|---|---|---|
| 1 | 80% | 0% | 0% | 0% |
| 2 | 64% | 5% | 10% | 15% |
| 3 | 51% | 10% | 19% | 27% |
| 4 | 41% | 14% | 26% | 37% |
Start with the second row. Two lines per agent at a 10 percent live-answer rate leaves 5 percent of answered calls with no agent when nobody else is free, which is already over the three percent cap. A ratio above one only stays under the cap when other agents finish their calls at the right moment to take the extra answers, and you can only count on that with a large pool. This is planning arithmetic, and your dialer's own pacing will behave differently, but it explains why vendors set agent floors for power and predictive modes and why a team of four usually belongs in progressive.
To run it on your own numbers, with r lines per agent and a live-answer rate p: the chance nobody answers is (1 - p)^r, and the share of answered calls left without the agent is (r × p - 1 + (1 - p)^r) ÷ (r × p).
How long does answering machine detection take?
Answering machine detection listens to the first moments after someone picks up and decides from the pattern of speech and silence whether it's a person or a recording. That listening takes time, and on a telemarketing call the time comes out of the same two seconds you have to connect a live person to an agent.
Twilio's documentation is specific about how its engine works. It "isolates human speech audio and measures periods between speech and silence in the greeting." By default it waits for 1,200 milliseconds of silence after someone stops talking before it treats the speech as finished, and raising that setting delays human detection by the same amount. It also names errors in both directions: a short two-second voicemail greeting can be read as a person, and a person who says "Hello," goes quiet for two seconds and says it again can be read as a machine (Answering Machine Detection, Twilio docs, read November 17, 2026).
Set those defaults next to the two-second rule. A person who says "Hello?" and stops has used 1.2 of their two seconds before the engine has decided they're a person, so the connection to an agent has to happen in what's left. Twilio can also run detection "in the background" and send your system the verdict once the engine reaches it. Whichever way your platform does it, time a sample of live answers from the end of the greeting to the moment an agent is on the line, before you tune detection for accuracy.
Is an auto dialer an "autodialer" under the TCPA?
"Auto dialer" is the industry's name for any system that places calls from a list. The Telephone Consumer Protection Act uses "automatic telephone dialing system" for something narrower. In Facebook v. Duguid the Supreme Court held that "a device must have the capacity either to store a telephone number using a random or sequential generator or to produce a telephone number using a random or sequential number generator" (Facebook, Inc. v. Duguid, No. 19-511, decided April 1, 2021; the statutory definition is 47 U.S.C. 227(a)(1)).
For a team running its own campaigns, that has a few practical consequences. Whether your system meets that definition is a question for your counsel, and a dialer vendor's page saying it doesn't is no substitute. The TCPA's separate rules for calls made with an "artificial or prerecorded voice" don't depend on the autodialer definition, and in February 2024 the FCC confirmed that those restrictions "encompass current AI technologies that generate human voices" (FCC 24-17, released February 8, 2024). And the telemarketing rules on abandonment, ring time and calling hours apply to telemarketing calls whatever system places them. This section summarizes federal sources and isn't legal advice, and states add rules of their own.
Where do AI agents fit in the call path?
A dialer's job ends when it hands a live person to an agent. When an AI agent places the call, the thing that dialed is also the thing that talks, so steps 4 and 7 stop being a constraint: there are no free people to pace against, and nobody who answers waits for one. Steps 1 to 3, 9 and 10 matter as much as before, which is why the list, the filters and the recycle rules are the first things to bring to an AI outbound program.
Which campaigns suit a dialer and which suit an agent comes down to how many whole agents each campaign has and how the conversation runs, and predictive dialer vs AI agents has the worksheet for that. What the agent does once someone picks up is in AI outbound calling.
How Telvana's outbound agents handle the call path
Telvana runs outbound two ways. You upload a contact list and the agent calls it within the days and hours you set for the campaign, or your own system places each call through the outbound calls API. The agent holds the conversation itself and transfers to a number you choose when your rules say a person is needed; that transfer is a cold one. Every call is recorded, transcribed, summarized and tagged by outcome, voicemail included, and calls placed through the API report a disposition such as completed, voicemail or failed.
Calling hours are set per campaign in that campaign's time zone, so a list that spans time zones goes in as one campaign per zone, and calls can pause on a schedule or on US public holidays. Each workspace keeps its own do-not-call list, and an agent can be set to add a number when the person asks not to be called again, but numbers aren't checked against the National Do Not Call Registry, so scrub the list before it goes in. There's no answering machine detection: voicemail is identified from the transcript after the call. A campaign calls each contact once unless you turn on "Call multiple times until answered," which calls back anyone who doesn't pick up, is busy or reaches voicemail, up to five calls in all, one to six hours apart. A call placed through the API gets one attempt, so your own recycle rules decide whether to call again. Pricing is from $399 a month based on call volume, plus a one-time setup fee. The outbound campaigns page shows the programs teams start with.
Frequently asked questions
What is the difference between an auto dialer and a predictive dialer?
A predictive dialer is one kind of auto dialer. "Auto dialer" covers any system that places calls from a list for your agents, and predictive is the pacing mode that dials before agents are free, based on a forecast of when they will be. Preview, progressive and power are the other common modes, and each dials fewer numbers per agent.
What is a power dialer?
A power dialer calls several numbers at once for each free agent and connects the first person who answers. Agents spend less time waiting than with one-at-a-time dialing, and in exchange some answered calls arrive when no agent is free. Good systems slow down on their own as abandonment nears the target you set.
Do auto dialers leave voicemails?
Many can. With answering machine detection on, a dialer can tell a recording from a person and then leave a recorded message, hang up, or mark the attempt for later, depending on your settings. Detection makes mistakes in both directions, and a prerecorded message brings in the TCPA's rules for prerecorded voice calls, so check consent before you turn it on.
How many times should an auto dialer call the same number?
Set it per campaign from your own dispositions. Look at which attempt number actually reaches people, stop at the attempt where reaches level off, and space retries across different days and times of day. Check the rules that apply to your calls as well, because some laws limit how often you can call the same person.
Can an AI agent replace an auto dialer?
For some campaigns. Where the conversation runs the same way nearly every time, such as confirmations, reminders and follow-ups, an AI agent can place the call and handle it, so there's no pacing to tune. Where a person has to negotiate or make a judgment call, keep the dialer and your people on it. Many centers run both side by side.
Fill in the call path before you pick a dialer
Before you compare dialers or AI agents, fill in the ten steps above for one real campaign: where the list comes from, how it's filtered, whose clock the calling window follows, and how many tries a record gets. If that campaign's calls run the same way nearly every time, book a demo and we'll show you an agent working a list like it.