AMD for debt collection calls: compliance, accuracy, and voicemail strategy

The compliance landscape

Three bodies of law govern how debt collectors use outbound dialers, and all three touch AMD in ways that most operators haven’t fully mapped.

FDCPA Section 805 restricts the timing of communications with consumers (before 8 a.m. or after 9 p.m. is prohibited) and imposes strict rules on third-party disclosure. A voicemail left for the wrong person — or a message that discloses the debt to someone other than the consumer — is a Section 805 violation regardless of whether it was intended.

TCPA caps abandoned calls at 3% of answered calls per campaign, per day. An “abandoned call” under the FTC’s TSR interpretation is one where no agent is connected within two seconds of the called party completing their greeting. AMD errors directly affect this number — a live person dropped because AMD called them a machine is an abandoned call, whether or not your dialer logs it that way. How the 3% rule works in practice →

CFPB Regulation F, effective November 2021, added call frequency caps (7 calls per week per debt per consumer), required meaningful consent tracking, and introduced new rules on voicemail treatment. Regulation F explicitly accommodates “limited content messages” left on voicemail — but only under the right conditions, discussed below.

Why AMD errors are expensive in debt collection

In most outbound verticals, a dropped live call is a missed revenue opportunity. In debt collection, it may also be a TCPA violation with statutory damages of $500–$1,500 per call. At scale, that changes the risk calculus entirely.

The math works like this: if your dialer makes 10,000 answered calls and heuristic AMD misclassifies 8% of live humans as machines (a conservative estimate for stock Asterisk), you’ve just dropped approximately 800 real people who picked up. If any of those calls occurred via an auto-dialer to a mobile number without proper consent, each one is a potential TCPA claim. Class action plaintiffs’ attorneys check dialer logs for exactly this pattern.

Even on consented landlines, the FTC’s 3% abandoned-call cap creates exposure. A call dropped by bad AMD doesn’t vanish — it goes into the denominator of your abandon rate calculation. Enough AMD errors can push you past the cap without your dialer ever flagging it, because the system thinks those calls were machines.

Voicemail strategy under Regulation F

Regulation F introduced a safe harbor for limited content messages (LCMs) — voicemails that include a business name and callback number but omit any reference to the debt or the collection purpose. An LCM left on a consumer’s own voicemail does not trigger FDCPA’s third-party disclosure rules.

The operative word is “consumer’s own voicemail.” The safe harbor requires that you actually reached a voicemail, not a live human. If AMD mistakes a live person for voicemail and plays an LCM to them directly, that is not a limited content message — it is a debt collection communication that may violate FDCPA’s disclosure requirements. The LCM framework only works if your AMD is accurate enough to know when you’re leaving a message versus speaking to a person.

This is the sharpest edge of AMD accuracy in debt collection: the same classification error that means a missed contact in other verticals means a potential FDCPA violation here.

The NOTSURE problem for debt collectors

Stock Asterisk AMD returns one of three values: HUMAN, MACHINE, or NOTSURE. Most dialers are configured to hang up on NOTSURE — it’s the safe default when you can’t tell what answered. Under TCPA, that hang-up on an answered call is an abandoned call.

The calls that come back NOTSURE are not random. They cluster around the hardest-to-reach consumers: people who answer fast, people in noisy environments, calls on compressed mobile audio. These are often the contacts you most need to reach — and every NOTSURE hang-up is both a missed contact and a potential abandoned-call violation. What NOTSURE actually means and where it concentrates →

Debt collectors need AMD that makes a definitive classification on every call, not a gray zone. A system that returns NOTSURE on 5–10% of calls is creating a compliance liability on every one of those calls, not just an operational inconvenience.

Mobile numbers and cell-scrubbing

TCPA’s restrictions on autodialed calls to mobile numbers mean most compliant debt collection operations run cell-scrubbing before dialing — removing or sequestering mobile numbers that lack express written consent. AMD doesn’t change that requirement.

What AMD affects on consented mobile numbers (and on landlines generally) is the abandoned-call denominator. Carrier false-answer supervision (FAS) is especially dangerous in debt collection contexts: the carrier signals “connected” when nobody actually answered, the dialer logs a connected call, and if no agent is waiting, the timer starts on an abandoned call that never existed. Your abandon rate climbs on phantom calls. How FAS distorts your abandonment stats →

Accurate AMD that identifies FAS calls keeps them out of both the numerator (abandoned calls) and the denominator (answered calls) of your compliance calculation. The math only works when you’re counting real connections.

How AI AMD reduces legal exposure

The compliance benefit of accurate AMD in debt collection comes down to three things:

• LCMs go to actual voicemail — when the classifier is right 99% of the time instead of 70–85%, your Regulation F limited content messages reach voicemail boxes, not live ears. The FDCPA disclosure risk shrinks proportionally.
• Abandoned call counts are accurate — live humans aren’t dropped and mislabeled as machine dispositions, so your abandon rate reflects your real operation, not a fiction created by bad AMD.
• FAS calls don’t inflate your violation denominator — phantom “connections” from carrier false answers are identified and excluded from answered-call counts, keeping your 3% calculation clean. The 3% rule and how AMD accuracy protects it →

No AMD system eliminates compliance risk — that requires proper consent management, call frequency tracking, and legal counsel. But the firms running the tightest compliance programs are also running the most accurate AMD, because AMD accuracy is the foundation that the rest of the compliance math sits on. A dialer that doesn’t know what answered the phone cannot reliably comply with rules that depend on knowing exactly that.