A predictive dialer is a bet. It dials more numbers than you have agents ready to answer, because most outbound calls end in no answer, a busy tone, or voicemail — and an agent who takes calls one at a time spends most of the day listening to ringback. The bet pays off when the over-dial cushion lands just right: by the time a call is answered, an agent has just freed up. It pays off badly when two people answer within seconds and there is one agent to take them. The surplus answered call, dropped back into silence, is an abandoned call — and the abandon rate is how regulators measure whether your pacing bet is conservative enough.
Devotel Orbit ships a predictive pacer that sizes that bet automatically, bounded by a hard ceiling: the FCC's 3% abandonment cap, with a tenant-set override that can only make it stricter. This guide walks through what the pacer actually does, what the ceiling setting means, and how to pick a dial mode without accidentally running a predictive campaign you didn't want.
What the pacer actually does
The pacing engine fires every 30 seconds. Each tick, for each active campaign in your workspace, it does the same arithmetic: how many agents are currently ready to talk, and how many calls should therefore launch right now.
The answer depends on the dial mode the operator picked per campaign:
- Preview — the engine dials nothing. An agent claims a contact, reviews the record, and clicks to dial it. Zero abandon risk exists by construction, because one human is attached to one number from claim to connect.
- Progressive — the engine dials at most one call per newly-ready agent. There is deliberately no over-dial cushion, so abandonment is structurally near zero. The cost is throughput: every ring phase is idle time for a waiting agent.
- Predictive — the engine launches roughly
idle agents × pacing ratiocalls this tick (floored to whole calls). Thepacing ratiois the over-dial multiplier, clamped between 0.8 and 3.0, starting at 1.2. A ratio of 1.5 with two agents ready launches three calls: the engine is wagering that at most two of them connect before an agent frees up. - Agentless — there is no agent leg at all. The batch plays a recorded message to answered recipients and is bounded by a concurrency limit, not by pacing math.
The predictive mode's agent-availability signal comes from the same real-time workforce state the queue dashboard reads — an agent toggled to available, with calls and wrap-up counted against them, stops being in the ready pool. When no agents are ready, the predictive engine launches zero calls. It never dials into an empty room, and a workspace whose headcount can't keep up simply sees the pacer throttle itself to nothing rather than abandon calls.
The abandon rate is the trade-off made visible
Over-dialing is what a predictive dialer is for, so some abandonment is the price of the throughput — but the price is regulated. The FCC hard-caps the abandonment rate at 3% of answered calls, measured over a rolling 30-day window: an answered call counts as abandoned when the recipient picks up and no agent connects to them.
Orbit's pacer continuously adjusts the pacing ratio against that ceiling using the campaign's own 30-day rolling abandonment rate:
- When the observed rate climbs above 80% of the ceiling, the ratio eases down by 5% per tick — the engine backs off before the legal boundary, not after.
- When the observed rate drops below 50% of the ceiling, the ratio nudges up by 2% per tick — reclaiming throughput the conservative pacing left on the table.
- When a campaign's 30-day rate actually trips the ceiling, the campaign is auto-paused. No further calls launch until an operator reviews the pacing settings and resumes it. Waiting for gentle EMA nudges to correct a real breach is not a defensible regulatory posture; stopping the dials is.
There are three dials, and they bind in this order: the operator-configured max_abandon_rate on the campaign (bounded by the FCC cap), the per-tenant ceiling override (also bounded by the FCC cap, and it can only tighten it), and the platform's own enforcement of the FCC cap as an absolute never-uncapped boundary. A tenant can decide to be stricter than the law. No tenant can configure their way past the cap, because the violation is not the tenant's to waive.
The abandon-ceiling setting: your workspace's posture
Until recently the FCC cap was the only ceiling available, and tightening it meant asking for an out-of-band change. The Settings → Voice dashboard now exposes it directly:
> "The FCC caps abandonment at {max}% — set a stricter target to dial more conservatively."
