Live platform data

What our answering machine detection data actually shows

Nobody can honestly quote an accuracy percentage for answering machine detection without a labelled ground-truth set, and we do not have one — so we do not claim a number. What we can show is what our detector saw: 2.99 billion detections across a trailing 30-day window, 13.0% of them live humans, 15.8% false answer supervision from carriers. Every figure below is queried live.

Window: Thu Aug 06 to Thu Sep 03 · recomputed hourly · aggregated platform-wide, no customer identified.

2,985,462,540
Detections in window
amd_carrier_daily, 30 days
13.0%
Live humans
387,129,027 calls
15.8%
False answer supervision
471,716,264 calls
0.092%
Spam-trap hits
2,733,796 numbers
122
Distinct outcome classes
across 3,605,089,985 classified events
4.0s
Average audio analyzed per verdict
all calls, not humans only — the aggregate carries no per-class split

How often does a real person actually pick up?

Across 2,985,462,540 detections in the window, 13.0% resolved to a live human and the rest to a machine — voicemail, an IVR, or a carrier recording. That ratio is the whole economic case for detection: about 8.7 in every 10 answered calls are answered by something that is not a person.

Human 13.0%Machine 87.0%
Human 387,129,027 · machine 2,598,333,459 · total 2,985,462,540. Separately, 15.8% of all calls (471,716,264) were false answer supervision, where the carrier signals an answer no person or machine ever gave.

Is false answer supervision getting worse?

In this window it fell. FAS moved from 15.8% of calls on the first day of the window to 11.3% on the last, while the human rate stayed inside a 11.8–14.0% band on full-volume days. The saw-tooth is the weekly cycle, not a data gap: a typical busy day runs 139,986,980 detections against 22,011,321 or fewer on the quietest days.

0%10%20%30%08-0608-1008-1408-1808-2208-2608-3009-03
Amber line: human rate. Blue line: FAS rate. Faint bars: daily detection volume (peak 154,818,458). Values: 2026-08-06 human 13.4% / FAS 15.8% · 2026-08-07 human 13.4% / FAS 16.5% · 2026-08-08 human 13.9% / FAS 15.3% · 2026-08-09 human 6.0% / FAS 10.4% · 2026-08-10 human 14.0% / FAS 17.2% · 2026-08-11 human 13.4% / FAS 18.1% · 2026-08-12 human 13.2% / FAS 17.9% · 2026-08-13 human 12.9% / FAS 18.6% · 2026-08-14 human 12.8% / FAS 18.7% · 2026-08-15 human 13.3% / FAS 15.7% · 2026-08-16 human 8.3% / FAS 6.7% · 2026-08-17 human 13.4% / FAS 19.4% · 2026-08-18 human 11.8% / FAS 19.3% · 2026-08-19 human 12.1% / FAS 18.7% · 2026-08-20 human 12.0% / FAS 18.0% · 2026-08-21 human 12.3% / FAS 17.7% · 2026-08-22 human 12.6% / FAS 17.6% · 2026-08-23 human 7.1% / FAS 15.5% · 2026-08-24 human 13.4% / FAS 17.1% · 2026-08-25 human 13.2% / FAS 15.8% · 2026-08-26 human 12.9% / FAS 13.6% · 2026-08-27 human 12.5% / FAS 12.0% · 2026-08-28 human 13.0% / FAS 12.1% · 2026-08-29 human 12.3% / FAS 15.7% · 2026-08-30 human 7.5% / FAS 6.0% · 2026-08-31 human 14.0% / FAS 12.1% · 2026-09-01 human 13.1% / FAS 11.4% · 2026-09-02 human 13.0% / FAS 10.6% · 2026-09-03 human 12.9% / FAS 11.3%

Which line types reach humans, and which burn calls?

Landlines still answer. RBOC and independent-telco numbers reach a live person at more than double the rate of PCS wireless, and carry a fraction of the false answer supervision. Only line types with over 100 million calls in the window are shown; the long tail is too thin to read.

Wireless16.3% FAS11.6% human1,348,193,679 callsPCS wireless20.4% FAS10.2% human652,386,332 callsCLEC14.7% FAS11.3% human351,549,097 callsRBOC landline10.4% FAS21.3% human348,942,035 callsVoIP (IPES)15.5% FAS11.7% human137,443,006 callsIndependent telco5.4% FAS24.7% human121,577,201 calls
LERG prefix type from the dialed NPA-NXX. WIRELESS: 1,348,193,679 calls, 16.3% FAS, 11.6% human · PCS: 652,386,332 calls, 20.4% FAS, 10.2% human · CLEC: 351,549,097 calls, 14.7% FAS, 11.3% human · RBOC: 348,942,035 calls, 10.4% FAS, 21.3% human · IPES: 137,443,006 calls, 15.5% FAS, 11.7% human · ICO: 121,577,201 calls, 5.4% FAS, 24.7% human

Does the hour you dial change who answers?

Barely. Between 09:00 and 20:00 US Eastern — the hours that hold 99.1% of all volume — the human rate sits in a narrow band. The excluded overnight buckets show higher percentages, but the largest of them holds 10,974,317 calls against 406,796,472 in the busiest hour, so those are small-sample artifacts rather than an argument for dialing at 4am.

