The 3% rule

Every person your AMD calls a machine is an abandoned call.

The Telemarketing Sales Rule caps abandoned calls at 3% of the calls a live person answers. Most dialers only count over-dialing against it. The bigger number is usually the detector: every person it mislabels as a machine is a hang-up on a real person, and it counts.

of calls answered by a person

The most a campaign may abandon, measured over each 30-day period of the campaign (or the whole campaign when shorter). Calls answered by a machine do not count in the base, which makes the rate harder to meet, not easier.

2 s

to connect a person to an agent

A call is abandoned when the person who answered is not connected to a sales representative within two seconds of their completed greeting. A detector that is still listening at three seconds has already used the budget.

$53,088

per violation

The FTC's maximum civil penalty for a Telemarketing Sales Rule violation, for penalties assessed after January 17, 2025, and adjusted for inflation every January. Each abandoned call can be a violation.

What counts as abandoned

Three ways a call to a live person ends without an agent. Only the last one shows up in most abandon reports.

A person your detector calls a machine. The call is dropped or gets a voicemail drop played at a live person. That is an abandoned call, and it counts against the 3% whether or not your dialer's pacing was perfect.
A person your detector takes too long on. If the verdict lands after the two-second window and no agent is on the line yet, the call is abandoned even when the verdict is right.
A person your dialer over-dialed for. The classic abandon: more answers than free agents. This is the only kind most dialers measure, which is why abandon reports look better than the truth.

The math

Why a detector that is "95% accurate" cannot keep you under 3%.

Say a campaign reaches 1,000 live people a day. A detector that wrongly calls 5% of them a machine has abandoned 50 calls before the dialer has over-dialed once. That is a 5% abandon rate, over the line, with the pacing set perfectly.

At 3% the detector alone spends the whole allowance, and any over-dial on top of it puts the campaign out of the safe harbor. To leave room for normal pacing, the detector's error on people has to sit well under 1%, and you have to know the number, not guess it.

This is why AMDSure reports people sent to voicemail as its own figure, with the sample size, and why people are decided first: a short hello is the clearest signal there is, and a person decided in 2.5 seconds is a person still inside the two-second window once the agent bridge is counted.

What AMDSure gives you

Your real number, before you switch. Shadow mode runs beside your built-in AMD() on every call. The dashboard shows people wrongly sent to voicemail for both, on your own traffic, with the count behind it. That figure is your detector's share of the abandon rate.
Two error rates, published apart. People sent to voicemail and voicemails caught, each with its sample size. One "accuracy" figure cannot tell you which side of the 3% line you are on.
People decided first. A live hello is decided in about two and a half seconds on average, inside the window, instead of waiting out a greeting timer.
A record for every call. Each verdict, its reason, its timing and what answered are kept and exportable, which is the record-keeping the safe harbor asks for on the detection side.
Fails open. Any fault on our side hands the call to your own AMD() in seconds. A detector that fails to "human" floods agents with machines; one that fails to "machine" abandons people. Ours does neither.

This page explains the rule as we understand it and is not legal advice. The sources are below; confirm the current penalty figure, which the FTC adjusts each January.

Find out your detector's share of the 3% today.

A 20 minute call. We show the console with live numbers, then set up shadow mode on your dialer so you can judge for yourself.

Start free