In a campaign report, some of the calls marked "connected" were never conversations. The line opened, a voice was heard, the agent started speaking — but that voice was a voicemail greeting.
This breaks reporting in two directions at once: the number of conversations looks higher than it is, and the conversion rate looks lower.
The signals that separate a machine from a human
Voicemail detection is not magic; it rests on a few observable signals:
- Length of the first sound after pickup. A human says "hello" and stops. A greeting talks without interruption.
- Absence of pauses. A human pauses to listen; a recording does not.
- The beep. The tone at the end of a greeting is the most definitive signal.
- Fixed patterns. Operator greetings are formulaic per country and recognisable.
Relying on a single signal produces wrong answers. A quiet human can look like a machine, and a talkative human can look like a greeting. In practice, requiring two signals together works well as a threshold.
Three options once detected
Hang up. The cheapest. Mark the record as "voicemail" and put it in the retry queue. This is the right choice in campaigns where leaving a message adds nothing.
Leave a short message. Wait for the greeting to finish and the beep, then leave ten to fifteen seconds. The message must say who is calling, why, and how to get back to you.
Send a text instead of a message. In most cases this performs best: the agent hangs up and a short WhatsApp or SMS goes to the same number. Voicemail messages have low listen-through rates; a written message stays visible.
A common mistake when leaving a message
Having the agent open to a machine as if talking to a person — "hello, how can I help you?" — is both funny and damaging. When the recipient plays it back, they cannot tell what was wanted.
The voicemail message needs its own script: one-way, short, asking no questions, and always containing one action.
Timing the retry
A number that reached voicemail is not "unreachable"; it means "not available right now". A retry rule has three parameters:
- Wait time. Calling the same number again half an hour later is an irritation. A gap of several hours performs better.
- A different time of day. The second attempt should fall in a different window than the first. Calling someone the next morning after failing in the morning usually produces the same result.
- A cap on attempts. Persisting beyond three attempts does not convert; it generates complaints.
How it shows up in reporting
Once voicemail detection is in place the campaign report looks worse — and that is a good thing. The "connected" count drops because it now counts actual conversations. The report you hand to a manager should keep this split explicit:
- Real conversation
- Voicemail
- No answer
- Busy / line error
Without those four lines separated, every number discussed about campaign performance is debatable.
Tomorrow: another invisible enemy of outbound calling — call rate limits on the operator side.