What Is a Ghost Call? How Predictive Dialers Create Silent Calls and How to Stop Them in 2026

Ghost calls are costing your call center real money. Not in a vague, hard-to-measure way. In a $120,000-per-month, agents-sitting-idle, live-customers-hanging-up way.

If you run an outbound center on ViciDial, Asterisk, or GoAutoDial, you've almost certainly seen it: a call connects, an agent picks up, and there's nothing on the line. Dead air. The contact hangs up. Your agent moves on. That was a ghost call — and it happened because your answering machine detection made the wrong call.

This article explains exactly what ghost calls are, why predictive dialers produce them, what the compliance exposure looks like in 2026, and how to stop them.

What Is a Ghost Call?

A ghost call is a connected call with no audio from the dialing side. The contact answers, hears silence, and ends the call. From their perspective, someone called and said nothing. From yours, a live connection was wasted.

Ghost calls aren't random. They're a direct output of how legacy AMD works.

When your dialer's AMD detects silence at the start of a call, it assumes a voicemail system is about to play a greeting. So it holds the call and waits. If a live human answered and paused before saying "hello," the AMD has already flagged that call as a machine. The agent never gets connected. The contact hears nothing. Ghost call created.

That's the core failure of silence-based AMD: it uses the absence of audio as its primary signal. Silence is not a reliable indicator of anything. Voicemails start with silence. Humans start with silence. The logic collapses under real-world conditions.

How Predictive Dialers Produce Silent Calls

Predictive dialers dial ahead of agent availability to keep agents busy. They estimate when an agent will finish a current call and pre-dial the next contact before that happens. When the math works, agents move from call to call with minimal wait time. When it doesn't, you get abandoned calls and ghost calls.

Here's the specific sequence that produces a ghost call in a ViciDial or Asterisk environment:

The dialer calls a number and the contact answers.

AMD runs on the audio stream: live human or voicemail?

Legacy AMD measures silence duration and basic audio energy.

The contact pauses briefly before speaking, or answers in a quiet environment.

AMD classifies the call as a machine. False positive.

No agent is connected. The contact hears dead air.

The contact hangs up.

That sequence repeats hundreds of times per day at scale. Legacy silence-based AMD produces false positive rates of 15 to 20 percent. At a 50-agent center, that's 1 in 5 live contacts being treated as voicemail.

The Difference Between a Ghost Call and an Abandoned Call

These two terms get conflated, but they're different problems.

An abandoned call happens when no live agent is available when a contact answers. The dialer connected the call, but no agent was ready. Ofcom in the UK mandates that abandoned call rates stay below 3 percent of live calls over any 24-hour period. The FTC's TSR has similar requirements in the US.

A ghost call is specifically caused by AMD misclassification. The dialer had an agent available, but AMD told it not to connect one. The contact gets silence, not a waiting agent. The revenue loss is the same. The compliance exposure is different but real in both cases.

Why Ghost Calls Are a Compliance Problem in 2026

Regulators treat silent calls as nuisance calling. In the UK, Ofcom can fine operators for persistent silent calls regardless of intent. The FTC's Telemarketing Sales Rule requires that abandoned calls play a specific message within two seconds of the contact saying "hello." A ghost call doesn't do that.

If your AMD is producing ghost calls at a 15 to 20 percent rate, you're not just losing revenue. You're generating a pattern of silent calls that regulators can identify and act on. TCPA exposure compounds this if your dialer is calling mobile numbers.

The practical risk in 2026 goes beyond fines. Carriers are increasingly using call behavior patterns to flag numbers as spam. A high rate of silent calls accelerates caller ID flagging — and once your numbers are flagged, contact rates drop across your entire operation.

Why Legacy AMD Cannot Fix This

The AMD built into ViciDial and Asterisk was designed when voicemail systems had predictable audio signatures. A greeting started with a specific pattern of silence followed by a recorded message. Silence detection worked well enough in that environment.

That environment no longer exists. Modern voicemail systems vary widely. Cellular voicemail behaves differently from landline voicemail. Regional carriers introduce different latency patterns. Live humans answer calls in noisy environments, with accents, with hesitation, or with a brief pause before speaking.

Silence detection can't distinguish between a human who paused and a voicemail that hasn't started yet. It uses a single variable. The real world has dozens.

Tuning silence thresholds in ViciDial helps at the margins. Operators spend hours adjusting AMDMAXWAIT, AMDSILENCETHRESHOLD, and related parameters. The false positive rate comes down a few points — not close to eliminating the problem. You're still hanging up on 10 to 15 percent of live contacts after tuning.

How ML Audio Fingerprinting Stops Ghost Calls

Machine learning AMD analyzes the actual audio fingerprint of a call in real time. Instead of measuring silence, it listens to the acoustic characteristics of what's on the line: pitch, cadence, breath patterns, the specific signature of a live human voice versus a recorded greeting.

This approach identifies live humans from the first fraction of a second of audio. It doesn't wait for silence to end. It doesn't guess based on a threshold. It classifies based on a trained model built on millions of real call audio samples.

