Predictive, Progressive, Preview and Power Dialers Explained

Choosing the wrong dialing mode is the most common reason a small campaign underperforms. Predictive dialing is not the advanced option that everyone should aspire to — below a certain size it is strictly worse than the simpler modes.

The four modes

Preview dialing

The agent sees the lead record first and chooses when to dial.

• Calls per agent: 1, initiated manually
• Agent idle time: highest
• Best for: B2B, high-value sales, complex accounts, regulated verticals
• Compliance risk: lowest — no abandoned calls are possible

Use when each conversation is worth enough to justify preparation. A £40,000 B2B deal warrants two minutes of research; a £30 consumer product does not.

Progressive dialing

The system dials automatically, one call per available agent, the moment the previous call is dispositioned.

• Calls per agent: exactly 1
• Agent idle time: moderate — agents wait through ringing and no-answers
• Best for: small teams, campaigns under ~8 agents
• Compliance risk: none — a call is only placed when an agent is already free

This is the right default for a new call center. It cannot abandon calls, and the efficiency loss versus predictive is small at low headcount.

Power dialing

Fixed multiplier — the system dials a set number of lines per available agent, typically 1.5 to 3.

• Calls per agent: fixed ratio you choose
• Agent idle time: low
• Best for: 5-20 agents with a reasonably stable answer rate
• Compliance risk: real — abandons occur whenever more calls connect than agents are free

Simple and predictable. You control the ratio directly and can tune it from the abandon rate.

Predictive dialing

The system continuously calculates how many lines to dial based on live statistics — current answer rate, average call duration, average wrap time, and how many agents will free up in the next few seconds.

• Calls per agent: dynamic, often 2-4
• Agent idle time: lowest
• Best for: 10+ simultaneously active agents
• Compliance risk: highest, and actively managed by the algorithm

Why predictive needs scale

The algorithm is doing statistical prediction. With three agents it is predicting from a sample of three, and the variance swamps the signal. The result is a dialer that alternately idles everyone and abandons calls.

Rough thresholds:

"Active agents" means simultaneously logged in and available on the same campaign — not your total headcount. Twenty agents split across four campaigns is four small campaigns, not one large one.

Abandon rate: the constraint that governs everything

An abandoned call is one where the dialer connected a live person and no agent was available. The person hears silence, or a recorded message, and hangs up.

This is not merely a courtesy issue:

• US (FTC Telemarketing Sales Rule): maximum 3% of answered calls, per campaign, per 30 days, with the agent connected within 2 seconds.
• UK (Ofcom): comparable 3% ceiling per campaign per 24 hours, with an information message required.
• Several other jurisdictions apply their own limits.

See the compliance guide for detail.

Tuning in practice

The correct approach is incremental and boring:

1. Start progressive. Establish baseline answer rate and average handle time.
2. Move to power at 1.5. Run a full day. Check abandon rate.
3. Step to 1.8, then 2.0, one step per day, watching abandon rate at each.
4. Stop increasing when abandon rate reaches roughly 2% — leave headroom below the 3% ceiling for variance.
5. Re-tune whenever the list changes. A new list with a different answer rate invalidates your ratio immediately.

In ViciDial, the relevant settings are Dial Method, Auto Dial Level, and the adaptive variants (ADAPT_HARD_LIMIT, ADAPT_TAPERED, ADAPT_AVERAGE) which manage the ratio automatically against a configured drop-rate target. See the installation guide.

Answering machine detection

AMD analyses the first seconds of audio to guess whether a human or a machine answered.

The trade-off is unavoidable:

Typical accuracy is 85-95%. Two consequences worth understanding:

False positives are expensive. A live prospect classified as a machine gets hung up on. They now have a missed call from a number that hung up on them — a strong complaint trigger and a reputation signal.

AMD creates silence. Detection requires listening for one to three seconds before connecting an agent. That silence is exactly what regulators class as a "silent call," and it counts against you under UK rules.

For small campaigns, consider turning AMD off entirely. Agents disposition voicemails in a couple of seconds, and you avoid both problems.

The list matters more than the algorithm

Pacing efficiency is bounded by data quality. A dialer cannot overcome:

• Disconnected numbers (raises dial attempts, lowers contact rate)
• Wrong time zones (calls outside legal hours — a compliance failure, not just inefficiency)
• Duplicates (repeated contact, complaint risk)
• Stale data (six-month-old leads convert far below fresh)

Clean the list before tuning the dialer. Operators routinely spend a week optimising pacing on a list that should have been rejected.

Where to go next

• ViciDial installation guide — configuring these modes.
• Compliance: TCPA, DNC, and GDPR — the abandon rate rules in full.
• Call center KPIs that actually matter — measuring pacing effectiveness.
• DIDs and caller ID reputation — why abandons hurt twice.