Off by default, the setting leaves the {max}% FCC cap in force workspace-wide. Enable it with a value of, say, 2% and every predictive campaign in your workspace paces as if 2% were the legal limit: the pacer's 80% warning band, the ease-down threshold, and the auto-pause all reference your number, and a value above the FCC cap is clamped back to the cap rather than honored. This is a tenant-owned compliance posture in the strict sense — the platform supplies the deterministic enforcement, the workspace decides where its own line sits, and the one thing no workspace can do is relax the federal boundary.
A stricter ceiling is worth setting when your calls carry consequences beyond a regulatory letter: a collections program with a consent-decree, outbound to vulnerable-population lists, or a brand where a dropped "hello, hello?" hurts more than a day of lost dialer throughput.
What this post does not cover — the sibling guide
Pacing is the how fast of outbound. It composes with three AI-assisted layers that answer different questions — when to dial each contact, which number presents to them, and how connected calls went — and those have their own walkthrough. The AI-assisted outbound dialing guide covers best-time-per-contact scheduling (which can't push a call outside the compliance envelope the operator sets), the caller-ID coverage report that names the uncovered area codes your number pool can't match, and post-call sentiment scoring at campaign level. The pacing layer described here is orthogonal to all three: best-time controls the dialing instant, the pacer controls how many dials happen at that instant, and both stay under the abandon ceiling.
Choosing a dial mode, honestly
The decision hinges on two facts about your operation: how valuable is agent idle time versus a dropped answered call, and how strict is the posture your list demands?
If your calls are high-stakes — the recipient is worth a pre-dial CRM read — stay on preview or progressive and accept the ring time. If throughput dominates and your workspace's ceiling has room, predictive earns its "expensive when wrong, invisible when right" place, and the pacer carries most of the tuning work. If nobody needs to talk at all, agentless broadcast skips the question.
Whichever you pick, the abandon ceiling is not a tuning knob to maximize against — it's the contract that keeps "we hit our throughput targets" from becoming a consumer-harm headline.
Frequently asked questions
What's the difference between predictive, progressive, and power dialing?
Progressive dialing launches at most one call per ready agent at a time — no abandon risk, but idle time during each ring phase. Predictive dialing over-dials by a multiplier of the ready-agent count, wagering that not every launched call connects, which buys throughput at the cost of some abandoned answered calls. "Power dialing" is the marketing label for the over-dial family; it describes the same over-dial cushion behavior predictive pacing implements, with the distinction usually being which vendor controls the multiplier.
What abandon rate do regulators care about?
In the US, the FCC's Telemarketing Sales Rule caps abandonment at 3% of answered calls, measured over a rolling 30-day window. An "abandoned" call connects a live recipient to nothing — no agent within the permitted window. Ofcom's corresponding cap in the UK is also 3%, with live monitoring expected. Orbit's pacer treats 3% as the absolute ceiling and lets tenants set a stricter workspace-wide value; it never lets a tenant relax the federal number.
Can I turn off the auto-pause if my campaign trips the ceiling?
No. Auto-pause is the only defensible answer to exceeding a regulator-set abandonment rate, and making it optional would defeat the control's purpose. The resume step exists so an operator can fix the pacing ratio or the agent pool — and only then continue dialing — rather than let the breach continue at a gentler pace while they read about it.
Does the ceiling apply to agentless or preview campaigns?
No. Pacing against the abandon ceiling only matters where answered calls can reach a busy agent pool — the predictive and progressive modes. Preview mode dials per-agent at click time and agentless mode plays a recording with no agent to abandon to, so the ceiling they need to respect is the one their own throughput budgets set.
---
Set the workspace abandon ceiling under Settings → Voice in the Orbit dashboard. The full campaign walk-through is in the outbound dialer campaign guide, the sibling AI-assisted outbound post covers best-time scheduling and sentiment scoring, and the IVR vs AI voice agents decision guide settles how answered calls route once they land. For the underlying telephony surface, see the programmable voice API.