913.01012.61112.81212.91313.21413.21513.01613.01712.91813.21914.22014.120%0Bars: call volume · amber line and lower row: human %
9:00 ET 199,405,675 calls / 13.0% human · 10:00 ET 341,921,620 calls / 12.6% human · 11:00 ET 376,149,832 calls / 12.8% human · 12:00 ET 406,796,472 calls / 12.9% human · 13:00 ET 353,495,157 calls / 13.2% human · 14:00 ET 317,627,543 calls / 13.2% human · 15:00 ET 342,903,213 calls / 13.0% human · 16:00 ET 378,562,257 calls / 13.0% human · 17:00 ET 343,461,033 calls / 12.9% human · 18:00 ET 297,017,860 calls / 13.2% human · 19:00 ET 116,720,671 calls / 14.2% human · 20:00 ET 46,575,530 calls / 14.1% human

Is false answer supervision spread out or concentrated?

Concentrated. Of 2,876 carriers seen in the window, the largest three account for 20.9% of every false answer, and the largest ten for 41.2%. We do not publish the names. The data shows where FAS is observed, not who is responsible for it, and naming third parties would be a claim we cannot settle with data we are able to open.

Top 3 carriers20.9%Top 5 carriers29.3%Top 10 carriers41.2%
Share of 471,716,264 false-answer events, across 2,876 carriers. Top 3 20.9% · top 5 29.3% · top 10 41.2%.

What do the machines actually say?

Detection is not one signal. These are the greeting types the classifier separates most often in the window. Two buckets are honest admissions rather than findings: “unclassified greeting” is audio the classifier could not place, and “no audio captured” is a call that ended before anything usable arrived.

Unclassified greeting14.6%Call forwarded, number read back11.5%Forwarded to voicemail10.9%No audio captured9.3%Forwarded, voicemail tone6.5%“Please leave your message for…”6.0%Google Voice call screening5.9%Number read back4.8%Custom personal voicemail3.8%Voicemail not set up3.5%
Unclassified greeting 14.6% (433,997,666) · Call forwarded, number read back 11.5% (342,199,121) · Forwarded to voicemail 10.9% (324,414,128) · No audio captured 9.3% (277,453,041) · Forwarded, voicemail tone 6.5% (193,620,568) · “Please leave your message for…” 6.0% (179,663,500) · Google Voice call screening 5.9% (176,208,417) · Number read back 4.8% (143,455,549) · Custom personal voicemail 3.8% (112,223,137) · Voicemail not set up 3.5% (104,952,149)

The key figures in one table

FigureValueDerived from
Detections analyzed (30-day window)2,985,462,540SUM(count), amd_carrier_daily
Live human answers387,129,027 (13.0%)SUM(human_count) / SUM(count)
Machine answers2,598,333,459 (87.0%)SUM(amd_count) / SUM(count)
False answer supervision471,716,264 (15.8%)SUM(fas_count) / SUM(count)
Spam-trap (honeypot) hits2,733,796 (0.092%)SUM(honeypot_count) / SUM(count)
Distinct outcome classes122COUNT(DISTINCT classification), amd_hourly_classification
Classified events3,605,089,985SUM(count), amd_hourly_classification
Average audio per verdict4.01svolume-weighted avg_duration, amd_hourly_stats
Carriers observed2,876COUNT of distinct company, amd_carrier_daily
FAS from the top 3 carriers20.9%top-3 fas_count / total fas_count
FAS from the top 10 carriers41.2%top-10 fas_count / total fas_count
Window coveredThu Aug 06 to Thu Sep 03MIN(day) / MAX(day), excluding the partial current day

What this data cannot tell you

These are our detector’s own verdicts, not externally labelled ground truth. That rules out a whole class of claim, and we would rather say so than fill the gap:

  • Accuracy, precision, recall, or a false-positive rate. There is no labelled ground truth here, so any percentage would be invented.
  • Time to identify a human specifically. The hourly aggregate carries one average duration for all calls, with no classification split.
  • Latency percentiles. The aggregates store averages only.
  • Connect-rate lift versus dialing without AMD. There is no control group and no pre-AMD baseline.
  • Revenue, ROI, or dollars recovered. Nothing financial exists in this database.
  • Customer counts or “trusted by N companies”. These aggregates hold IPs, not identities, and IPs are not customers.

Methodology

All figures on this page come from AMDY’s own detection logs, queried live from our analytics database when the page is rendered. Unless stated otherwise, each number covers a trailing 30-day window ending the day before the page was last built, and the window is printed alongside the charts. Because it is a rolling window, the numbers move: a figure published in an earlier post or a previous build reflects a different 30 days and will not match. That is expected, not an error — our State of AMD 2026 post reports 12.5% human on its own earlier window, against 13.0% here.

Everything is aggregated across the entire platform. No individual customer, account, IP address, phone number, or campaign is identified, and no figure is broken out in a way that would isolate one account. Carrier-level analysis is published as concentration and by regulatory line type, never by carrier name — the finding holds either way.

Two limits worth stating plainly. First, these are our detector’s own verdicts, not externally labelled ground truth, so nothing here measures detection accuracy and we make no accuracy claim from this data. Second, the carrier and line-type breakdowns cover only calls whose number maps to a known North American NPA-NXX in LERG (2,985,462,540 calls), which is a smaller set than total classified events (3,605,089,985); the two denominators are never mixed inside a single chart. Every figure drawn from the daily table excludes the current day, which is always partial; the hour-of-day, greeting and classification figures come from hourly aggregates over a rolling 30×24-hour clock instead.

Run it against your own traffic

Sandbox is free at 50,000 detections a month, no card. Your dashboard shows the same breakdowns for your own calls.

Start free