The State of AMD 2026
We looked at 2.3 billion answered outbound calls across the AMDY network over 30 days. Here’s what outbound answering actually looks like in 2026 — and what it should change about how you dial.
Everyone has a hunch about their connect rate. Almost nobody has the data underneath it. So we pulled it — anonymized and aggregated across the whole network, no single account, no phone numbers. Roughly 2.3 billion detections in 30 days. The numbers are blunt.
Only one in eight “answers” is a person
That’s the headline. 87.5% of answered outbound calls are not a live human. Three quarters are machines — voicemail, IVR, carrier recordings. The rest is something worse, which we’ll get to. If your dialer connects an agent on “answer” and leans on stock detection to sort it out, your agents are spending most of their connected time on things that can’t buy anything.
Carriers are faking one in seven answers
The number that surprises people: 14.2% of answered calls were carrier false answer supervision (FAS) — the network returns a “connected” signal when nothing actually picked up. Your dialer thinks it got an answer; your agent gets dead air; your stats get polluted. Across the network, one specific carrier pattern alone accounted for over 200 million of those fake answers in the month. If your AMD only knows “human or machine,” it has no bucket for this — so it guesses, and you pay for the guess.
The best time to reach a human isn’t when everyone’s dialing
Human rate moves with the clock, and not the way the dialing does. The heaviest dialing window — roughly 10am to 4pm ET, where the network sees 200M+ detections an hour — is also one of the worst for human rate, hovering around 12%. Everyone is hammering the same lists at the same time.
- 7–8am ET runs ~15–19% human — the best window with real volume, before the mid-day crush.
- ~8pm ET ticks back up to ~15%.
- After 9pm human rate falls off (sub-10% by 11pm) and keeps dropping overnight.
The takeaway isn’t “dial more.” It’s “dial when humans are actually answering, and lighten up on the crowded mid-day window.”
Humans announce themselves fast
When a real person answers, they’re usually obvious almost immediately. 17.9% of all human answers were identifiable in under half a second, and the large majority within about a second and a half. Only a stubborn tail takes three or four seconds. That matters for one reason: any detection method that needs to wait for a sentence of speech is, by construction, last to the party on the easiest, fastest humans — the ones you most want to connect live.
Honeypots are rare per call, and expensive in aggregate
Spam-trap (honeypot) numbers were just 0.09% of detections — sounds trivial, until you multiply it out: roughly 2 million dials into trap numbers in a single month across the network. Each one is a small deposit into the “this caller ID is an autodialer” ledger that carriers use to slap “Spam Likely” on your numbers. The per-call rate is a rounding error; the reputation damage compounds.
What to do with this
- Treat AMD as mandatory, not a nicety. 87.5% of answers aren’t people. The whole game is sorting the 12.5% out fast.
- Make sure FAS has its own bucket. 14% of your “answers” are carrier fakes. Two-bucket AMD can’t see them.
- Move volume toward the morning. 7–8am ET beats the 10–4 crush on human rate.
- Watch your honeypot exposure. It’s tiny per call and ruinous over a quarter.
See these numbers on your own traffic
AMDY logs every detection and ships 23 reports — human rate by hour, by carrier, by area code, plus honeypot exposure. The Sandbox plan is free: 50,000 detections a month, no card, 5-minute Vicidial install.
Figures are anonymized, aggregated platform-wide detection counts over a trailing 30-day window (~2.3 billion detections); no individual account or phone number is identified. Hour-of-day is in US Eastern Time. Your own mix will vary by list, carrier, and campaign.