Written for the operations lead deciding how each campaign should dial: the four dialing modes explained precisely, the abandon math shown properly, and straight answers on which mode fits which list.
What is the difference between predictive, progressive, power, and preview dialing?
Preview shows the agent the record and lets them launch the call. Progressive dials exactly one call per ready agent, so a live answer lands on a waiting agent and abandonment stays near zero. Power dials a fixed admin-set ratio per ready agent, which can exceed 1:1. Predictive paces statistically ahead of agent availability, trading a controlled abandon rate for the highest occupancy.
The operational definitions, anchored to the moment an agent goes ready.
Vendors recycle these four labels loosely, which is how a power dialer in a demo turns out to be predictive pacing with a friendlier name. The definitions that hold up are operational: what the system does the moment an agent becomes ready, and who decides when a call launches.
The agent sees the record before anything dials. They read the history, then launch the call themselves, or skip it where the campaign allows skipping. One record, one human decision, one call. Throughput is the lowest of the four modes and control is total.
The system dials exactly one call per ready agent, automatically, and only when that agent is actually free. A live answer therefore lands on an agent who is already waiting. Abandonment is near zero by design, not zero in practice: carrier answer signaling and the occasional timing collision at handoff leave a small residue that survives even a clean configuration.
The system dials a configured ratio of calls per ready agent, say two or three lines each. The ratio is set by an administrator and stays put; nothing statistical adjusts it in the moment. Anything above 1:1 means some live answers will find nobody free.
The system paces statistically, launching calls ahead of agent availability based on live answer rates, talk times and wrap-up, so the next connect arrives roughly as an agent frees. It is the highest-occupancy mode and the only one where abandonment is structural rather than incidental.
Five9's own product material draws the same lines: predictive adjusts dial rates from predicted agent availability, power dials a user-configured number of calls per available agent, progressive paces adaptively for smaller teams, and preview delivers the record with CRM history to the agent before any dialing happens. The vocabulary is standard across the industry. What varies between platforms is the quality of the pacing math and the abandon controls wrapped around it.
One curve, and every pacing decision is a position on it.
Dial conservatively and agents wait for connects. Dial aggressively and connects wait for agents, except a person who has answered will not wait, so the call gets abandoned. No setting escapes this trade; you are choosing where on the curve to sit.
Before any numbers, the definition that everything else depends on: abandon rate is abandoned calls divided by calls answered by a live person. Not divided by dials. The US federal rules and Ofcom's abandoned call rate calculation both use the live-answer denominator, and the difference is enormous in practice because most dials never reach a person.
A worked illustration, with invented round numbers rather than client data. A campaign makes 10,000 dials in a day and 2,000 are answered by a live person. The pacer overshoots through the morning and 60 of those live answers get the abandonment message instead of an agent. Measured against dials, that is 0.6 percent and looks harmless. Measured correctly, it is 60 divided by 2,000: a 3.0 percent abandon rate, sitting exactly on the US federal ceiling with no margin left for a bad afternoon. Dashboards that report abandonment against attempts flatter the operation by roughly the inverse of the answer rate, which is why we recheck the denominator on every floor we audit.
The occupancy side is what you buy with that risk. On progressive, the same floor triggers one dial per free agent, and with one live answer per five attempts an agent sits through four dead attempts between conversations. The US safe harbor conditions require an unanswered call to ring at least 15 seconds or four rings before disconnecting, so those dead attempts cannot be rushed; the idle time between connects is real and structural in that mode. Predictive overlaps the dead time across the whole pool, which is where its extra conversations per hour come from.
| Mode | What sets the pace | Occupancy | Abandon exposure |
|---|---|---|---|
| Progressive | Agent readiness, locked 1:1 | Lowest of the automated modes | Near zero by design, never quite zero |
| Power | A fixed admin-set ratio | Higher, but blind to conditions | Spikes when answer rates swing |
| Predictive | A statistical model, continuously updated | Highest | Structural, must be managed to a target daily |
Power deserves its own warning. A 2:1 ratio tuned at 10am against a cold list behaves very differently at lunchtime, or the moment a fresh list drops in and answer rates jump. Nothing in the mode notices. On occupancy gains generally: they depend on your answer rate and handle time, and anyone quoting a universal minutes-per-hour uplift without seeing your numbers is quoting marketing.
Our free Five9 Health Check runs 17 questions across pacing, abandon management and campaign structure, then returns a scored report with the fixes ranked by impact.
Minimum agents for a predictive dialer, and why the math destabilizes below it.
Five9's campaign administration guidance is unusually blunt here: predictive mode should never run with fewer than 15 consistently active agents in the campaign, and for stable pacing those agents should not be shared across multiple campaigns. The public product pages put the line lower, at around ten active agents. We plan floors around the stricter figure, because it comes from the documentation written for the people who configure the thing.
The reason is small-number statistics, not vendor caution. The pacer forecasts when the next agent will free up from averages: average talk time, average wrap, average answer delay. Averages need volume to mean anything. With 40 agents on a campaign, one conversation running twenty minutes long is noise the model absorbs. With eight agents, that same call silently removes 12.5 percent of the capacity the model thinks it has, and the batch of calls already in flight against that capacity has nowhere to land. Those become abandoned calls. The pacer then overcorrects and throttles, agents sit idle, the model relaxes, and it overshoots again. Below the threshold the algorithm does not fail gracefully; it oscillates.
