Auto dialer vs predictive vs progressive vs power vs parallel dialer: what is the difference?

An auto dialer is the umbrella term for any software that places outbound calls automatically from a list; predictive, progressive, power, preview and parallel describe specific ways that automation can pace calls and connect them to a person. The difference that matters when you are buying is who is reserved to handle each answered call, and when: a predictive dialer gambles on a forecast, a progressive dialer waits for an available agent, a power dialer usually works a queue one line at a time, and a parallel dialer rings several numbers at once for a single rep and connects the first answer.

The dialer types at a glance

Type	How it dials	Lines per rep	What the person who answers hears	Best for	Main risk
Auto dialer (umbrella)	Whatever mode it is set to: preview, progressive, power or predictive	No defining number	Depends entirely on the mode and implementation	Any automated outbound calling, sales or service	The label alone tells you almost nothing about pacing or the recipient's experience
Predictive	Starts calls based on a forecast of when agents will free up	Variable; can exceed the number of currently available agents	Silence or a recording while the system classifies the answer and finds an agent	Large contact-center campaigns with pooled agents	More human answers than agents, creating abandoned calls and dead air
Progressive	One call per agent who is actually available, right now	One per available agent	Call analysis runs first, then an agent is bridged; a short pause is possible	Controlled-volume, higher-value outreach	A blended agent can take an inbound call before the outbound prospect answers
Preview	The rep reviews the record first, then chooses to call	One	The rep, already prepared, from the first word	Personalized, complex or sensitive accounts	Idle time while reps review records
Power	Works a queue, usually one line at a time, next number dials itself	One in most sales products; some contact-center versions dial more	The rep, often already on the line at "hello"	Warm B2B lists, follow-ups, insurance outreach	"Power" is not standardized: Kixie's Multi-Line PowerDialer runs up to four lines, so verify the algorithm, not the name
Parallel	Rings multiple numbers at once for one rep	Configured; Apollo recommends three and permits five	The first detected answer is connected; other lines are canceled	B2B SDR prospecting where most attempts hit voicemail or no answer	If a second person answers before the first is connected, there is no rep for them

Auto dialer: the umbrella, not a pacing mode

An auto dialer automates outbound calling, usually from a contact list. Cisco Webex's glossary groups preview, power and predictive dialing underneath it, so the label says nothing about pacing or lines. It should not be confused with automatic call distribution, which routes inbound calls.

Because the label is that broad, comparing two "auto dialers" is meaningless until both suppliers tell you four things: what triggers a dial, how many simultaneous attempts it makes, what happens when a human answers, and whether a person, a recording or an AI voice handles the conversation. Genesys documentation last edited October 2, 2020, for example, describes an agentless mode that plays interactive messages at a configured rate with no human connection at all, which is very different from the human-connected sequential products most sales teams picture.

Predictive dialer: dialing ahead of agent availability

A predictive dialer uses an algorithm to estimate when agents will become free and when calls will connect, then launches attempts so that human answers land just as agents finish their previous calls. Genesys Cloud lists average handle time, after-call time, idle time, script stage and historical contact rates among its pacing inputs, and the dial rate varies with campaign conditions rather than a fixed lines-per-agent number.

The trade: predictive gives up the certainty of waiting for a specific agent in exchange for a forecast about the pool. When the forecast holds, utilization goes up. When answer rates, handle times or staffing shift suddenly, human answers arrive with no agent ready, and the prospect hears silence, a message, or nothing before a disconnect. Genesys recommends at least 15 agents for its Predictive mode, which is product guidance, not a legal or industry-wide minimum.

On figures, a PinnacleVoice page labeled January 2025 lists 50-80 calls per agent per hour without distinguishing attempts from connected conversations, so treat it as a vendor estimate. A Nextiva article published January 19, 2026 and updated February 17, 2026 repeats a claim of 40 to 57 minutes of talk time per hour, citing Wikipedia rather than any original study. Predictive pacing is a fit for large, stable agent pools with solid abandonment controls, not something to choose off a headline calls-per-hour number.

Progressive dialer: one attempt per available agent

A progressive dialer starts exactly one call for each agent who is genuinely available, counting after-call work as part of availability. Six ready agents produce six attempts, no speculative surplus.

The 1:1 ratio reduces the risk of more human answers than you can staff, but it does not eliminate it, and it does not guarantee the rep hears the prospect's first hello. Genesys Cloud runs call analysis before matching the call to an agent, so answering-machine detection and routing can still add a pause. Genesys also warns that a blended agent marked available for an outbound attempt can take an inbound call first, leaving the outbound prospect unanswered; its "Precise" option addresses this by reserving agents before dialing so they cannot take inbound work until released. When a vendor promises "one call per agent," ask whether that is an approximate capacity count or an actual reservation.

