How Five9 predictive pacing actually works, and the settings, timers and hourly checks that keep it fast without tripping the abandonment cap. Written by operators who run these campaigns daily.
What are the best practices for Five9 predictive dialer settings?
Run predictive only with roughly 15 or more consistently active agents, let the pacing algorithm work against a dropped-call target set with headroom below the regulatory 3 percent cap, and measure abandon rate as abandoned calls divided by calls answered by a live person. Then manage the inputs the algorithm depends on: campaign profile filters, disposition redial timers and list freshness, checked hourly.
The algorithm is a statistical bet, and it is only as good as its inputs.
A predictive campaign dials more numbers than there are agents free to take the answers. That is the entire trick. Most outbound dials go unanswered, so the dialer places calls ahead of agent availability, betting that by the time a person picks up, an agent will have just finished wrap and be sitting ready.
To size that bet, Five9 runs on live and recent statistics. Per Five9's campaign documentation, pace is driven by predicted agent availability built from call handle times, ring times, the percentage of calls answered and the dropped-call percentage, and the calls-to-agent ratio adjusts automatically to maximize agent usage while attempting to stay below the configured drop limit. Four inputs do most of the work: the live answer rate on the list right now, average handle time, ring time, and the number of agents ready or about to come out of wrap.
The ratio is the visible output. If one dial in five gets a live answer, the pacer can place several calls per expected free agent; if the answer rate climbs, it backs the ratio down, and as drops approach the configured cap it retreats further. None of this is magic. It is a rolling statistical estimate built from the recent past, which is exactly why it can be excellent and fragile at the same time.
Fragile matters. Feed the pacer a volatile list, a tiny agent pool or a filter that chokes the record supply and it does not fail loudly. It just paces badly, and the first symptom you see is either idle agents or an abandon spike, usually mid-morning.
Predictive pacing is an averages game, and averages need volume. With 40 agents on a campaign, one unexpectedly long call barely moves predicted availability. With eight agents that same call is 12.5 percent of capacity gone in a single event, the prediction swings, and the pacer lurches between over-dialing and idling.
Five9's own product page puts the threshold at more than 10 active agents for predictive mode, and its campaign administrator documentation goes further: predictive should never run with fewer than 15 agents consistently active in the campaign. Fifteen is the floor we hold on campaigns we run ourselves, and our Five9 implementation guide works to the same number. The gap between 10 and 15 lives in the word consistently. Twenty agents logged in is not twenty active once you subtract the coaching session, the long wrap, the two people pulled onto inbound and the one at lunch. The pacer counts who is genuinely in the rotation, and so should you.
Below the floor, the honest answer is that predictive is the wrong mode, not a mode that needs cleverer settings. Power or progressive dialing gives up some throughput in exchange for an abandon rate that stays put, and for a team of eight that trade is nearly always correct. We set out the comparison properly in predictive vs progressive, power and preview; the short version is that a small team on predictive collects the worst of both worlds, abandons when the pacer overshoots and idle time when it overcorrects.
The free Five9 Health Check is 17 questions covering pacing, abandonment measurement, dispositions and list hygiene. It takes a few minutes and shows where your configuration diverges from what we would run.
Regulators picked the denominator for you. Abandon rate is abandoned calls divided by calls answered by a live person. Not divided by dials, not divided by connects of every kind, and the difference is not cosmetic.
The FTC's Telemarketing Sales Rule safe harbor requires technology that keeps abandonment to no more than 3 percent of calls answered by a person, measured per calling campaign over each successive 30-day period, with a call counted as abandoned when no sales representative is connected within two seconds of the person's completed greeting (16 CFR 310.4). The FCC's parallel rule works to the same 3 percent of calls answered live by a person over a 30-day period per campaign (47 CFR 64.1200). Both texts also expect calls to ring at least 15 seconds or four rings, and the FTC safe harbor requires a recorded message naming the seller with a telephone number when no agent is available in time. Rules get amended and enforcement practice moves, so verify against the current text and take advice from qualified counsel where the exposure warrants it.
Now the worked example. A campaign places 5,000 dials in a day. 1,000 are answered by a live person, a 20 percent live answer rate. Of those 1,000, 35 hit silence or the recorded message because no agent was free within two seconds. Abandon rate: 35 divided by 1,000 is 3.5 percent, over the cap. Measured against dials it would be 0.7 percent, which looks comfortable and means nothing. We still find floors reporting the second number. Machine answers, unanswered calls and dead numbers sit outside the calculation entirely, so a heavy voicemail day shrinks your denominator and pushes the true rate up while the against-dials figure barely moves. The deeper treatment, including what counts as a single campaign, is in our guide to abandonment rules for predictive dialing.
