Dialing Guides
Which dialer mode fits which campaign, how to set pacing without breaching your abandonment ceiling, and why connect rate is a list and reputation problem long before it is a software problem.
In short
There are four outbound dialing modes. Preview shows the agent the record and lets them dial. Progressive dials one number per free agent. Power dials a fixed number per free agent. Predictive forecasts availability and dials several ahead of the agents.
Talk-time efficiency rises left to right; abandonment risk and regulatory exposure rise with it. Only predictive dialing generates meaningful dropped calls, which is why regulated verticals usually stop at progressive.
The four dialer modes
| Mode | Numbers per free agent | Abandonment risk | Best for |
|---|---|---|---|
| Preview | Agent decides | None | Complex, high-value or sensitive calls needing preparation |
| Progressive | 1, after the agent is free | None | Regulated verticals; smaller, higher-quality lists |
| Power | Fixed ratio, e.g. 3 | Low to moderate | Predictable lists where simplicity beats optimisation |
| Predictive | Forecast, dials ahead | Material | Large lists, clean data, high agent counts |
Predictive dialing needs volume to work. Its forecast is a statistical model of agent availability, and below roughly 8–10 agents on a campaign the model has too little signal — you get either idle agents or dropped calls. Small teams are usually better served by progressive.
Predictive vs progressive
This is the decision most outbound teams actually face, and it is a compliance decision disguised as an efficiency decision.
Progressive dialing cannot abandon a call, because it never dials a number that does not already have an agent waiting. Predictive dialing can, by design — it is dialing ahead of availability, and sometimes the forecast is wrong. Under US outbound rules a call answered by a live person where no agent connects within two seconds of the greeting is a dropped call, and dropped calls are capped.
How to choose
Use progressive if you operate in a regulated vertical, your list is small or expensive, your conversations are consultative, or you have fewer than about ten agents on the campaign. Use predictive if you have volume, clean data, enough agents for the forecast to be stable, and the monitoring in place to hold your abandonment rate inside its ceiling.
Setting the pacing ratio
The pacing ratio is how many numbers a predictive dialer places per available agent — 2:1, 2.5:1, and so on. The common mistake is tuning it against talk time. Talk time will always improve as you raise the ratio, right up until you are in breach.
Tune against the abandonment ceiling instead:
- Decide your ceiling and set it below the regulatory limit. The gap is your margin for a bad forecast day.
- Start conservative — around 1.5:1 on a new campaign with unknown list quality.
- Raise in small increments, holding each for a full shift so you see the whole intraday pattern rather than one quiet hour.
- Stop at the ratio where abandonment sits at roughly two-thirds of your ceiling.
- Re-check after any list change. New list, new answer behaviour, new correct ratio.
Pacing is per campaign, not per account. A collections list and a warm inbound-lead follow-up list will not share a correct ratio.
Answering machine detection
AMD decides whether a call was answered by a person or a voicemail system, so live answers reach agents and machines get a voicemail drop. It saves roughly 30 seconds per unanswered call, which is substantial across a campaign.
It is also a probabilistic judgement made in the first second or two of audio, and it is wrong sometimes. The tuning trade-off:
- Aggressive AMD — classifies faster, keeps agents busier, and hangs up on more real people. Someone who answers slowly or quietly gets treated as a machine.
- Conservative AMD — passes more machines through to agents, costing idle time, but rarely abandons a live person.
Verticals where the person answering may be elderly, unwell or distressed should run conservative and accept the cost. A hang-up on a real person is not a neutral efficiency loss — it is the behaviour that generates complaints, and complaints drive caller ID labeling.
Raising connect rate
Connect rate is the share of dial attempts that reach a live person. Teams usually look at dialer settings first. That is almost never where the problem is.
| Lever | Why it moves the number |
|---|---|
| Caller ID reputation | A labeled number collapses answer rates regardless of everything else. Check this first. |
| List age | Speed to lead dominates. A lead contacted in minutes converts at a different order of magnitude from one contacted days later. |
| Time-of-day strategy | Answer behaviour varies by segment. Dialing everyone at 10am wastes the windows where each segment actually answers. |
| Number hygiene | Disconnected and reassigned numbers burn attempts and inflate your dial volume for no connects. |
| Local presence | Area-matched caller ID lifts answer rates, but rotating aggressively looks like spoofing to carrier analytics. Use it carefully. |
| Attempt cadence | Six attempts across varied days and dayparts beats six attempts in one afternoon. |
| Dialer settings | Last, and smallest. Pacing changes how efficiently you work the list, not how willing people are to answer. |
Caller ID reputation
Carriers and handset apps score inbound calls and label the ones they judge unwanted. A labeled number does not ring the way an unlabeled one does — many are silenced entirely. When connect rate falls off a cliff overnight with no campaign change, this is almost always the cause.
