What dead air and ghost calls are
From the customer's side, dead air is simple: they answered, said "Hello?", and heard nothing. Maybe an agent arrives three seconds later and opens with an apology. Maybe nobody arrives and the line goes dead. That second case is a ghost call, also called a phantom call.
From the agent's side, "dead air" often means something else: the call connected, and there is nobody there, or the person on the other end has already given up. Both views describe the same failure. A live person and an agent did not meet in time.
On a VICIdial predictive dialer, these calls rarely have a single cause. They come from a few places, and fixing the wrong one wastes weeks. The good news is that each cause leaves its own trace.
The six common causes
1. AMD listening too long
While answering machine detection decides, the person hears nothing. VICIdial uses Asterisk's AMD() by default, and its shipped defaults allow a total analysis time of 5000 ms, with 2500 ms of initial silence before it gives up on a quiet answer. Many people say "Hello?" once, wait, say it again and hang up inside that window.
2. NOTSURE
When AMD() cannot decide within its total analysis time, it returns NOTSURE with cause TOOLONG. Those calls used the full analysis time before anything happened. Whatever your VD_amd.agi does with them next, the person has already waited the longest possible time.
3. False machine verdicts
Timing-based AMD judges by silence and word counts. A person who answers with a short "Hello?" and waits can look like a machine; a person who answers with a longer sentence can trip the word limits. When AMD says MACHINE, VICIdial hangs up or plays your voicemail message. To the person, that is a ghost call, or a recording that makes no sense.
4. No agent free
A predictive dialer places more calls than it has free agents, betting that many will not be answered. When too many are answered at once, some wait in the queue in silence or are dropped. This is the dial level at work, not AMD, although slow AMD makes every wait longer.
5. Dial level set too high
A close relative of cause 4: a manual dial level that suited a full shift can be far too high when half the agents are on break. Adaptive methods help, but a high maximum dial level still lets the dialer overshoot.
6. Trunk and network problems
A carrier that signals answer late, or a path with one-way audio, adds silence that no dialer setting can fix. The person hears nothing because the audio is not getting through at all.
How each cause looks in your reports
Before changing anything, find out which cause you have. This table maps each one to where it shows up.
| Cause | What the person hears | Where to see it |
|---|---|---|
| AMD listening too long | Several seconds of silence, then an agent | Agents report customers saying "hello?" repeatedly; Asterisk AMD results of HANGUP when people hang up mid-analysis |
| NOTSURE | The longest silence of all | Count of NOTSURE / TOOLONG results in the Asterisk log |
| False machine verdicts | Silence then hang-up, or a voicemail message | Not visible: logged as answering machine (AA). Only listening to AA calls reveals them |
| No agent free | Silence while queued, maybe a drop | DROP status; dropped percentage in the Real-Time Main Report |
| Dial level too high | Same as above, worse at shift edges | Drop rate rising when logged-in agent count falls |
| Trunk problems | Silence even with an agent on the line | Agent complaints of "dead" calls on calls AMD sent as human; specific carriers or routes |
The third row is the one that catches teams out. VICIdial cannot report a mistake it does not know it made, so a campaign can show a healthy drop rate while AMD hangs up on live people all day. Our post on the abandoned call rate shows how to estimate these hidden calls.
A quick diagnosis routine
Check the drop rate per campaign
Open the Real-Time Main Report. If a campaign's dropped percentage is high, start with the dial level and agent staffing.
Count NOTSURE results
Look at Asterisk's log for a day of AMD results. A high share of
NOTSUREmeans people are waiting the full analysis time.Listen to AA calls
Pull a random sample of calls marked as answering machine on one campaign and listen to them. Every live person you hear is a ghost call your reports did not count.
Ask agents
Agents know when customers greet them with "Hello? Hello?". Ask which campaigns, which times of day and, if they can tell, which number ranges.
Isolate the carrier
If dead air remains on calls that AMD sent straight to waiting agents, compare carriers and routes, and check for slow answer signalling and one-way audio.
Fixes for each cause
Slow AMD and NOTSURE
You can shorten AMD()'s total analysis time and initial silence in amd.conf or the AMD() arguments. That cuts waiting, but timing rules decide faster by guessing sooner, so more calls are misjudged in one direction or the other. Tuning moves the errors around rather than removing them.
