Auto dialer vs power dialer vs predictive dialer

Every dial mode is the same machine with one number changed: how many lines it opens for each agent who is free. Preview opens none until a human says go. Power opens one. Parallel opens a fixed several. Predictive opens a forecast. Connect rate, the pause the called party hears, the abandonment you have to account for and the tuning burden you inherit are all downstream of that single ratio, which is why arguing about features misses the decision entirely.

The four modes differ in one number, and everything else follows

Write the ratio down for each mode and the rest of the comparison stops needing a table.

• Preview opens zero lines automatically. The agent sees the record, reads it, and triggers the call. Lines per free agent: nothing until a human acts.
• Power opens exactly one line per free agent, immediately, as soon as the previous call ends. Lines per free agent: one.
• Parallel opens a fixed number greater than one — three is the common default — and bridges whoever answers first. Lines per free agent: N.
• Predictive opens a number derived from historical answer rate, average handle time and how many agents are about to become free. Lines per free agent: a forecast, recomputed continuously.

Two consequences fall straight out of that list and they are the whole argument. Any mode with a ratio of one or less can never abandon a call, because there is always an agent attached before the phone rings. Any mode with a ratio above one abandons calls by construction, and the only open question is how many.

Preview dialing: the agent decides, and the system waits

Preview is the slowest mode and the only one where the agent knows who they are calling before it rings. The system presents the record — history, last disposition, the note from six weeks ago — and does nothing until the agent chooses.

Use it when preparation changes the outcome and the conversation is worth the preparation time. Named accounts, renewals, anything where opening with the wrong context costs you the meeting. In Autocloz, preview is the default mode on a new dialer campaign, and its dial strategy is deliberately a no-op: the engine never auto-dials in preview, so the mode exists as a uniform shape for the engine rather than as an automation.

The trade is arithmetic. If an agent spends ninety seconds reading and four minutes talking, research is 27% of the working hour. At a deal value where 27% of an agent's time is worth less than the lift in conversion, preview wins. Below that, it does not, and no amount of enthusiasm for research changes the sum.

Power dialing: one line per free agent, and no dead air

Power removes the seconds between calls and adds no risk, because the ratio never exceeds one. There is no bridge pause at pickup — the agent is already on the line — and there is no abandoned call to account for, because no call was ever placed without an agent behind it.

The shipped strategy is as plain as that description: it pairs ready agents with dialable queue rows in order, one each, capped by whatever pacing budget remains for the tick. Deterministic, and it cannot over-dial by design.

For most business-to-business outbound this is the correct default, and the reason is that the thing power dialing removes — inter-call dead time — is genuinely a large share of an hour, while the thing it does not remove is not the mode's fault. Power dialing eliminates gaps. It does not manufacture pickups. If eight in ten of your numbers ring out, power dialing gets you to the same two conversations faster, and that is the whole benefit.

Parallel dialing: more lines than agents, and the abandonment that buys

Parallel opens N lines for one agent and bridges the first person to answer. The others are dropped. That is not a side effect to be tuned away; it is the mechanism.

Work the arithmetic, because the numbers are worse than most teams assume. Take three lines per agent and a live-answer rate of 25%, which is generous for cold outbound.

Each burst of three dials produces on average 0.75 live answers. The probability that at least one person answers is one minus 0.75 cubed, or 0.578 — so 57.8% of bursts connect somebody. The agent can take exactly one of them. Expected answers per burst is 0.75; expected connections is 0.578; the difference, 0.172 per burst, is people who picked up the phone and got nobody. Divide that by the 0.75 answers and the abandonment rate is 22.9%.

Drop to two lines per agent at the same answer rate and it is still 12.5%. At one line it is zero, which is power dialing. There is no setting between "abandons a fifth of your answers" and "is a power dialer" that a single agent can occupy, and that is the real finding: parallel dialing at small scale cannot be tuned into compliance, because the collisions are a property of the ratio and not of the tuning.

