What Is a Predictive Dialer? Operator's Guide

If you run an outbound call center without a predictive dialer, you are paying agents to listen to ringing. In manual or preview dialing, a large share of every agent hour goes to waiting — for the phone to ring, for someone to answer, through voicemails and disconnected numbers. A predictive dialer shrinks that idle time by managing outbound call volume algorithmically, so that an agent is connected to a live person as soon as they become available.

This guide is written by SIPNEX, an FCC-licensed carrier that provides SIP trunks for predictive dialers and VICIdial operations specifically. What follows is not a software review — it is the carrier-side perspective on how predictive dialers work, what they demand from the underlying trunk, and why operators who blame their software for bad performance are often looking at the wrong layer.

What a predictive dialer does

A predictive dialer is software that places outbound calls on behalf of a pool of agents, using statistical algorithms to determine how many calls to place simultaneously so that a live human is ready to connect to an agent the moment that agent finishes their current conversation. The dialer does not wait for an agent to become free before initiating the next call. It predicts when the agent will become free — hence the name — and begins dialing in advance so the connection is ready the instant the agent’s line opens.

The math behind predictive dialing is more sophisticated than most operators realize. The dialer maintains a real-time model of several variables: the current answer rate (what percentage of outbound calls are being picked up by a live person), the average call duration (how long a typical conversation lasts once connected), the average wrap-up time (how long the agent spends on after-call work before they are ready for the next call), and the number of agents currently available or about to become available.

From these inputs, the dialer calculates a dial ratio — the number of simultaneous outbound calls to place per available agent. If the answer rate is 20 percent (one in five calls gets picked up), the dialer might place five calls per available agent. If the answer rate drops to 10 percent, it dials more aggressively — maybe eight or ten per agent. If the answer rate spikes to 40 percent, it backs off to two or three per agent to avoid overwhelming the agent pool.

The target is to keep the abandon rate — calls that are answered by a live person when no agent is available to take them — at or below 3 percent. The FCC rule, 47 CFR 64.1200(a)(7), caps abandonment at 3 percent of telemarketing calls answered live, measured per calling campaign over each 30-day period. A call counts as abandoned if no live agent is on the line within two seconds of the called person’s completed greeting. Exceeding the cap violates the FCC’s TCPA rules. The dialer’s algorithm constantly adjusts the dial ratio to maximize agent utilization while staying under that ceiling.

In VICIdial specifically, this behavior is controlled through campaign settings: the dial level (auto or manual ratio), the adaptive dial methods (which pace against a Drop Percentage Limit you set), and the hopper (the queue of leads ready to be dialed). The adaptive algorithm is highly configurable, and according to VICIdial’s own PREDICTIVE.txt documentation it recalculates the dial level every 15 seconds based on actual performance.

How predictive dialing works at the SIP level

Understanding what happens at the trunk level explains why carrier choice matters so much for predictive dialer performance.

When VICIdial (or any predictive dialer) decides to place a call, it sends a SIP INVITE through the configured trunk to the carrier. Each simultaneous call attempt is a separate SIP session — a separate INVITE, a separate call setup, a separate media stream if the call is answered. If the dialer is running 10 agents at a 3:1 ratio, up to 30 call attempts can be in progress at once. If the answer rate drops and the ratio increases to 5:1, that becomes up to 50 simultaneous sessions. A 100-agent floor at a 4:1 ratio can generate 400 concurrent SIP sessions.

Each of those sessions requires the carrier to authenticate the trunk, validate the caller ID, perform STIR/SHAKEN signing, route the call, and handle the media. If the call is answered, an RTP audio stream is established. If the call goes to voicemail and the dialer detects it via AMD (Answering Machine Detection), the session is torn down. If the called number is disconnected, the far end typically returns a SIP error such as 404 Not Found, sometimes after playing an intercept recording as early media, and the session ends.

The carrier’s capacity to handle this burst traffic without degradation determines your dialer’s effective performance ceiling. Here is where the problems show up:

Post-Dial Delay (PDD). This is the time between the SIP INVITE leaving your dialer and the called party’s phone starting to ring. On a good carrier, PDD is under 3 seconds. On a congested or poorly peered carrier, it can reach 5 to 8 seconds. In a predictive dialing environment, excessive PDD cascades through the algorithm — the dialer’s predictions about when agents will become free are based on expected call setup times. If PDD varies wildly, the algorithm makes bad predictions: too many calls connect at once (agents overwhelmed, abandon rate spikes) or too few connect (agents idle, efficiency drops).

Concurrent channel limits. Many SIP trunk plans cap concurrent channels per trunk or per account. A 50-agent operation running a 4:1 ratio needs 200 concurrent channels. If the carrier caps at 50, the dialer cannot execute its algorithm — calls beyond the cap are rejected, often with SIP 503 (Service Unavailable), throttling the operation to a fraction of its potential. On SIPNEX, concurrent channels are unlimited. Your capacity is your bandwidth and your server.

STIR/SHAKEN attestation level. Every one of those SIP INVITEs gets signed by a carrier. If your provider does not hold its own certificate, the signing happens upstream, and the upstream carrier can give full (A) attestation only if it can verify your right to use the calling number through that provider. When it cannot, the calls go out with partial (B) attestation. Analytics engines treat B attestation as a lower-trust signal; no carrier publishes a specific answer-rate gap.

Codec and media quality. G.711u (ulaw), which is uncompressed, is a common primary codec for US VICIdial trunks. Each G.711 call uses approximately 85 kbps of bandwidth. At 200 concurrent calls, that is 17 Mbps of sustained audio traffic in each direction. The carrier’s media infrastructure must handle this without packet loss or jitter. Oversubscribed media gateways introduce audio problems the dialer cannot compensate for — choppy or clipped audio and dropouts — all of which cause agents to lose calls and consumers to hang up.