Our operational experience matches the documentation. Somewhere in the mid-teens, predictive stops being a statistics engine and starts being an abandon generator with good marketing. Below that line, power at a modest ratio gives you most of the throughput with a lever you actually control, and progressive gives you the compliance posture. If you are running predictive with nine agents because that is how the campaign was built two years ago, that is not a tuning problem, it is a mode problem. The pacing parameters that sit underneath this decision, abandon targets, answer-machine handling and ring time among them, are a separate subject covered in our guide to Five9 predictive dialer settings.
The same two-second event, two very different legal frameworks.
Regulators on both sides of the Atlantic define abandonment around the same moment: a person answers, and no agent is connected within two seconds of their completed greeting. What happens after that moment differs sharply by country, and it changes which modes are sensible where.
The FTC's Telemarketing Sales Rule (16 CFR 310.4(b)(4)) gives sellers a safe harbor if they abandon no more than 3 percent of calls answered by a live person, measured over the duration of a campaign if it runs under 30 days, or over each successive 30-day period if longer. The same provision requires unanswered calls to ring at least 15 seconds or four rings, and any abandoned call must promptly play a recorded message naming the seller and giving a phone number. The FCC's parallel rules (47 CFR 64.1200(a)(7)) carry the matching 3 percent live-answer ceiling per 30-day campaign period, with a prerecorded identification and opt-out message requirement. Together these make predictive dialing lawful to engineer in the US: the rules set a measurable ceiling, and operators typically configure the platform's abandon target below it to leave margin for bad days inside the measurement window.
Ofcom's persistent misuse statement, published in December 2016 and applying from 1 March 2017, dismantled what much of the industry had treated as the UK equivalent. The 3 percent figure in the older 2010 policy was only ever a criterion for prioritizing enforcement, and respondents to Ofcom's consultation acknowledged it had created a belief that staying under 3 percent meant staying compliant. Ofcom declined to formalize that belief, on the reasoning that it could not permit a certain amount of unlawful conduct or fetter its own discretion to act. The abandoned call rate still exists in the policy, calculated per calendar day against calls connected to a live individual, on a per-campaign or per-call-center basis, but it now functions as a factor in choosing which cases to pursue, with penalties available up to 2 million pounds. The statement also sets out what Ofcom expects of dialer users: dialing rates appropriate to the number of agents employed, competent management of the technology, and records kept for at least six months.
The practical consequence is that mode selection is itself the compliance control. A US floor can run predictive against the safe harbor and evidence its 30-day numbers. A UK floor running predictive is making a documented risk decision with no numeric defense if Ofcom comes asking, which is why regulated UK brands so often default to progressive and reserve predictive for segments where the risk case has been made deliberately. We keep separate, current pieces on the US abandonment rules and the Ofcom dialler rules. Both regimes get amended and enforcement priorities move, so verify against the current texts and take advice from qualified counsel where the stakes justify it.
Preview dialing use cases, predictive territory, and the segments in between.
Defaults, not doctrine. Every row has exceptions, but if you are staring at a new campaign wondering where to start, start here.
| Use case | Start with | Why | The catch |
|---|---|---|---|
| Fresh inbound leads, minutes old | Progressive, or power held at 1:1 | Contact rates on new leads decay fast, and the person who just filled in a form deserves an agent on the line the instant they answer. Abandoning a hot lead is the most expensive abandon there is. | Keep fresh leads out of any shared campaign with cold list. Our speed-to-lead work is built around this segment. |
| Aged and cold B2C lists | Predictive, with 15 or more agents (US floors) | Answer rates are low, so agent-triggered dialing leaves the team idle for most of the hour. This is the segment predictive was invented for. | Abandon performance needs watching daily. In the UK, treat this pairing as a deliberate, documented risk decision rather than a default. |
| Collections segments | Preview or progressive | Contact rules in collections attach to the account, not the list. Regulation F presumes a violation past seven calls in seven days per debt, which makes volume dialing across a portfolio a liability engine. | Frequency caps must be enforced per account. See our piece on Regulation F collections dialling. |
| High-value B2B prospecting | Preview | When one conversation can be worth five figures, thirty seconds reading the account history beats two extra attempts per hour. | Set preview time limits, or preparation quietly becomes browsing. |
| Callbacks and booked appointments | Preview | The prospect agreed to a specific call at a specific time. An abandoned callback is a broken promise with your brand attached. | Use scheduled auto-delivery so commitments do not depend on agent memory. |
The pattern underneath the table is worth naming: the warmer and more valuable the contact, the more human judgment belongs in the loop; the colder and cheaper the record, the more the math should run things.
Dialing mode is a campaign property, and the platform will happily run several at once.
The most common architecture we find in audits is one campaign, one mode, everything in it: fresh leads, ancient records, callbacks, the lot, all pacing predictively because that is how go-live was configured and nobody has touched it since. It works in the sense that calls happen. It also fails every segment in the table above simultaneously.
