Answering machine detection (AMD) tells you whether a person or a machine answered each call in your broadcast. It listens to the first moments of an answered call and labels it human, machine or unknown. That turns "answered" into "heard by a person", so you can report honest reach and retry voicemail contacts later. On Edesy, AMD is an opt-in setting on each campaign.
Turn on AMD for your next campaign: Open the Voice app and enable it in the campaign wizard.
Why "answered" isn't enough
In a broadcast, a call counts as answered when the network says it was picked up. But in India a lot of "answered" calls are not a person:
- the network's own voicemail on a phone that's switched off or busy,
- a recorded announcement,
- an office phone system's greeting,
- a caller-tune or message that plays before anyone speaks.
If 70% of calls show as answered and 20 points of that is voicemail, your real reach is much lower than the dashboard suggests. Worse, you won't retry those contacts because the campaign thinks they were reached.
What AMD gives you
| Verdict | Meaning | What to do with it |
|---|---|---|
human |
A person answered | Count as reached |
machine |
Voicemail or a recorded greeting answered | Retry later, at a different time of day |
unknown |
AMD ran but couldn't decide (silence, very short greeting, noise) | Treat as its own group; check a sample of recordings |
| (empty) | AMD was off, or the call was never answered | n/a |
The verdict is stored per contact, so your results show who was reached by a person and who landed in voicemail.
How it works
- The call is answered.
- AMD listens to the opening audio. A person usually says a short "hello?" and waits. A machine plays a longer, continuous greeting.
- AMD returns a verdict: human, machine or unknown.
- Your message plays and the result is saved with the contact.
AMD uses speech analysis on the opening of each answered call, which is why it's opt-in: it adds work to every answered call in the campaign.
When to turn it on
| Campaign | AMD? | Why |
|---|---|---|
| Payment or EMI reminders | Yes | You must know who heard it before planning the next attempt |
| Appointment reminders | Yes | Voicemail contacts need a second try |
| Emergency or service alerts | Yes | Reach must be measured, not assumed |
| One-off announcement to a small list | Optional | Plain answered rate may be enough |
| Agent dialer (bridging to a live agent) | Consider it | Avoid putting an agent on a voicemail |
Using AMD results to plan retries
The simplest use is a two-pass campaign:
- Pass 1: run the broadcast with AMD on.
- Export or filter contacts with
machineandunknownverdicts, plus no-answer and busy. - Pass 2: use "run again" (duplicate) on that subset at a different time of day. Many people who can't pick up mid-morning will pick up in the early evening.
- Stop after a sensible number of attempts and respect your calling hours and DNC list.
For list cleaning and retry rules in general, see why campaign calls fail: retries, DND and list hygiene.
Cost
AMD doesn't make a machine-answered call free. The network reports it as answered, so it's billed like any answered broadcast call.
| Item | Price (plus 18% GST) |
|---|---|
| Broadcast call | Rs 0.50 per minute, 60-second pulse |
| Unanswered call | Free |
| Number rental (mobile or 92-series) | Rs 499 per month |
The value of AMD is knowing which of those paid calls actually reached a person. That lets you stop paying for repeat calls to people who already heard the message, and focus retries on the ones who didn't. See voice broadcasting pricing, and how to estimate a broadcast's cost before launch.
Honest limits
- AMD is a classifier, not a guarantee. Some people answer with a long greeting; some voicemails are short. Expect some calls in the wrong bucket.
- Unknown is normal. Silence, background noise and very short greetings won't always classify.
- It doesn't change the message. On Edesy, AMD labels the call; your message plays the same way. Plan your follow-up from the labels.
- Check samples. Turn on recording for a test campaign and listen to a handful of
machineandunknowncalls before trusting the split for a big list.
Setting it up
- In the Voice app, start the 4-step campaign wizard.
- Add your message: upload WAV or MP3, or use text-to-speech.
- Upload a CSV, or send leads through the API (
/v1/campaigns/:id/leads). - Choose a caller ID and a backup caller ID, set max concurrent calls and dialing hours.
- Turn on answering machine detection (and recording, if you want samples).
- Use test call to hear it on your own phone, check the cost estimate, and launch.
- Read the results: answered, human, machine, unknown, no-answer, busy.
Single calls through POST /v1/calls do not run AMD today (the option is accepted but not applied), so use a campaign when you need it. Campaign
endpoints are in the campaigns docs.
More on the product: voice broadcasting and broadcast features.