Predictive vs progressive vs preview dialers

Predictive dialing is not the only mode, and it is not always the right one. Understanding the three primary modes helps you match the technology to the campaign.

Predictive dialing places multiple calls per available agent based on algorithmic predictions. Highest efficiency — agents spend the maximum percentage of their time talking to live humans. Highest compliance risk — the algorithm can miscalculate and cause abandoned calls. Best for: high-volume B2C campaigns with large lead lists, commodity offers, appointment setting, collections. In VICIdial, a compliance-minded setup uses an adaptive dial method with the Drop Percentage Limit at or below 3 percent. ADAPT_HARD_LIMIT is the strictest of the three adaptive methods; ADAPT_TAPERED lets the drop rate run over the limit early in the shift.

Progressive dialing (sometimes called power dialing) places exactly one call per available agent. When an agent becomes available, the dialer places one new call for that agent, so it cannot abandon calls by over-dialing (though answering-machine detection that holds a live answer past two seconds still counts). Lower efficiency than predictive — agents wait through each ring cycle — but far less exposure to the FCC’s 3 percent cap. Best for: regulated industries (healthcare, financial services), campaigns with compliance-sensitive leads, and any operation where a single abandon-rate violation could be catastrophic.

Preview dialing shows the agent the lead record before the call is placed. The agent reviews the information — name, account history, previous interactions, notes — and decides whether to dial or skip. The agent initiates the call manually. Lowest efficiency of the three modes because every call includes a human review step. But the highest quality conversations because the agent is prepared. Best for: complex sales (insurance, real estate, B2B), high-value leads where personalization matters, and any campaign where the conversion value justifies the lower dial volume.

Many operations use more than one mode across different campaigns. A collections shop might run predictive for right-party-contact campaigns (high volume, low touch) and preview for settlement negotiation calls (high value, preparation required). VICIdial supports all three modes and allows switching between them per campaign without changing the underlying trunk configuration.

Why your carrier matters more than your dialer software

This is the point many operators miss, and it is why this article takes the carrier perspective rather than comparing software. VICIdial is a capable, long-running open-source dialer with a well-tested adaptive algorithm, but it — or any dialer — is only as good as the SIP trunk feeding it. Four carrier-layer problems undo dialer tuning:

Channel caps invalidate the pacing. If the carrier returns 503 errors because you hit a cap, the dialer’s predictions break — it expected those calls to connect or fail normally, not get blocked at the carrier. Agent utilization and drop-rate calculations both go wrong.

Partial attestation is outside the dialer’s reach. VICIdial can optimize pacing, AMD, and lead order. It cannot change the trust signal a B-attested call carries to the recipient’s carrier and its analytics engine. That is decided by who signs the call.

Variable PDD makes the predictions wrong. Adaptive dialing assumes a fairly consistent call setup time. If PDD swings between 2 and 8 seconds, the dialer over-dials on some cycles (abandons) and under-dials on others (idle agents), and never converges on the right pace.

Poor media loses the calls you connected. An agent reaches a live person, but the audio is choppy or drops out; the consumer says “hello” twice and hangs up. The campaign looks worse than it is — not because the leads or agents are bad, but because the trunk is degrading the audio.

SIPNEX’s answer to those four: unlimited channels, a sub-3-second PDD target for US outbound, A-level STIR/SHAKEN attestation signed with our own SP certificate on numbers we verify, and media capacity sized for high-concurrency dialing.

Choosing a carrier for your predictive dialer

Based on everything above, here is what to evaluate when selecting a carrier for your dialing operation.

Concurrent channel policy. Ask explicitly: is there a per-trunk or per-account concurrent channel limit? If the answer is anything other than “unlimited” or a number higher than your peak predicted concurrency, the carrier will throttle your operation.

Post-dial delay. Ask for average PDD data. Under 3 seconds is acceptable. Under 2 seconds is good. If the carrier cannot tell you their PDD metrics, they probably do not monitor them, which means they do not optimize for them.

STIR/SHAKEN attestation. Ask the critical question: do you hold your own STIR/SHAKEN Service Provider certificate, or does an upstream carrier sign your calls? A carrier that signs with its own certificate and verifies your right to use your numbers can give A-level attestation. If signing happens upstream, ask how the upstream carrier verifies your numbers — without that verification, calls go out at B. Attestation is one of the few answer-rate factors your carrier choice controls directly.

Codec support. Confirm G.711u (ulaw) support as primary. G.729 as fallback. Opus if your platform supports it. Ask about transcoding — does the carrier transcode between codecs, or does it pass through natively? Transcoding adds latency and can degrade quality.

Pricing transparency. If you cannot see rates before talking to sales, you are paying for their sales team. Per-minute rates fall with volume and depend on destination mix: SIPNEX’s published tiers run from $0.025–$0.030 per minute under 100K minutes a month down to as low as $0.005 at 10M+. Per-channel fees are a red flag — they charge for capacity rather than actual usage.

Support depth. Call the support number and ask a VICIdial-specific question: what codec and DTMF mode do you recommend for VICIdial AMD detection? If they cannot answer without escalating, their support team does not understand your workload.

Frequently asked questions

SIPNEX is an FCC-licensed carrier that signs calls with its own STIR/SHAKEN certificate, with unlimited channels and a sub-3-second PDD target for US outbound. Request a dialer-grade trunk or see our published rates.