Most days, on most lists, the right move is to let the pacing algorithm run and resist the urge to steer. It reacts to answer-rate shifts faster than any supervisor watching a wallboard, and the worst-performing setups we audit are usually the ones carrying the most manual intervention.
The exception is when the algorithm has nothing truthful to learn from. Statistics need history, and there are specific moments when history is missing or misleading.
Steady lists with a day or more of dialing history, a stable agent count and answer rates that drift rather than jump. Set the dropped-call target with headroom below the regulatory cap (we run 2.5 percent or lower, not 3) and let the ratio float. Look at it hourly. Touch it rarely.
Fresh lists with no history, the first 30 to 60 minutes after a campaign starts or resumes, short windows at the end of a shift, and any day when the active agent count is swinging. A conservative maximum calls-to-agent ratio limits how wrong the first hour can be while real data accumulates. Lift it once the answer rate settles.
A cap is a guardrail, not a steering wheel. If you are changing it several times a day you are managing a symptom, and the cause is upstream in one of three places: the agent pool, the record supply, or the redial architecture. The next two sections cover the second and third, because that is where the quiet failures live.
Pacing gets blamed for problems it did not cause. The pattern we see most: agents idle, someone nudges the ratio up, abandons climb, and the actual fault is the campaign profile offering the dialer only a few hundred dialable records.
A campaign profile controls which records qualify for dialing and in what order: contact field filters, list sorting, dialing hours and schedule restrictions. Each rule is individually sensible. Stacked, they can quietly cut a 50,000 record list to a few thousand dialable records, and nobody notices because nobody looks at the combined effect. The usual suspects:
The check is simple and almost never automated: dialable records remaining, per campaign, against the size of the team. A predictive campaign needs a deep pool relative to its agent count, and on our floors we treat a thinning pool as an end-of-list condition well before the list is technically exhausted. If the campaign is barely dialing at all rather than dialing slowly, work through why a Five9 campaign stops dialing first; the causes overlap but the fixes differ.
Every disposition in Five9 can carry a redial timer that decides when a record becomes dialable again, and the sum of those timers decides what the dialer has to work with hour by hour. That makes disposition architecture a pacing input, whether or not anyone designed it as one.
The classic pattern looks like this. A fresh list answers well first thing, so the morning runs hot. No Answer is set to redial after 30 minutes, so by mid-morning the dialable pool is refilling with recycled no-answers, which pick up at a fraction of the fresh rate. The pacer responds correctly to the falling answer rate by raising the ratio, and abandon risk peaks exactly when everyone has stopped watching. Nothing malfunctioned. The timers built that curve.
The design work is deciding what each outcome means and how long reality takes to change it. A busy signal can come back in minutes. An unanswered ring deserves an hour or more, and a different daypart on the next pass. A voicemail you actually left should rest for hours or a day, depending on cadence strategy. A disconnected number should never return. Five9's disposition documentation covers the mechanics of attaching redial timers to disposition types; the mechanics were never the hard part. Most estates we open up are still running the timers configured at go-live by whoever happened to be in the room, and the answering machine outcomes are usually the worst offenders because they are the most common result on any consumer list.
Lists do not announce that they are finished. They fade, and pacing fades with them, because a tired list drags the answer rate down and pushes the ratio up against the abandon cap.
Penetration is measurable with three numbers per list: the share of records attempted at least once, the share that has reached its attempt cap, and connects per agent hour on the list compared with its first day. The third is the one that pays the bills. A list is effectively done when the next hour spent dialing it earns fewer connects than the next hour on whatever you would load instead. That is an economic definition rather than a technical one, and it usually arrives well before every record hits its attempt cap. Holding teams on exhausted lists to reach a round-number penetration target is one of the more expensive habits in outbound.
Two practical notes. First, measure per list, never blended across a campaign; a fresh list stacked on a dead one produces averages that hide both. Second, a naming trap: Five9 also offers a list dialing option called List Penetration Dialing, which dials only the first number on each record. It shares a name with the metric and nothing else. How you sequence, rest and retire lists is its own discipline, and we cover it in outbound list strategy.
An hourly pass over five numbers, owned by a named person.