The result is 99% accuracy — the difference between a 15 to 20 percent false positive rate and a sub-1 percent false positive rate. At a 50-agent center, that gap is worth approximately $120,000 per month in recovered live connections.

AMDY.IO uses exactly this approach. The engine replaces legacy silence-based AMD with real-time audio fingerprint analysis. Every detected live human is routed to an agent immediately. Ghost calls stop because the misclassification that creates them stops.

How to Stop Ghost Calls in ViciDial and Asterisk

You have three practical options in 2026.

Option 1: Tune Legacy AMD Parameters

Adjust silence thresholds in your existing ViciDial or Asterisk configuration. Free, no new software required. The ceiling on improvement is low. You'll reduce false positives but not eliminate them — and you'll spend hours tuning and re-tuning as call patterns shift.

Option 2: Disable AMD Entirely

Some operators turn AMD off and connect every call to an agent. This eliminates ghost calls from false positives, but agents spend significant time sitting through voicemail greetings. Utilization drops. Voicemail drop functionality is gone. That's a trade-off, not a fix.

Option 3: Replace Legacy AMD with ML-Based Detection

This is the only option that solves the root problem. ML audio fingerprinting classifies calls accurately enough that false positives become rare rather than routine. Agents connect to live humans. Voicemails are handled automatically. Ghost calls drop to near zero.

AMDY.IO installs into ViciDial, Asterisk, GoAutoDial, FreePBX, Issabel, and 3CX with a single line of code. No platform migration. No developer project. If your stack is already running, you can have ML AMD active the same day. Any SIP-based dialer can connect via the WebSocket API without touching your existing infrastructure.

The 14-day free trial at amdy.io has zero setup fees. You can measure the false positive reduction against your current baseline within the first week.

The Revenue Math on Ghost Calls

A 50-agent outbound center running a predictive dialer at standard volume will connect roughly 5,000 to 7,500 calls per day across the team. At a 15 percent false positive rate, 750 to 1,125 of those daily connections are live humans who heard silence and hung up.

Each one is a lost contact opportunity. In debt collection, insurance, solar, or home services, a single live connection can be worth $50 to $500 in pipeline value depending on conversion rates. The math reaches $120,000 in monthly losses quickly — and $1,440,000 annually.

That's the cost of keeping legacy AMD in place. It's not a technology problem. It's a revenue problem.

Frequently Asked Questions

What exactly is a ghost call in a call center context? A ghost call is a connected outbound call where the contact answers but hears only silence. It happens when AMD misclassifies a live human as a voicemail and fails to connect an agent. The contact gets dead air and hangs up.

Why does my ViciDial setup produce so many silent calls? ViciDial's built-in AMD uses silence detection to distinguish live humans from voicemail. Silence is an unreliable signal. Live humans pause before speaking. Quiet environments produce low audio energy. The result is a 15 to 20 percent false positive rate — meaning 1 in 5 to 7 live contacts is misclassified as a machine.

Are ghost calls a compliance risk under TCPA or Ofcom rules? Yes. Ofcom treats persistent silent calls as a regulatory violation regardless of intent. The FTC's TSR requires that abandoned calls play a disclosure message within two seconds. Ghost calls generated by AMD false positives create a pattern of silent calls that regulators and carriers can identify and act on.

Can I fix ghost calls by tuning AMD settings in ViciDial? Tuning parameters like AMDMAXWAIT and AMDSILENCETHRESHOLD can reduce false positives at the margins. It won't eliminate them because the underlying method — silence detection — is fundamentally limited. The false positive rate after tuning typically stays above 10 percent.

How does ML-based AMD eliminate ghost calls? ML AMD analyzes the audio fingerprint of a call in real time, identifying the acoustic characteristics of a live human voice versus a recorded greeting. It classifies calls from the first fraction of a second of audio without relying on silence thresholds. AMDY.IO achieves 99% accuracy with this approach, reducing false positives to near zero.

How hard is it to integrate a new AMD engine into an existing ViciDial or Asterisk setup? AMDY.IO installs with a single line of code into ViciDial, Asterisk, GoAutoDial, FreePBX, Issabel, and 3CX. No platform migration required. Any SIP-based dialer can connect via a WebSocket API. Most centers are running the new engine the same day they decide to install it.

How quickly can I see ROI after switching to ML AMD? At a 50-agent center losing $120,000 per month to false positives, the recovery starts on day one of accurate detection. Most centers see measurable improvement in live connection rates within the first week of the trial period.

Stop Accepting Ghost Calls as a Cost of Doing Business

Ghost calls are not a predictive dialer side effect you have to live with. They're a direct output of inaccurate AMD — and inaccurate AMD has a known fix.

Legacy AMD guesses. AMDY.IO knows.

If your center is running ViciDial, Asterisk, or any SIP-based dialer, you can replace silence detection with ML audio fingerprinting today. One line of code. No migration. No setup fees.

Start your 14-day free trial at amdy.io or use the ROI calculator to see exactly what ghost calls are costing your team each month.