ASR: What It Tells You and What It Hides

 Key takeaways

• ASR alone is gameable. A rising ASR means nothing without checking ACD and PDD alongside it.
• A rising ASR combined with a falling ACD is the clearest fraud signal in wholesale voice.
• Sub-second PDD on a distant route means the call never left the vendor's server.
• PDD over 7 seconds on a mobile destination is a strong indicator of SIM farm termination.
• The three metrics form a triangle. Manipulating one always distorts the others.

---

ASR — Answer-Seizure Ratio — is one of the most commonly reported metrics in VoIP wholesale. It is also one of the most commonly misread. Understanding what it actually measures, and more importantly what it does not, is essential for anyone who routes traffic professionally.

ASR is simply the percentage of call attempts that result in an answer. If we send 100 calls to a vendor and 60 of them are answered, the ASR is 60%. That is the whole formula. Nothing sophisticated about it. The trouble begins when operators treat a good ASR as proof that a route is performing well — because a route can post an excellent ASR while robbing you blind.

Let's start with the honest picture. We have a vendor — call them CleanVoip — terminating traffic to a busy destination. We send 100 calls. Perhaps 55 connect, 30 get a proper busy signal (SIP 486) or a ring-no-answer (SIP 408/487), and 15 fail outright (5xx errors). ASR is 55%. Not spectacular, but that is the destination. The calls that answer ring properly, conversation happens, the callee hangs up naturally. Average call duration — ACD — sits somewhere around 3–4 minutes. PDD (Post-Dial Delay, the time from sending INVITE to hearing the first ring) is a normal 3–4 seconds. Everything is consistent with a legitimate termination route.

Now let's say we switch that same traffic to a new vendor, ShortVoip. They quote us a better rate and our ASR climbs to 78%. Nice, we think. The route is performing better. Except our ACD has quietly dropped from 3.5 minutes to 22 seconds. And our customers start complaining that calls connect but nobody is there.

This is the FAS vendor's fingerprint. ShortVoip is not terminating the calls to the actual destination. They are answering them locally — sending a 200 OK back to our softswitch the moment the call arrives — and playing our callers a few seconds of audio: ambient noise, a recorded voice, silence, or a tone designed to sound like a connection in progress. The caller waits a few seconds, gets nothing, and hangs up. ShortVoip has manufactured a call with zero termination cost, billed us for 20–30 seconds of answered time, and walked away with clean-looking ASR numbers because the call was, technically, "answered."

The math is worth spelling out. Say we route 1,000 calls per day to ShortVoip at $0.07/minute. At a fraudulent ACD of 22 seconds, we are paying $0.07 × (22/60) × 1,000 = $25.67 per day for calls that connected to nobody. At a real ACD of 3.5 minutes, those same 1,000 answered calls would have cost $408 — but our customers would have actually reached the other end. We are not just losing quality; we are paying for fraud and taking the customer satisfaction hit simultaneously.

Now here is the important part. ASR, ACD, and PDD are not independent signals — they form a triangle, and any manipulation of one shows up as a distortion of the others. A fraudulent vendor who inflates ASR by connecting calls to a local IVR must, by definition, deflate ACD. Real conversations last minutes. Fake ones last seconds. If your ASR is rising while your ACD is falling, that combination is the tell. It is almost never explainable by legitimate traffic patterns.

PDD adds a third data point. On a real route, PDD reflects actual signalling travel time through the destination network. As a practical reference:

• Excellent: under 2 seconds — typically a direct, high-quality interconnection
• Good / Normal: 2–4 seconds — standard for well-maintained routes, most interconnections fall here
• Acceptable: up to 7 seconds — still workable, common on longer routing chains or certain geographies
• Poor: over 7 seconds — likely to generate customer complaints, and a strong indicator of a SIM farm somewhere on the route

That last point is worth dwelling on. A SIM farm is a device — sometimes a rack of smartphones, sometimes a dedicated GSM gateway — loaded with local SIM cards. Traffic routed through a SIM farm is technically terminated to the destination network, but it arrives via a cellular GSM leg rather than through a proper interconnection. The INVITE has to traverse the full signalling chain to the SIM farm, the farm then physically dials out over a mobile network, and the call only connects once that GSM leg establishes — which takes several seconds on its own on top of whatever routing delay already accumulated. The result is PDD that regularly exceeds 10–15 seconds. Customers hear extended silence before the ring starts, and many hang up before the called party even has a chance to answer. A vendor with chronic high PDD on mobile destinations is almost always running SIM farm termination, whether they disclose it or not.

A vendor who is answering calls locally — without ever routing them into the destination network — will often show suspiciously low PDD. There is no need to traverse multiple networks. The INVITE arrives, the 200 OK fires almost instantly, and the fake audio starts. Sub-second PDD on a route that should be touching a mobile network on the other side of the world is a red flag in the opposite direction.

Taken alone, each of these metrics is gameable. ASR tells you what fraction of calls were answered — it does not care what the caller heard after the answer. ACD tells you average conversation duration — it does not tell you whether a conversation actually happened. PDD tells you signalling speed — it can drop artificially when the vendor short-circuits the routing path, or climb suspiciously when the route takes an undisclosed detour through a SIM farm. Together, they are much harder to fake. A vendor who tries to inflate ASR while keeping ACD realistic would have to synthesise genuine-sounding multi-minute conversations for thousands of calls per day — not practical at scale. A vendor who deflates PDD to look efficient will inevitably struggle to maintain a plausible ACD. The patterns betray each other.

One nuance worth acknowledging: ACD varies legitimately by destination and traffic type. Calls to a consumer mobile line in a country with expensive international rates will naturally have lower ACD than calls to a local business line — callers self-select, they hang up faster when costs are high. Call center traffic often has shorter ACD than direct-dial traffic. So there is no universal threshold for "good ACD" any more than there is a universal threshold for "good ASR." What you are looking for is consistency and coherence within a given destination, and sudden changes in the relationship between the three metrics over time.

If you are monitoring routes manually — pulling CDR summaries and doing mental arithmetic — this kind of pattern recognition is genuinely difficult to do reliably. It is exactly what RouteInspector is designed for: making test calls through your routes, recording signalling and media, and surfacing the ASR/ACD/PDD profile alongside FAS detection so that the pattern — not just a single number — is visible at a glance.

The summary is short. ASR is a useful metric, but it is the metric most easily gamed by a motivated vendor. Treat it as a starting point for investigation, not a conclusion. If your ASR is climbing while your ACD is dropping, that is not a route getting better. That is a route being worked.

Feel free to drop questions in the comments.