The pacing math behind a predictive dialer that doesn't drop calls

Every predictive dialer advertises a "pacing ratio" setting. Most teams twiddle it until the abandon rate stops complaining and then leave it there. That works until it doesn't — a Monday morning, a campaign with unusual answer patterns, a new list — and suddenly your compliance team is on the phone.

The honest framing: pacing is a closed-loop control problem. You are trying to keep an agent pool saturated (every agent on a call) without exceeding a hard ceiling on abandon rate. The signal you're controlling against — time-to-first-hello — is noisy, non-stationary, and operator-dependent.

VOIPONE's pacing engine measures three things on a rolling 60-second window: observed connect rate, agent readiness distribution, and post-answer dead-air before agent pickup. From there it solves for a dial velocity that keeps the expected abandon rate under the TCPA ceiling with a safety margin — and decays back toward 1:1 as soon as the signal gets noisy.

The effect our customers see is counterintuitive: higher talk time and a lower abandon rate at the same time. That's because the engine is willing to dial less aggressively when the signal says "I don't know yet" — and more aggressively when it has high confidence the next call will land on a ready agent.

We'll go deeper in future posts on AMD tuning, post-call wrap-up modeling, and how we treat timezone-aware pacing as a first-class citizen.