SIP trunks and false answer supervision: what every call center needs to know

What false answer supervision is, technically

Every SIP call follows a well-defined signal sequence. Your dialer sends an INVITE. The carrier responds with 180 Ringing while the destination phone rings. When someone picks up, the carrier sends 200 OK — the signal that the call is answered and the billing clock starts.

False answer supervision happens when the 200 OK fires before a genuine human or voicemail pickup event occurs. The carrier has “answered” the call in the signaling layer while the destination phone is still ringing, while a carrier announcement is playing, or while the connection is dead. Your dialer has no way to know. From its perspective, the call connected.

The result: you’re charged for connect time that produced no conversation, your AMD receives audio it wasn’t designed to classify (often silence, a ringback tone, or a carrier intercept announcement), and an agent may be briefly tied to the call before it becomes obvious nothing is there.

Why it happens

FAS is not always intentional fraud. Several legitimate carrier behaviors produce the same observable effect:

• Early media delivery — carriers send ringback tones or in-network announcements via early media, then fire a 200 OK to simplify billing handoff before the true answer event propagates back through the network.
• Network topology delays — on international routes with multiple transit carriers, the 200 OK from a downstream carrier can arrive at your SIP trunk before the actual pickup signal from the far end catches up.
• Carrier intercept injection — instead of routing to a disconnected number, the carrier “answers” with a recording (“the number you have dialed is not in service”) and bills for the connect time.
• Silent FAS — the 200 OK fires and the call produces no audio at all. The carrier logs it as connected; your side hears nothing.
• Profit-driven FAS — in some wholesale markets, especially international, carriers deliberately trigger 200 OK to increase billable duration. This is where FAS shades from technical artifact into fraud.

The five FAS patterns AMDY classifies

Not all FAS sounds the same. AMDY’s acoustic model has been trained to recognize five distinct carrier false-answer patterns that each require a different downstream response:

Heuristic AMD cannot reliably detect any of these. It is built to classify silence and beep patterns as human or voicemail — it has no concept of a carrier audio signature. When FAS produces silence, heuristic AMD often returns NOTSURE or MACHINE, and the call is logged as a detection event rather than a billing error. Full breakdown of FAS and what it costs →

The cost math

Across the AMDY network, roughly 14% of “answered” calls are carrier false-answers. On a high-volume operation dialing millions of calls per month, that number translates directly to money leaving the operation through at least three channels:

• Carrier billing — most SIP trunks bill per-minute from the 200 OK. A $0.007/min connect charge on a call that produces 8 seconds of carrier audio before disconnect is still a billable event. At 14% FAS across 5 million answered calls a month, that’s 700,000 phantom connect charges.
• Agent idle time — if a FAS call passes AMD undetected, an agent may be connected for the seconds it takes to realize no one is there. At 30 agents each absorbing a few phantom connections per shift, idle time compounds quickly.
• Inflated abandonment denominators — FAS calls that count as “answered” in your dialer inflate the denominator of your abandonment rate calculation, making compliance metrics look better than they are (a separate problem, detailed in our abandonment rate guide).

Which SIP trunk providers have higher FAS rates

Provider-level FAS rates vary widely. Rather than calling specific carriers out by name, the pattern in AMDY’s network data is clear enough to give useful general guidance:

• International routes have materially higher FAS rates than domestic US calls. The more transit carriers in the path, the greater the chance of signaling timing drift and the greater the incentive for an intermediate carrier to inject early answers.
• Budget wholesale providers — carriers competing primarily on per-minute price — tend to show higher FAS because they transit through carriers that engage in profit-driven false answer. The margin pressure is passed through the wholesale chain.
• Carriers without STIR/SHAKEN implementation show higher FAS rates in AMDY’s data. STIR/SHAKEN adoption correlates with overall network quality investment; carriers that have not implemented attestation tend to sit on wholesale routes with weaker signaling discipline.
• Carriers that do not support SIP 183 early media properly sometimes substitute a 200 OK to deliver pre-answer audio, producing FAS as an unintended side-effect of a missing feature.

If you are dialing primarily US domestic numbers via a Tier 1 carrier with STIR/SHAKEN, you will likely see FAS below 5%. If you are running international campaigns or using budget wholesale trunks, 14–20% is not unusual. The only way to know your actual rate is to measure it — your dialer’s call logs will not tell you, because your dialer trusts the 200 OK.

How AMD deals with FAS

The fundamental limitation of heuristic AMD in the face of FAS is architectural. Heuristic AMD classifies silence and syllable patterns. It was designed to answer: “Was the audio after answer a human voice or a voicemail beep?” FAS often produces audio that is neither — it is a carrier tone, an intercept announcement, or nothing at all.

AI-based acoustic AMD approaches the problem differently. Instead of applying rules to what it hears, it classifies the acoustic signature of the audio against a model trained on the full range of what actually comes back on a modern SIP answer. Carrier ringback tones, intercept announcements, busy tones, and silence each have distinguishable acoustic properties that a trained model can separate from genuine human pickup or voicemail greeting.

AMDY classifies FAS into the five patterns above and returns a FAS result rather than HUMAN or MACHINE. Your dialer can then treat FAS calls as billing anomalies, exclude them from abandonment calculations, and feed the data back to your carrier for credit disputes. That feedback loop — carrier-level FAS reporting grounded in acoustic evidence — is the only way to pressure providers into fixing the problem.

What to do with FAS data

• Exclude FAS calls from your abandonment denominator. They were not answered by a person or a machine; counting them inflates your denominator and creates false compliance comfort.
• Build a FAS credit report by carrier. Most carrier contracts include provisions for billing disputes on failed calls. A monthly report of classified FAS events by route gives you the documentation to request credits.
• Use FAS rate as a trunk quality metric. If a trunk’s FAS rate climbs above your threshold, it is a signal to route traffic elsewhere or escalate with the carrier before the rate compounds further.
• Don’t log FAS as AMD detection events. If your reporting rolls FAS into your “machine” count, your accuracy numbers and cost-per-contact metrics are wrong at the foundation.