Predictive dialing: a forecast that needs a pool to forecast from

Predictive is the mode that is supposed to solve exactly the problem the previous section describes, and at genuine scale it does. The reason it fails below that scale is statistical rather than architectural.

The number of people who answer at once out of N open lines is a binomial variable. Its mean grows in proportion to N; its standard deviation grows with the square root of N. So the noise, expressed as a fraction of the signal, shrinks as the pool grows — and a pacing algorithm can only leave a safe abandonment margin when the noise is small relative to the margin.

Put numbers on it at a 25% answer rate. Four open lines: mean 1.0 answers, standard deviation 0.87 — the variation is 87% of the average, so any forecast is guessing. Eighty open lines, which is twenty agents at four lines each: mean 20, standard deviation 3.87, or 19% of the average. The same algorithm that is useless in the first case is workable in the second, without a single line of it changing.

This is why "we have a predictive dialer" is close to meaningless as a purchase criterion for a team of three. With three agents the model has nothing to predict from and degrades into an aggressive parallel dialer with a marketing name, carrying all of parallel's abandonment and adding a tuning surface you now have to own.

The regulator has already set the number, and it is three percent

The United States rules are specific enough to design against, and they are the reason the mode question is a compliance question rather than a preference.

Under 47 CFR 64.1200(a)(7), a telemarketer may not abandon more than three percent of all telemarketing calls that are answered live by a person, as measured over a 30-day period for a single calling campaign. The same paragraph defines the term: a call is abandoned if it is not connected to a live sales representative within two seconds of the called person's completed greeting. Paragraph (a)(6) adds that an unanswered call must not be disconnected prior to at least 15 seconds or four rings, and paragraph (c)(1) restricts telephone solicitation to residential subscribers to the hours between 8 a.m. and 9 p.m., local time at the called party's location.

Hold the 22.9% from the parallel section against the three percent ceiling. That is more than seven times the limit, produced by a default configuration and a plausible answer rate, in a mode many vendors ship switched on. The number is not marginal and it is not a tuning problem.

Two further points that get lost. The ceiling is measured per campaign over 30 days, so a rolling five-minute safety valve is a useful circuit breaker but is not the regulatory metric and cannot substitute for the record-keeping the rule requires. And the calling-hours rule is stated in the called party's local time, which means a compliant system needs the recipient's timezone, not the caller's. Rules differ by country and, inside the United States, by state — verify against your own jurisdiction rather than a summary like this one, and treat state-level rules as additive.

Which mode, decided by a rule rather than a preference

• Fewer than five agents: power. Predictive has nothing to predict from at that size, and parallel puts an abandonment obligation on a team too small to monitor it hour by hour.
• High deal value, long cycle, named accounts: preview, or power with a preview step reserved for the named list.
• Large agent pool, low-value high-volume list, and somebody whose job is watching the abandonment rate daily: predictive, with the 30-day per-campaign record as the metric you actually report on.
• Anything else, including most mixed motions: power, plus list hygiene. Verifying numbers and fixing which hours you dial in is nearly always the cheaper improvement, and it is the one that survives a regulator's question.

Autocloz's free plan covers 5 users and 10 mailboxes with the calling, LinkedIn, SMS and WhatsApp channels included alongside email — start free if you want to test a power-dial motion against your own list before committing to a dialer contract.

What actually caps your connect rate, and it is rarely the mode

Teams reach for a more aggressive ratio when the binding constraint is somewhere else entirely.

Number reputation. Once a carrier's analytics engine labels an originating number as likely spam, calls from it stop connecting, and the label is applied to the number rather than to you. The diagnostic is to plot connect rate per originating number rather than per campaign; a campaign average of 14% can be two good numbers at 22% and one dead one at 2%. A parallel dialer makes this worse faster, because it puts three times the unanswered attempts against the same number in the same hour.

List quality. Disconnected lines, switchboards and wrong numbers consume dial capacity without ever producing a conversation. A smaller verified list routinely beats a larger unverified one on conversations per hour, which is the metric that pays.