The architecture you actually want is several campaigns, each carrying the mode its list deserves: a progressive campaign for fresh leads, predictive for the aged list with the full agent pool behind it, preview for callbacks and promises. Each campaign then gets its own pacing settings, its own caller ID strategy, its own abandon management, and its own place in the priority order, so the fresh-lead campaign always outranks the cold list for the next available agent.
This split is usually the first structural change we make in an optimization engagement, ahead of any parameter tuning, and it does more work than any parameter. Moving each segment onto the mode its economics and rules actually call for is a large part of how we have cut abandon rates by 35 percent across our client portfolio; that is a portfolio aggregate, not a promise about any single floor. The campaign-splitting exercise itself costs a planning afternoon and some list hygiene, which makes it about the cheapest performance improvement available in outbound.
Predictive assumes a stable agent pool. Inbound traffic breaks that assumption hourly.
Everything above assumes the agent pool holds still. Blended floors violate that constantly, and predictive is the mode that punishes it. The pacer launches calls against capacity it expects to exist a few seconds from now; when an inbound spike pulls three agents away mid-batch, the calls already ringing have nothing behind them, and the people who answer become your abandon rate.
Five9's guidance about keeping at least 15 consistently active agents on a predictive campaign, unshared across campaigns, is really a statement about stability rather than headcount. Fifteen agents who are sometimes twelve is not fifteen agents.
Three arrangements hold up on real blended floors:
The honest answer for small blended teams is less comfortable: predictive pacing, heavy blending and a clean abandon record do not coexist, so pick the two that matter most to your operation. We run blended floors daily, and the ones that perform have made that choice explicitly rather than letting the dialer make it for them at the worst possible moment.
No. Progressive launches one call per ready agent, so a live answer almost always finds an agent waiting, but carrier answer signaling delays and occasional timing collisions during the handoff mean a small residue of abandons survives even a clean configuration. Treat progressive as near-zero by design rather than zero, keep the abandonment message configured anyway, and monitor the numbers. Floors that assume zero tend to stop watching, which is how surprises happen.
Five9's campaign administration guidance says predictive mode should never run with fewer than 15 consistently active agents on the campaign, and that those agents should not be shared across campaigns. Its product pages put the line lower, around ten. The stricter figure is the safer planning number: below the mid-teens the pacing model works from too few data points, oscillates between overshoot and idle, and generates abandons. Smaller teams get better results from power at a modest ratio, or from progressive.
Abandoned calls divided by calls answered by a live person, never by total dials. The US federal rules measure it this way over 30-day periods per calling campaign, and Ofcom's abandoned call rate formula likewise uses calls connected to a live individual as the denominator, calculated per calendar day. Reporting abandons against dials understates the true rate by roughly the inverse of your answer rate, which is how a dashboard showing 0.6 percent can describe a floor actually running at 3 percent.
Progressive is locked to 1:1, one call per ready agent, launched only when that agent is free. Power dials a configured ratio per ready agent, say two or three lines each, set by an administrator and unchanged until someone edits it. Above 1:1 some live answers will find no agent available, so power carries real abandonment risk, and because nothing statistical adjusts the ratio in the moment, that risk spikes whenever answer rates swing, such as when a fresh list arrives mid-shift.
Preview or progressive for most segments. Collections contact rules attach to the individual account: in the US, Regulation F presumes a violation when a collector calls a person about a debt more than seven times within seven days, or within seven days of a conversation about it. Ratio-based and predictive dialing across a portfolio make per-account discipline hard to evidence. Agent-controlled or 1:1 dialing keeps frequency caps, right-party contact and documentation manageable, and any more automated segment should be a deliberate exception.
There is no UK numeric shield. Ofcom's persistent misuse statement, in force since March 2017, declined to create a safe harbour: the older 3 percent figure was only a criterion for prioritizing enforcement, and Ofcom said it cannot permit a quota of unlawful activity. Any abandoned call can in principle attract action, with penalties up to 2 million pounds, and the abandoned call rate now serves to prioritize cases rather than to define compliance. UK floors treat abandonment as something to minimize, not a budget to spend.
Yes. Dialing mode is set per campaign, and running several campaigns with different modes at the same time is the intended architecture. A typical structure runs fresh leads progressive, aged lists predictive with the main agent pool, and callbacks preview, each with its own pacing settings, caller ID plan and abandon management, ordered by campaign priority. The common mistake is the opposite: one campaign, one mode, every record type mixed together, which serves every segment badly at once.
It is the lowest-throughput mode, and that is fine where conversation value justifies it. High-value B2B, collections accounts with history, and booked callbacks all repay thirty seconds of preparation more than they repay two extra dials an hour. The failure mode is unbounded preview time drifting into browsing, so set a preview time limit with auto-dial at expiry. For cheap, cold records preview genuinely is the wrong tool; that is what pacing automation is for.
AL Performance is a Five9 Certified Implementation Partner, and we run outbound floors of our own every day. Bring us your abandon numbers, your team size and your list mix, and we will tell you which modes belong where and what we would change first.
Talk to us