Voicemail detection: why your SDRs still hear voicemails

Every dialer claims it filters out voicemails. And yet your SDRs still hear several per session.

It’s not a defect of your tool in particular. Voicemail detection rests on a few seconds of listening and a decision that has to be made fast. It gets it wrong in both directions, and the setting that limits one error increases the other.

How a system tells a human from a voicemail

When a line picks up, the system listens to the very first seconds and looks for patterns.

A person who answers says a short word, then goes quiet and waits. A voicemail launches into a long sentence, without a pause, often with the same intonation from start to finish. A fax emits a continuous tone. An out-of-service number triggers a carrier announcement, usually preceded by three beeps.

The decision plays out in one to three seconds depending on the settings. That’s short, and it’s deliberate: beyond that, the person who picked up starts finding the silence odd.

The cases where it gets it wrong

Three families of situations come up all the time.

The voicemail that talks like a human. “Hello, you’ve reached Marc Lefebvre.” A real voice, a real hello, a first name. Nothing distinguishes it from a person during the first second. Those greetings get through, and the SDR ends up talking to a recording.

The person who picks up without saying anything. It happens more often than you’d think: someone takes the call while walking, picks up by reflex, takes two seconds to speak. A strict setting reads that silence as a voicemail and cuts the call. You just lost a conversation with someone who had picked up.

Switchboards and phone menus. “Welcome to Dupont Industries, for sales press 1.” Technically, it’s a recorded announcement. But behind it, there’s a reachable company. Depending on how your tool classifies this case, you filter them all out, or your SDRs spend their session in menus.

There are also the rare but costly cases: the person who picks up and only says “yes?”, the prospect in a noisy environment, the call that goes through forwarding.

The numbers that mislead

Part of the problem has nothing to do with the audio. It comes from the file.

A switchboard number filed in a “mobile” column, a fax number never cleaned out, a line that changed hands since the export: in all these cases, detection does its job correctly on a number that shouldn’t have been called. The rate of voicemails heard then serves as an indicator of the list’s quality, not the tool’s.

Before touching the detection threshold, look at where the file comes from and when it was last enriched. That’s often where the real fix is.

Tuning the threshold

The setting arbitrates between two errors that don’t cost the same.

A strict threshold cuts fast. You hear few voicemail greetings, but you lose the people who take two seconds to speak. A loose threshold lets more through. Your SDRs hear more recordings, and lose a few seconds each time.

The right trade-off depends on your list. On a file of direct mobile numbers, a strict threshold makes sense: people pick up and talk. On a file of company landlines, full of switchboards and welcome greetings, a loose threshold avoids filtering out reachable accounts.

And the trade-off also depends on the team. A session where one voicemail in three gets through ends up wearing people down, even if the math works on paper. It’s something to keep an eye on.

What it costs, concretely

A conversation cut by mistake shows up nowhere. It appears in your reports as a voicemail, along with the others. You’ll never know the person had picked up.

That’s why we advise against tuning this setting blind. The simplest method is still to put a few numbers you control into a test list, pick up with a deliberate delay, and listen to what happens. You’ll see immediately where your tool’s limit sits.

What we do at Ubic

Detection can be switched on and tuned per list, because a file of mobiles and a file of landlines don’t call for the same threshold.

When a call is classified as a voicemail, it doesn’t stop there: you can trigger a text message or leave an automatic voicemail. The contact then goes back into automation with a delay before they can be called again.

The rest of the filtering follows the same logic. Faxes and invalid numbers are set aside before the call ever reaches the SDR, as described in how the parallel dialer works, and the outcome goes up into the CRM with the matching status. You keep track of what was filtered out in the reports, which is exactly what lets you check whether the threshold is set right.

Frequently asked questions

What detection rate is normal?

No serious vendor will give you a single figure, because it depends entirely on your file. A file of direct mobiles and a file of company switchboards don’t produce the same results with the same tool. Compare on your list, not on a product page.

Can you turn detection off completely?

Yes, and it’s sometimes the right choice. On a short list of high-stakes accounts, where every attempt counts, better to hear a few voicemails than to risk cutting off a conversation.

Does the prospect hear anything during detection?

They hear a short silence after their “hello”, the time it takes to decide. That’s the delay that has to stay brief: a blank that runs too long makes people hang up. Test it on your own number before signing with a tool.

Does detection work on company voicemails with hold music?

It’s one of the hardest cases. Music has neither the profile of a human voice nor that of a recorded announcement, and the classification depends a lot on the tool. If your file contains a lot of company numbers, that’s precisely the case to test during your trial.