Predictive campaigns are not set-and-check-on-Friday systems. On the floors we manage, a named person owns an hourly pass over a short list of numbers, and the discipline matters more than the dashboard it runs on.
| What we check | How often | What it tells us |
|---|---|---|
| Abandon rate, computed as abandoned divided by live answers | Hourly, and within the hour after any settings change | Whether headroom under the cap is real; a jump means the answer rate moved or the pool changed |
| Live answer rate trend | Hourly | The pacer's main input; a steady fall usually means recycled records or a tiring list, not a network problem |
| Occupancy | Hourly | Idle agents point to starvation or an over-tight cap; healthy predictive floors we run typically sit somewhere in the high 70s to mid 80s percent, depending on handle time |
| Connects per agent hour | Hourly, plus a daily close-out by list | The output that pays for everything; it falls before revenue does, and per-list views catch a dying list a day early |
| Dialable records remaining | Start of day and again by early afternoon | Whether filters and redial timers have already decided how your afternoon ends |
One habit worth stealing: recompute abandon rate from raw counts until you have verified what your report column divides by. Reporting tools offer several percentages that look like the regulatory one, and the wrong denominator is the most common serious error we find in outbound reporting.
The same handful of misconfigurations turns up in most of the Five9 estates we open, and none of them look like problems from the inside.
None of this is exotic, which is rather the point: it is configuration work, not a project, and fixing this class of problem is a large part of how engagements across our client portfolio have averaged around a 35 percent reduction in abandon rate. Go-live was supposed to be the start of the tuning phase. If that phase never happened on your estate, it is not too late to run it, and the first pass usually takes days rather than months.
It continuously estimates agent availability from recent statistics: live answer rate, average handle time, ring time and the number of agents ready or finishing calls. From those it sets a calls-to-agent ratio, dialing several numbers per expected free agent when answer rates are low and fewer when they rise, and it automatically backs the ratio off as the dropped-call percentage approaches the limit configured on the campaign.
Five9's product page says predictive works best with more than 10 active agents, and Five9's campaign documentation is stricter: never fewer than 15 agents consistently active in the campaign. Our implementation experience backs the 15, because smaller pools make the availability statistics too volatile: one long call swings the prediction and the pacer lurches between over-dialing and idling. Below that, power or progressive dialing usually produces better results with far less abandonment risk.
Abandoned calls divided by calls answered by a live person, not by total dials. Under both the FTC's Telemarketing Sales Rule and the FCC's rules, a call is abandoned when no agent connects within two seconds of the person's completed greeting, and the 3 percent threshold is measured per calling campaign over each 30-day period. Machine answers, unanswered calls and dead numbers are excluded from the denominator.
The FTC safe harbor at 16 CFR 310.4 and the FCC rule at 47 CFR 64.1200 both work to no more than 3 percent of calls answered by a live person, measured per campaign over each successive 30-day period, alongside requirements to ring at least 15 seconds or four rings and to play a recorded identification message when no agent is available in time. These texts get amended, so check the current versions and take proper advice where it matters.
As a guardrail, yes; as a steering wheel, no. Manual calls-to-agent caps earn their keep on fresh lists with no dialing history, in the first half hour after a campaign starts, and on days when active agent counts swing. Once the pacer has real answer-rate data, a steady list is usually better served by the algorithm with a dropped-call target set below the regulatory cap. Frequent intraday cap changes are a sign the real problem is upstream.
Check record supply before pacing settings. Campaign profile filters, dialing-hour and time zone restrictions, and disposition redial timers can shrink a large list to a small dialable pool, which starves the dialer no matter what the ratio says. Look at dialable records remaining, then at what share of the list is sitting in redial cooldown, then at the profile's filters. Pacing settings are the last place to look, not the first.
List penetration is how deeply you have worked a list: the share of records attempted at least once, the share at their attempt cap, and how yield has decayed since the list was fresh. A list is effectively done when an hour spent dialing it returns fewer connects per agent hour than a replacement list would, which usually happens well before every record reaches its attempt cap. Five9's List Penetration Dialing feature is a dialing mode, not this metric.
Hourly: abandon rate computed as abandoned divided by live answers, the live answer rate trend, occupancy, and connects per agent hour. At least twice a day: dialable records remaining, so filters and redial timers do not quietly end your afternoon. Recompute abandon rate from raw counts until you have verified what your report column actually divides by, and re-check everything within the hour after any settings change.
If your abandon rate whipsaws or your agents sit idle mid-afternoon, the cause is nearly always in the settings covered above. We are a Five9 Certified Implementation Partner, we run outbound floors daily, and you get a dedicated pod rather than a ticket queue. Tell us what the dialer is doing and we will tell you what we would change.
Talk to us