How call abandonment is controlled in predictive dialing

What counts as an abandoned call

An abandoned call is not a call nobody picked up. It is the opposite: someone answered and there was no one on the other end. A predictive dialer places more calls than its floor can handle — that is its entire reason to exist — betting that most will not be answered. When the bet goes wrong and more people answer than predicted, someone sits in silence until the system hangs up.

That person will not answer your next call the same way. And the metric that counts them is the only one on the dialing board with legal consequences in several markets: the US Telemarketing Sales Rule caps it at 3% of calls answered by a live person, measured over thirty days and per campaign.

Telling it apart from what it is not keeps the measurement honest:

• Abandoned — a person answered, no agent was free, the system hung up.
• No answer — nobody picked up. Does not count.
• Voicemail — a machine answered. Does not count, if the detector was right.
• Hung up by the contact — answered, an agent was there, and they hung up anyway. That is a different conversation and a different problem.

The answering machine detector is the fragile part of that list. An aggressive detector classifies people as voicemail and drops them without ever appearing in an abandonment metric. The number looks fine and the campaign is throwing contacts away.

The three levers

The abandonment target

Every predictive dialer has a configurable target, and it is the first thing to check because it governs the algorithm. Set to 3%, the dialer adjusts how many calls it places per free agent so the abandonment ratio stays there. Set to 8%, it will do exactly what you asked.

What gets forgotten: the target is met on average, over the window you measure. A 3% target measured across a whole month allows 9% mornings offset by 1% afternoons. If the rule that applies to you is measured daily, the configuration that satisfies it is a different one.

The wait window

This is how much silence the system tolerates before hanging up. Shortening it lowers counted abandonment — the call drops before the clock classifies it — without improving anything for the person who answered. It is the lever most often used to dress up the number and the one that fixes the problem least.

Lengthening it does the opposite: more calls reach an agent who freed up at the last second, and abandonment falls for real. The price is that anyone waiting four seconds in silence has already worked out that a machine called them.

Two seconds is usually where the balance sits, and it is a decision better made by listening to the recording than by reading the spreadsheet.

The size of the connected floor

This is the lever almost nobody touches and the one that pays best. A predictive algorithm aims better the more agents it has: with forty agents, the variance in how many calls get answered at once is proportionally far smaller than with six. It is the same effect that lets a large telephone exchange need fewer lines per user than a small one.

The practical consequence is counterintuitive: splitting a floor of thirty agents into three campaigns of ten makes abandonment worse on all three at once, with nobody having changed a single setting. If you have small campaigns with high abandonment, the first question is not how to tune the dialer, but whether they need to be separate campaigns at all.

Below eight agents, predictive stops having a statistical edge over progressive — and progressive abandons no calls at all.

Why cutting the pace comes last

Cutting the pace always works. Dial less, abandon less, and at the extreme — one call per free agent — abandonment is zero, because you have just turned your predictive dialer into a progressive one.

It comes last because it is the only lever paid for in contacts. The other three correct how the dialer works; this one asks it to work less. If pace is still the problem at the end of the tuning, the honest answer is usually that the campaign needs more agents or fewer records, not another setting.

How to measure it without fooling yourself

Three things that make the number on the board lie:

1. The denominator. The limit is defined over calls answered by a live person, not calls placed. Measuring over calls placed divides by a much larger number and returns a reassuring, false percentage.
2. The window. A thirty-day average hides the peaks. If the rule is measured daily, measure it daily.
3. The answering machine detector. A miscalibrated detector pulls people out of the denominator. Listening to a sample of what it called voicemail, once a month, is worth the time.

In PulseDial, abandonment is shown per campaign and live, next to the cost per contact for that same campaign, because they are two faces of one decision: you can nearly always lower one by raising the other, and the useful question is which of the two is costing you more today.