Progressive fits controlled-volume, higher-value work. Vonage's March 13, 2026 article points to renewals, follow-ups and mid-volume campaigns; Genesys positions it for smaller, higher-value campaigns. No credible mode-specific dials-per-hour benchmark exists in the reviewed sources.

Preview dialer: review before you call

Preview dialing shows the rep the contact record first, and the rep decides to dial or skip. It appears in auto-dialer product families even though the call itself may need a manual click, and some implementations use a configured preview timer, though there is no standard number of seconds. Genesys flags excessive idle time as the drawback when the timer is disabled.

The point of preview is preparation and account control, and it pairs well with automation after the dial: a Genesys case study of US wireless retailer Round Room describes advisors no longer waiting to leave voicemail because the system detects the approaching voicemail and releases them for the next call. The undated study reports a 20% performance improvement with 10% fewer agents, without defining the performance metric, so read it as evidence of a workflow gain, not a connect rate.

PinnacleVoice's January 2025 page lists 15-25 calls per agent per hour for preview, with the same caveat about missing methodology.

Power dialer: usually sequential, but verify the meaning

In most sales products, a power dialer works through a queue one contact at a time and dials the next number automatically when the previous call or wrap-up finishes. PhoneBurner says its rep is already connected when the prospect says hello, so there is no post-answer hunt for a free agent. Aircall describes collecting numbers from a web page or CRM view into a queue, dialing sequentially and writing outcomes back to integrated systems.

But "power" is not a standardized category. Genesys Cloud's Power mode can dial multiple numbers per idle agent and adjust speed against abandonment risk, and Kixie sells a product called Multi-Line PowerDialer with up to four simultaneous lines. Webex treats power and progressive as equivalent terms. Ask for the actual dialing algorithm, not the name on the tin.

On figures, PhoneBurner advertises up to 80 contacts per hour (undated page, no test protocol), and Aircall's September 24, 2026 guide claims more than two minutes saved per call and three times more leads reached per hour. Those are vendor claims about removing number entry, logging and navigation, not independently established benchmarks for every team.

Parallel dialer: several attempts for one rep

A parallel (or multi-line) dialer rings multiple numbers at once for a single rep, then connects the rep to the first detected human answer and cancels the remaining attempts. The concurrency is configured per rep, not per pool. Apollo recommends three simultaneous attempts and permits five; Orum lets organizations and reps set their own maximum, and setting it to one turns parallel dialing off. These are product settings, not industry limits.

Parallel dialing does not remove the concurrency problem, it just moves it. Orum documents that the rep never hears the ringing or the prospect's initial greeting: AI classifies the answer first, then transfers the rep in with an alert. If the prospect has already said hello during that classification, the prospect experiences a pause even though the rep perceives a clean connection. And when two people answer at nearly the same moment, one of them gets canceled. A canceled ringing attempt is not the same event as a dropped human-answered call, and compliance measurement has to keep them separate.

Undated Orum documentation reports its own team at roughly a 5% connect rate using four to five simultaneous calls, and recommends a connection every 90-120 seconds. Those are operating examples from one vendor, not expected results for everyone.

Parallel fits B2B prospecting where most attempts reach voicemail or nothing, and the rep's scarce resource is live conversations. Apollo recommends power dialing instead when the rep wants context review, manual voicemail drops or careful outcome logging.

CallCloud's own Parallel Dialer is an example of a different parallel design: the rep sits on a live audio bridge for the whole session, so an answered line joins the rep with a measured median of 22 ms (99% within 50 ms, measured on live calls October 8, 2026), with no answering-machine detection window before the rep is let in, and the other lines keep ringing until the rep presses Connect, then return to the queue rather than being cut off mid-ring. See the Parallel Dialer and the zero-lag measurement.

AI dialers: who actually speaks?

Two very different products get sold as "AI dialers." The first is AI-assisted human dialing: the AI classifies answers, prioritizes contacts or drafts notes, and a human conducts the conversation, as in Orum's human-detection workflow. The second is a conversational AI voice agent: software that recognizes speech, generates and speaks responses and executes tasks, with a human joining only if the workflow transfers the call. Retell documents this second model, including scheduled campaigns, pacing, retries and live-agent transfer.