False machine verdicts
This needs detection that listens to what the answer sounds like, not just how long the silences are. That is what AI answering machine detection does.
No agent free and dial level
- Use an adaptive dial method with a Drop Percentage Limit set below your compliance target.
- Lower the maximum dial level, especially at the start and end of shifts and around breaks.
- Catch more machines: every voicemail AMD stops is an agent who is free for the next live answer.
Trunks
Raise tickets with specific call examples: time, number dialed and what the agent heard. If one route is worse than others, move traffic while it is investigated.
Measure after every change
Change one thing at a time, on one campaign, and compare a full week before and after. Track the dropped percentage, the share of NOTSURE results, agent-connected calls per hour and, from your listening sample, how many AA calls were really people. If you change the dial level and AMD on the same day, you will not know which one helped.
Why dead air damages caller ID reputation
Every ghost call teaches the person who answered that your number is not worth picking up. Some block it. Some report it. Carriers' call analytics and call-labelling apps take hang-ups and complaints into account, and numbers that make many short silent calls are more likely to be labelled "Spam Likely".
Once that label appears, answer rates fall, so the dialer places more calls to reach the same number of people, which produces more short calls. Rotating to fresh numbers treats the symptom. The numbers stay clean longer when the calls stop sounding like spam, which means fewer silent seconds and fewer people hung up on.
The cost adds up quickly: lost conversations on the calls themselves, angry callbacks that tie up inbound staff, and lower answer rates on every call afterwards. Our dead air calls page covers the wider picture.
How nomoVM helps
nomoVM targets the AMD causes of VICIdial dead air directly:
- No long waits. Most calls are decided about 2 seconds after answer; only unclear calls listen up to 3.2 seconds. There is no
NOTSUREresult that burns the full analysis time. - Fewer people hung up on. In our measured test of 400 real outbound calls checked by a person, 94.9% of live people were sent to agents, and 88% of calls sent to agents were live people.
- More machines caught. Voicemail, call screening assistants, silence, busy, SIT, fax and ringback are recognised, so agents stay free for real answers.
- Visible decisions. The dashboard lists every call with its verdict, cause, timing and the audio heard, so you can check calls instead of guessing.
It plugs into the VICIdial you already run: one script and a copy of your 8369 AMD extension as 8370, then one campaign's Routing Extension changed to 8370. It takes about 10 minutes, needs no new servers and no carrier change, and if nomoVM is unreachable the call falls back to Asterisk's AMD(). Rolling back means setting the Routing Extension back to 8369.
Plans start free at 30,000 calls a month, and paid plans start at $99 for 1,000,000 calls; see pricing. Start free and follow the VICIdial AMD setup guide to test it on your worst dead-air campaign first.
Frequently asked questions
What causes dead air on VICIdial?
The main causes are answering machine detection listening too long, AMD returning NOTSURE after using its full analysis time, AMD wrongly deciding a person is a machine, no agent being free when the call is answered, and carrier problems such as slow answer signalling or one-way audio.
What is a ghost call?
A ghost call is an outbound call where the person answers and nobody is there. The dialer dropped it, AMD hung up on them, or it waited in a queue with no agent. People also call them phantom calls.
How do I find ghost calls in VICIdial reports?
Dropped calls show as DROP and in the Real-Time Main Report's dropped percentage. Calls where AMD hung up on a live person show as answering machine (AA) and are invisible in reports, so you have to listen to a sample of them.
Can dead air get my numbers marked as spam?
It can contribute. Carriers' call analytics and call-labelling apps take hang-ups and complaints into account, and numbers that make many short silent calls are more likely to be labelled Spam Likely.
Will lowering the dial level fix dead air?
It fixes the part caused by no agent being free. It does not fix dead air caused by slow AMD or by AMD hanging up on people, which happens even when agents are waiting.
How does nomoVM reduce dead air?
It decides most calls about 2 seconds after answer and only listens up to 3.2 seconds on unclear calls, so there is no long NOTSURE wait. In our measured test it sent 94.9% of live people to agents, so fewer people are hung up on.