Picture a sales floor at ten past nine. Twelve operators are logged in, and half of them are staring at a screen waiting for the next call. Now picture the same floor at four in the afternoon, when the dialer suddenly connects three people at once and two of them hear silence before someone picks up. Same team, same list, same software. What changed is the share of people who answer, and a dialer that does not notice that change will either waste your operators' time or annoy your prospects. That is the whole story of predictive dialing in two scenes.
What a predictive dialer actually does
A manual dialer waits for the operator. A progressive dialer calls one number when one operator is free. A predictive dialer goes further: it calls more numbers than there are free operators, because most calls never turn into a conversation. Some ring out, some go to voicemail, some numbers are dead, some people simply decline.
The software estimates how many of the calls it places will end up with a live person, and it times the dialing so that a human answers at roughly the moment an operator becomes free. When the estimate is good, the operator moves from one conversation to the next with only a short gap. When it is bad, you get one of two problems: idle operators or abandoned calls.
Calls per operator: the setting that matters most
The core number is the dialing ratio, often called calls per operator. If it is set to 1, the system behaves like a progressive dialer. If it is set to 3, then for every free operator the system has three lines ringing.
Imagine that, on a given morning, one in three dialed numbers produces a live person. A ratio of 3 would then deliver about one answered call per free operator. That is the sweet spot. Now imagine the answer rate drops to one in five, because it is lunchtime and nobody picks up unknown numbers. The same ratio of 3 now delivers fewer answers than you have operators, and people wait. If the answer rate climbs to one in two, the same ratio delivers an overflow: more humans on the line than operators to talk to them.
This is why a fixed ratio is a guess dressed up as a setting. The right value depends on the list, the region, the time of day and even the weekday.
Abandon rate: the cost of getting it wrong
An abandoned call is a call where a person answered and no operator was free to speak to them. From the prospect's side it looks like silence, a click, or a recorded message from a company they have never heard of. It is the most visible failure of a predictive dialer.
There is a direct trade-off. Push the ratio up and operators spend more of their shift actually talking, but the abandon rate climbs. Pull it down and the abandon rate falls, but operators wait longer between calls. Nobody can remove this trade-off. What you can do is choose where to sit on the curve and make sure the system holds that position.
Many markets have rules or industry codes that limit how many answered calls may be abandoned, and what must happen when one is. Check what applies where you call and where your prospects live, and set your own internal target below that limit rather than at it. A margin gives you room on the bad hours.
Why pacing has to adapt by the hour
Answer rates are not flat across a day. A few patterns show up in almost every outbound operation, though the exact shape depends on your market:
- Early morning: many people are commuting or not yet at their phone, so answer rates can be uneven.
- Late morning: often a steady stretch, especially for B2B lists, where decision makers are at their desks.
- Lunch: phones are in pockets or on tables, and unknown numbers get ignored more often.
- Late afternoon and early evening: for consumer lists this is frequently the best window, as people are home and willing to talk.
Your own side moves too. Operators go on breaks, a training session pulls three people off the floor, a long conversation ties up someone for twenty minutes. The number of free operators changes minute by minute, and the dialer has to react to that as well, not only to the answer rate.
So a ratio you tuned on Tuesday morning is wrong by Tuesday afternoon. Good pacing is a feedback loop: measure recent answers, measure free operators, measure abandoned calls, then adjust the number of calls in flight. Do it continuously, not once a day.
Floor, ceiling and line cap
Automatic pacing still needs guardrails. A floor stops the ratio from dropping so low that operators sit idle even when conditions are fine. A ceiling stops it from rising to a level where one lucky streak of answers produces a pile of abandoned calls. A line cap limits the total number of simultaneous outbound calls so you never exceed what your telephony channels or carrier agreement can carry. And an emergency stop lets a supervisor halt dialing immediately when something looks wrong, such as a bad list, a wrong script or a broken caller ID.
Answering machine detection changes the maths
Voicemail is the quiet distortion in all of this. If the dialer counts a voicemail greeting as a live answer, it sends it to an operator, who then listens to a beep and hangs up. That wastes the very time the dialer was supposed to save. If it wrongly treats a real person as a machine and drops the call, you have created an abandoned call by a different route.
Detection is therefore a trade-off between speed and certainty. The longer the system listens, the surer it is, but the longer a real person waits in silence. Setting a time limit per queue makes sense, because a B2B queue and a consumer queue have different greeting styles and different patience. The outcome should be clear: human, voicemail, unavailable or invalid number, each handled differently by your retry rules.
A practical checklist for your next review
- Pull the answer rate by hour for the last few weeks, per queue, not as one average.
- Compare it with idle time per operator in the same hours. Long waits with a low ratio mean you are too cautious.
- Check abandoned calls by hour. Spikes usually follow a jump in answer rate or a sudden drop in free operators.
- Set a floor, a ceiling and a line cap for each queue, and write down who may change them.
- Review your detection time limit by listening to a sample of calls labeled as voicemail.
- Make sure the calling windows, retry ladders and attempt limits match when your prospects really pick up.
- Test the emergency stop before you need it.
Where AI fits into pacing
Everything above assumes humans are the bottleneck, and in a classic setup they are. That is also where pacing gets hard: people take breaks, vary in speed and cannot absorb a sudden burst of answers. At Alphacall the dialer, with auto-pacing based on calls per operator and the guardrails described above, sits inside a full contact-center engine, and the AI voice agent can take the conversations that would otherwise be lost to a burst or to a quiet hour. It can uncover needs, handle objections and pass the call to a human colleague with the context already in place. The people on your team then spend their time on the calls that truly need them, and the dialer has more room to keep abandoned calls low.
See it on a live call
Settings and charts only tell you so much. The clearest way to judge a dialer is to hear how a call connects, how quickly it detects a voicemail, and how a handover feels to the person on the line. If you would like to see how Alphacall handles pacing and a real sales conversation, ask for a live call demo and bring your toughest queue.