Reputation is driven by behaviour the analytics engines can observe:
- Short-duration call ratio — lots of sub-6-second calls looks like scanning
- Complaint volume against the number
- Dial velocity from a single number
- Calls to disconnected and reassigned numbers
- STIR/SHAKEN attestation level
Remediation is monitoring, branded calling registration, sensible number rotation, and removing the behaviours that generated the score. Rotating numbers without fixing the behaviour just spreads the reputation problem across more numbers. More on scam likely labeling →
Timing and calling windows
The federal outbound window is 8am to 9pm in the called party's local time, which means a national list needs timezone-aware dialing rather than one schedule. Several states impose narrower windows, and some restrict weekends. Build the constraint into the campaign rather than into a policy document.
Within the legal window, timing is empirical. Rather than adopting someone else's best-time-to-call chart, segment your own list and measure connect rate by daypart for each segment. B2B and B2C answer behaviour barely resemble each other, and a shift worker's reachable window is not a salaried office worker's.
List segmentation
One list dialed uniformly is the most expensive way to run outbound. Segmentation lets each group get the mode, cadence and script that suits it.
- By age — fresh leads to progressive or preview for immediate, prepared contact; aged leads to predictive for volume.
- By source — consent provenance differs, and so does intent. Keep sources separable so you can suppress one without rebuilding the list.
- By attempt history — a record on attempt six should not be treated like a record on attempt one.
- By score — work the highest-likelihood records in the best windows, not whatever the import order happened to be.
- By timezone — a compliance requirement, not an optimisation.
Deep-dive guides in progress
Each of these takes one topic above into configuration-level detail, including the settings and the numbers to watch while you change them.
- Predictive vs progressive: a decision worksheetIn progress
- Pacing ratio tuning, shift by shiftIn progress
- Answering machine detection: tuning by verticalIn progress
- Connect rate diagnostics: a fault treeIn progress
- Caller ID reputation recovery: a 30-day planIn progress
- Best time to call: measuring it for your own listIn progress
- Lead list segmentation and cadence designIn progress
- Local presence dialing without looking like a spooferIn progress
These are listed without links on purpose — we would rather show you what is coming than send you to a page that does not exist yet. Tell us which to prioritise.
Frequently asked questions
What is the difference between a predictive dialer and a progressive dialer?
A predictive dialer forecasts agent availability and dials several numbers per available agent at once, maximising talk time at the cost of some abandoned calls. A progressive dialer dials one number per agent only after that agent is free, which eliminates abandonment risk but produces fewer connects per hour.
Predictive suits high-volume lists with clean data and enough agents for the forecast to stabilise. Progressive suits smaller lists, regulated verticals and higher-value conversations.
What is a good pacing ratio for a predictive dialer?
There is no portable number — the correct ratio is whichever one holds your abandonment rate at roughly two-thirds of your ceiling for that specific list. Start near 1.5:1 on an unknown list and raise in small increments, holding each for a full shift.
Tuning pacing against talk time instead of against abandonment is how campaigns end up in breach.
Why has my outbound connect rate dropped suddenly?
A sudden drop with no campaign change is almost always caller ID reputation — your numbers have been labeled, so they are being silenced or warned against before they ring.
Check labeling status first, then STIR/SHAKEN attestation, then whether your carrier is still in good standing in the Robocall Mitigation Database. Pacing and AMD settings are the last place to look, not the first.
Does answering machine detection hang up on real people?
Sometimes, yes. AMD is a probabilistic judgement made from the first second or two of audio, and aggressive settings misclassify people who answer slowly or quietly as machines.
In verticals where the person answering may be elderly, unwell or distressed, run conservative settings and accept the extra agent idle time. Hang-ups on live people are what generate the complaints that drive caller ID labeling.
How many times should you attempt a lead?
Most programmes see the return flatten somewhere between the fifth and eighth attempt, but the pattern matters more than the count: attempts spread across different days and dayparts materially outperform the same number made in one afternoon.
Measure it on your own list. Attempt curves differ enough by vertical and list source that a borrowed number is not worth much.
Four dialing modes, one platform
DialedIn runs preview, progressive, power and predictive dialing on the same campaign engine, so you can change mode as a list matures instead of rebuilding it.