Local hour. Dialing at the wrong local time is both the cheapest connect-rate loss to recover and, per the rule above, a compliance exposure rather than merely an inefficiency.

If the phone is one channel in a wider cadence rather than the whole motion, multichannel sequence templates covers how a call step sits between an email and a LinkedIn touch, and cold calling scripts that book meetings covers what the agent says once the line connects. The dial mode decides how often that moment arrives; it decides nothing about what happens in it.

Why Autocloz stopped selling parallel dialing, and never shipped predictive

We have an opinion here and it cost us a feature, so it is worth stating plainly rather than as a positioning line.

Autocloz's dialer defines exactly three modes — preview, power and parallel — with preview as the default on a new campaign. There is no predictive mode anywhere in the product. On 30 May 2026 the create and update endpoints for dialer campaigns began refusing parallel for new campaigns with a 422 and the message *"parallel dialing is no longer offered — use 'power' or 'preview'"*. Campaigns that were already running in parallel keep running and remain editable; the engine and the bridging orchestrator were left untouched deliberately, because withdrawing a mode from sale is not the same as breaking it under customers who were relying on it.

The parallel strategy that remains carries an abandon-rate cap of 0.03 — the FCC's three percent — computed over a rolling five-minute window, and when the rate is above it the strategy places no new lines while already-ringing lines complete normally. That is a genuine safety valve. It is not a compliance proof, and calling it one would be the sort of claim this page exists to avoid: the regulation measures answered-live calls per campaign over 30 days, and a five-minute rolling ratio over queue dispositions is a different quantity with the same number attached to it.

The reasoning for withdrawal is the arithmetic in the parallel section. A mode whose default configuration produces roughly a fifth of answered calls abandoned, at a scale where the statistics cannot rescue it, is a liability we would rather not sell to a five-person team. If your volume genuinely justifies predictive dialing, buy a contact-centre platform built around it and staffed to monitor it; the Autocloz and Aircall comparison sets out where a CRM-attached dialer stops being the right shape.

What the dialer does not check before it places a call

The limits below are specific, and knowing them is the difference between a control and an assumption.

The calling-window gate is off unless you turn it on. The org setting calling_window_enforcement defaults to off, and a warn setting is explicitly non-blocking. Only strict prevents a call. A workspace that never opened the dialer settings page has a campaign engine that will dial at any hour.

When strict, the window is compared against UTC. The evaluator reads the configured start and end as HH:MM and compares them against the current UTC wall clock — not the recipient's local time, and not the workspace's timezone. An 08:00 to 21:00 window is therefore 08:00 to 21:00 UTC, which is 13:30 to 02:30 in India. Set the window in the timezone the comparison actually uses, and do not read it as an implementation of the 8 a.m. to 9 p.m. local-time rule, because it is not one.

A malformed window fails open. If either bound does not parse, the gate logs a warning and allows the call rather than jamming the campaign on a typo. That is a deliberate trade — a strict-mode misconfiguration stopping every call is its own outage — but it means "we set a window" and "a window is in force" are two different claims.

Quiet hours on the messaging channels use the sending account's timezone. For SMS, WhatsApp and call touches the quiet-hours evaluator resolves a campaign override, then the workspace default, then the sending account's own timezone, falling back to UTC. It is the sender's clock, by design, so a US dialer and an Indian dialer on one campaign do not bleed into each other's night — but it is not the recipient's clock, and any claim that it is would be wrong.

Do-not-call is checked, and it is checked twice. Queued rows matching the legacy voice-only list are excluded in SQL at claim time, and every claimed row is then re-checked against the cross-channel list — which is where a STOP keyword from an SMS reply or a manual entry lands — with a match terminally skipped and dispositioned dnc. That one is a real control rather than a caveat, and it is the reason the two stores cannot silently disagree.

What none of that does is decide whether the person on the other end wanted the call. The compliance gates in the dialer and the wider consent and suppression controls can stop a call that should not be placed. They cannot make a call worth answering, and no dial mode ever has.