A Retell case study with Matic insurance reports 8,000-plus AI-operator calls in Q1 2025, 3,000-4,500 scheduled calls monthly, and an 85-90% transfer success rate to licensed human agents. It also attributes better answer rates in A/B testing largely to calling at the exact scheduled minute, without publishing a pickup-rate comparison, so the evidence does not isolate a benefit caused by synthetic speech. Transfer success is a handoff metric, not a dial-to-answer rate, and "calls," "leads reached," "human answers" and "successful transfers" are not interchangeable units when you compare vendors.

The pickup delay problem: why the prospect says hello twice

A detected human answer and an audible rep are two separate events, and the gap between them is where dialers earn or lose their reputation. Delay can come from several places:

Answer classification. Answering-machine detection (AMD) listens for speech and silence patterns to tell a human from voicemail. Twilio's documentation (modified August 19, 2026) describes configurable speech and silence thresholds and acknowledges misclassification.
Blocking versus background analysis. Twilio's default synchronous AMD blocks the call until classification finishes; asynchronous AMD runs detection in the background. Either way, the application decides what the prospect hears during detection.
Agent selection and bridging. Reserving an agent before dialing costs less recipient-side time than finding one only after the contact answers, a distinction Genesys has documented across its older and newer dialing material.
The rep's entry point. In Orum-style parallel dialing the rep joins after classification, so the prospect's greeting happens before the rep is on the line.

Historical timing evidence is sparse. A June 27, 2002 California Public Utilities Commission decision records industry submissions of 150-400 ms to identify a human answer and 600 ms to 1.4 seconds to transfer to an agent, which are equipment figures from that era, not current product benchmarks. Apollo's undated documentation states roughly three seconds of latency when voicemail drops are enabled and substantially less without them. No reliable percentage of prospects hanging up specifically because of answer-to-agent delay was verified in the research; PhoneBurner's 2022 report on more than 100 million calls made in 2021 warns about delay-related hangups qualitatively but publishes no percentage isolating that cause.

The practical decision rule: treat dead air as an end-to-end routing problem. Faster AMD helps, but it cannot compensate for a rep who was never available. When you test a dialer, measure answer-to-first-rep-audio from call recordings, not from the rep's screen, and count simultaneous answers separately.

What the rules say about abandoned calls

The legal measure of an abandoned call is narrower than most teams assume. Under the FCC's rule (47 CFR 64.1200(a)(7), covered telemarketing calls), a call is abandoned when a live sales representative is not connected within two seconds after the recipient's completed greeting, and no more than 3% of calls answered live by a person may be abandoned, measured per campaign over the applicable 30-day period. If no representative is available in time, a limited identification and opt-out message must play, and playing it does not turn the call into a connected one. A separate rule, 64.1200(a)(6), prohibits disconnecting an unanswered covered telemarketing call before 15 seconds or four rings.

The FTC's Telemarketing Sales Rule has its own abandonment prohibition (16 CFR 310.4(b)(1)(iv)) using the same two-second completed-greeting test, plus a safe harbor (310.4(b)(4)) requiring all of: 3% or less abandonment among live-answered calls, the 15-second or four-ring minimum, the required recorded identification message, and compliance records. The safe harbor is conditional relief from that prohibition, not permission to ignore the rest of the TSR or the FCC rules.

In the UK, do not import the US 3% figure. Ofcom's persistent-misuse statement (published December 20, 2016, applying from March 1, 2017) deliberately does not make 3% abandonment a lawful threshold or safe harbor, and its January 22, 2026 refresher reiterates that organizations must prevent silent and abandoned calls, with penalties up to GBP 2 million for persistent misuse. Ofcom specifically identifies AMD false positives, mistaking a human for voicemail and disconnecting, as a source of silent calls.

Two legal summaries in circulation are wrong. The FCC's one-to-one consent restriction was vacated by the Eleventh Circuit on January 24, 2025 (Insurance Marketing Coalition v. FCC), and the FCC removed the invalidated amendment on July 14, 2025, restoring the prior definition of prior express written consent, which still requires a signed written agreement with the required disclosures. And Ofcom offers no 3% safe harbor at all. Keep US and UK abandonment policy separate.

Legal basics for dialer buyers

This section is general information, not legal advice.

