Predictive dialer abandonment rate: what it is and how AMD affects it

How abandonment rate is calculated

The FTC’s TSR definition is precise: abandoned calls divided by answered calls, measured over a 30-day rolling period, per campaign. An abandoned call is one where the system connected to a live person but no agent was available within two seconds of the person’s greeting.

Written out:

Both numbers come from the answered pool — calls where a human picked up. Calls that went to voicemail, were classified as MACHINE, or never connected do not enter either side of the fraction. That distinction is the source of most compliance errors.

The denominator trap: why most operators compute this wrong

The most common mistake is using all outbound calls as the denominator instead of answered calls. If you dial 100,000 calls, 40,000 get answered, and 1,200 of those answered calls are abandoned, the correct abandonment rate is 3% (1,200 / 40,000). Using total dials as the denominator gives you 1.2%, which looks compliant on paper and is wrong under the TSR.

This matters because the FTC measures what happened to people, not what happened to dial attempts. Every regulatory enforcement action in this space starts with the agency recomputing the denominator on the operator’s own call logs — and finding that the “compliant” campaign was actually running at 8–12%. Full guide to the 3% rule →

The AMD connection: hidden violations your metric never shows

This is where answering machine detection becomes a compliance issue in a way most operators don’t anticipate.

When your AMD misclassifies a live human as a machine, the dialer hangs up. In ViciDial, that call is logged with disposition MACHINE (or equivalent). It exits the call flow as a machine detection event, not as an abandoned call. It never enters your abandonment numerator, and it never enters your abandonment denominator.

The result: you dropped a real person who answered your call, you did not connect them to an agent, and your abandonment rate metric never reflected it. The person’s experience was identical to an abandoned call. The regulatory exposure is real. The number on your report is clean.

This is not a minor edge case. Heuristic AMD misclassifies live humans at a rate of 5–15% on modern call lists. On a campaign that genuinely answers 40,000 humans per month, that is 2,000–6,000 people who answered your call, got hung up on immediately, and generated zero visibility in your compliance reporting. If a regulator subpoenas your call records and reinterprets fast MACHINE dispositions, the hidden picture can be significantly worse than your reported rate.

The FAS distortion: phantom “answered” calls inflating your denominator

False answer supervision (FAS) works in the opposite direction — it makes your abandonment rate look better than it is.

FAS occurs when your SIP carrier fires a 200 OK (the “call answered” signal) before a genuine human or voicemail pickup. In some dialer configurations, that 200 OK increments the “answered” counter. The call connects to nothing, produces dead air or a carrier announcement, and is then classified by AMD as MACHINE or FAS — but if your dialer already counted it as “answered” in the denominator, you now have a phantom entry that reduces your apparent abandonment rate.

Across the AMDY network, roughly 14% of “answered” calls are FAS events. If those calls inflate your denominator without a corresponding increase in your abandonment numerator, your reported rate can sit at 2.8% while the true rate on genuine human answers is above 3%. How FAS works and what it costs →

How accurate AMD improves the compliance math

The fix operates on both distortions simultaneously:

• Fewer false-positive drops. When AMD accuracy goes from 80% to 99%, the number of live humans misclassified as MACHINE shrinks by roughly 95%. Hidden violations that were accumulating in machine dispositions move into properly counted abandonment events — or, more often, get connected to an agent correctly.
• Correct FAS classification. When FAS calls are identified as FAS rather than MACHINE or counted as answered humans, they can be excluded from both sides of the abandonment fraction, or at minimum excluded from the denominator. Either way, the phantom entries that inflated your denominator disappear.
• Clean audit trail. AI AMD returns a per-call classification with confidence and disposition. If the FTC asks to see your call records, you can show exactly how each call was classified and why it was or was not counted as abandoned.

Best practices for compliant abandonment rate calculation

Use answered-call denominator, always

Configure your reporting to divide by calls where a human answered, not total dials and not total connects. If your dialer uses a broad “connected” count that includes FAS and machine answers, build a separate query that filters to human-answered calls only.

Exclude FAS from both numerator and denominator

A FAS call was not answered by a person. It should appear in neither the abandoned count nor the answered-call denominator. This requires your AMD to return a FAS classification that your dialer can act on — a binary HUMAN/MACHINE result is not sufficient.

Audit MACHINE dispositions monthly

Pull a sample of fast MACHINE dispositions (calls classified as machine in under 1.5 seconds) and manually review the audio. If a material percentage of those calls shows a human voice, your AMD is generating hidden violations. The sample size does not need to be large — 50–100 calls per month is enough to detect a problem.

Measure by campaign, not by account

The TSR applies the 3% cap per campaign. An aggregate abandonment rate across all campaigns can mask a single campaign running well above compliance while others run below. Report and track each campaign independently.

Keep 30 days of raw call records

The FTC’s rolling 30-day window means you need granular call records for the full period, not just summary metrics. Store the per-call AMD disposition, the call duration, the connect timestamp, and the agent assignment timestamp. If you are ever asked to demonstrate compliance, these are the records you’ll need.