VICIfast
All guides

Predictive dialer guide — what it is, when to use it, how to tune it

Complete operator guide to predictive dialing — math, TCPA compliance, tuning, and when manual wins.

A predictive dialer places more outbound calls than agents currently available, predicting some won't connect (busy, no answer, voicemail). The goal: keep agents talking continuously, not sitting idle waiting for the next pickup.

This is the most efficient outbound dialing method when configured right. It's also the most regulated — the FCC caps abandonment at 3% specifically because predictive dialers, when poorly tuned, drop live calls.

How predictive dialing actually works

Walk through a simple scenario:

You have 10 agents. Dial level set to 2.0. The dialer places 20 calls simultaneously.

Three buckets emerge from the 20:

  1. No connect (~50% on cold lists): busy, no answer, voicemail detected by AMD. The dialer hangs up these.
  2. Connected to a person — within drop SLA window (typically 2 seconds): routed to an available agent.
  3. Connected to a person — but no agent available within the drop SLA: abandoned. Counts toward the FCC's 3% cap.

When tuned right, ~10 connects align with 10 available agents. When over-tuned, abandons spike. When under-tuned, agents idle.

The math, formally

expected_connects = dials × connect_rate
expected_drops    = max(0, expected_connects − available_agents)
abandon_rate      = expected_drops / expected_connects

Goal: abandon_rate ≤ 3%. Variable: dials (i.e. dial level).

If your steady-state connect rate is 25%, with 10 agents available, the dial level that perfectly matches is 4.0 (40 dials × 25% = 10 connects). At 4.0, prediction noise will sometimes overshoot (15 connects, 5 abandoned, 33% abandon — way over the cap).

In practice, operators run dial level lower than the theoretical math says, to leave headroom for connect-rate variance. Typical: 1.5-2.5.

When predictive wins

  • High volume outbound (>200 calls/agent/day)
  • Stable, large lists where prediction can train on real data
  • Cold or warm B2C: low connect rates make prediction matter
  • Drop SLA enforced + monitored

When manual wins

  • B2B prospecting (low volume, high value, idle agent fine)
  • Hot leads (you want the call through, abandons are unacceptable)
  • Compliance-heavy verticals (collections in some states; political non-exempt calls)

Tuning, step by step

Day 1: Baseline

  • Set dial level = 1.0
  • Run for 1-2 hours with full agent pool
  • Observe: agent idle %, drop %, calls per agent per hour

Day 1, hour 3+

If idle is over 30% AND drops stay under 1%, raise dial level to 1.5.

Day 2

If still idle and drops still low, raise to 2.0.

If drops spike past 2%, lower 0.2 and watch.

Steady state

Most teams settle at 1.5-2.5. Some aggressive teams push to 3.0+ with tight monitoring + manual intervention when abandonment climbs.

Adaptive dial methods

VICIdial has dial methods that auto-tune:

  • ADAPT_HARD_LIMIT: dialer adjusts but never exceeds your set max
  • ADAPT_AVERAGE: smooths over short spikes
  • ADAPT_TAPERED: aggressive adjustment

Adaptive is great if your connect rate varies through the day (mornings vs afternoons) but it can overshoot during sudden agent dropouts (lunch breaks, etc.).

TCPA compliance

The FCC 3% cap is a 30-day rolling window per campaign. You can blow it on a single bad day and recover; you can't sustain.

Real-time monitoring is critical. VICIdial's Real Time Main Report shows abandon % live; if you see it climb past 2%, drop dial level immediately.

Common mistakes

1. Treating agent count as the variable

Adding agents doesn't reduce abandonment proportionally — adaptive dialers spread across more agents. The variable that matters is dial level.

2. Running predictive on small lists

Prediction needs many calls to train. Lists under ~5K leads with predictive often see wild abandonment swings. Use progressive or manual instead.

3. Ignoring connect rate variance through the day

Cell phones answer more in afternoons than mornings. Predictive tuned for noon connect rate over-dials at 9am, under-dials at 3pm. Adaptive helps.

4. Over-recycling

Recycling NA and BUSY leads aggressively spikes connect rate temporarily, breaks predictive's calibration. Spread recycling.

Hardware sizing

Predictive dialing is RAM-hungry. vicidial_live_agents + vicidial_auto_calls carry a lot in memory. Budget:

  • 200MB / agent on top of base
  • 100 agents on Pro plan (8GB RAM) is the practical max
  • 200+ agents need Business plan (16GB) or Scale (32GB)

Related guides