What counts as an autodialer. In Facebook, Inc. v. Duguid (decided April 1, 2021), the Supreme Court held that an ATDS under the TCPA must have the capacity to use a random or sequential number generator to store or produce numbers to be called. A product is not an ATDS just because its marketing calls it an auto dialer, and the ruling is not a general exemption for list dialing: the equipment's actual capabilities and the applicable interpretation matter.
Prerecorded and AI voice. Artificial or prerecorded voice is a separate TCPA trigger from the ATDS question. The FCC's declaratory ruling released February 8, 2024 treats AI-generated voices as artificial voices under the TCPA, including technology that purports to act as a live agent. Prerecorded or artificial-voice telemarketing to protected numbers requires prior express written consent, plus caller identification and opt-out requirements.
DNC and calling hours. Covered residential solicitations are prohibited before 8 a.m. or after 9 p.m. local time, and callers must honor national and seller-specific do-not-call requirements, screening registry data no more than 31 days old.
B2B versus B2C. Most genuine B2B calls are exempt from the general TSR provisions, but the exemption preserves prohibitions on misrepresentation (the FTC's 2024 amendments narrowed it for deceptive conduct), and it is not blanket immunity. A business purpose does not erase TCPA protection for cellular numbers against covered ATDS or artificial-voice calls, so "B2B," "business landline" and "employee mobile" are three different classifications.

How to choose by team and use case

The recurring axis across all five modes is what each one treats as the scarce resource. Preview protects preparation time. Progressive protects available-agent capacity. Power protects the rep-centered flow while removing manual work. Parallel protects live conversations by overlapping the dead time between attempts. Predictive protects pooled utilization. Choose for your actual constraint:

Small B2B SDR team on a long list: parallel or power dialing. Parallel when most attempts go unanswered, power when reps want context and control per call. Apollo's own guidance makes this split.
Follow-ups and warm lists: power or preview, where the first hello and the quality of the conversation matter more than attempts per hour.
Large pooled contact center with 15+ agents and stable staffing: predictive, with reservation-based progressive as the lower-risk alternative.
Sensitive or high-value accounts: preview.
Scheduled confirmations and qualification at scale: conversational AI voice agents, which change who speaks, not just how fast you dial, and carry their own consent obligations.

CallCloud sits deliberately in the B2B part of this map. Our Parallel Dialer runs up to five lines in the browser with the rep on a live bridge, and our Power Dialer works a queue one line at a time ($60 per seat per month; the Parallel Dialer is $125 per seat per month, both unlimited US and Canada calling, month to month, no seat minimum; you can start a free trial with no card). See the Power Dialer and pricing. We do not do predictive dialing for large contact centers, IVR or call-center routing, and we say so rather than stretching the product. If your team is the pooled-contact-center case, a contact-center platform like Genesys Cloud or Five9 is the better fit.

FAQ

Is a parallel dialer a type of auto dialer?

Yes. Auto dialer is the umbrella for automated outbound calling, and parallel describes one pacing behavior underneath it: multiple simultaneous attempts for a single rep, with the first detected answer connected. Cisco Webex's glossary groups several such modes under the auto-dialer label.

Does power dialing mean one line?

Usually, but not always. Most sales power dialers work a queue one number at a time, but Genesys Cloud's Power mode can dial multiple numbers per idle agent and Kixie's Multi-Line PowerDialer runs up to four lines, so ask for the dialing algorithm rather than trusting the name.

Why do predictive and parallel dialers sometimes leave dead air?

Because detecting a human answer and putting a rep on the line are separate steps. Answering-machine detection has to classify the answer, and in designs where the rep joins only after classification, the prospect's greeting can happen before any rep is present. If answers outpace available reps, the gap becomes silence or a disconnect.

What is the legal limit on abandoned calls in the US?

For covered telemarketing, the FCC caps abandonment at 3% of calls answered live by a person, per campaign over the applicable 30-day period, and a call counts as abandoned if a live sales representative is not connected within two seconds after the completed greeting. The FTC's Telemarketing Sales Rule has a matching prohibition with a conditional safe harbor. The UK's Ofcom sets no 3% threshold at all.

Which dialer gives the rep the fastest connection to the prospect?

Sequential power dialing, where the rep is already on the line when the prospect answers, and parallel designs that keep the rep on a live bridge rather than transferring them in after answer classification. CallCloud measures a median 22 ms answer-to-rep bridge because the rep is on the bridge for the whole session, before any answer.

Are AI voice agents legal for outbound sales calls?

It depends on consent and purpose, not on the technology alone. The FCC's February 2024 ruling treats AI-generated voices as artificial voices under the TCPA, which requires prior express written consent for advertising or telemarketing to protected numbers, plus identification and opt-out rules. An AI agent also does not automatically satisfy rules written around a live sales representative, so get advice for your specific implementation.