ASR, NER, ACD and PDD: How Wholesale Voice Route Quality Is Actually Measured
- Yashar Sabunchi
- 5 hours ago
- 7 min read
A single answer-seizure ratio figure tells a wholesale buyer almost nothing on its own. ASR is the number quoted most often in route negotiations, and also the number most easily moved by things that have nothing to do with the route: the hour it was sampled, the destination mix underneath it, and the behaviour of the people being called. Route quality is a set of measurements read against each other, or it is not a measurement at all.
Sigma Telecom has worked in telecom since 2003 and in wholesale voice since 2014, carrying 150M+ minutes a month over SIP and H.323 across 1,000+ interconnections into 190+ countries, with a NOC staffed 24/7. Route quality disputes are routine work, and the first job is always agreeing on what was actually measured.
This article sets out the metrics that decide a wholesale voice relationship — ASR, NER, ACD, PDD and MOS — what each counts, what each cannot prove, and what to ask before accepting any of them as evidence.
Seizures, and why the counting decides the answer
Every one of these metrics is a ratio, and every ratio is decided by what went into the denominator. A seizure is a call attempt the far end took responsibility for — in SIP terms, an INVITE that produced a final response. Whether a particular failure counts as a seizure changes the percentage without anything changing on the route: a 404 from the terminating network, a 503 from a congested transit carrier, an attempt your own softswitch rejected before it ever left. Two carriers can look at the same traffic, produce different numbers, and both be telling the truth. Sigma reports international voice services per destination and per interconnect rather than as one blended figure, because a blend conceals precisely the destinations a buyer is asking about.
ASR: the number everyone quotes
Answer-seizure ratio is 100 times the number of seizures that produced an answer signal, divided by the total number of seizures. It derives from ITU-T Recommendation E.411 on international network management, and almost every rate discussion opens with it.
Its weakness is written into the definition. A called party who is busy, or who simply does not pick up, counts as a failed call. So does a disconnected number. ASR therefore measures the network and the called population together: a dialler campaign into a market with poor answer rates will show a low ASR over a route working exactly as it should. The reverse is worse — a route can be pushed toward a high ASR by answering calls that should never have been answered.
NER: the same traffic, without the called party
Network effectiveness ratio, defined in ITU-T Recommendation E.425, measures the ability of the network to deliver a call to the terminal and deliberately excludes what the called party then does. User-busy, ring-no-answer and terminal rejection all count as network successes, because the network did its job.
That makes NER the more honest measure of a carrier's delivery, and the pair is more informative than either figure alone. NER high with ASR low points at the destination market or the dial list. Both low points at the route. NER low is a carrier problem and nothing else.
ACD: where false answer shows up
Average call duration is total conversation time divided by the number of answered calls. Standing alone it is close to meaningless, because it is dominated by what the traffic is rather than how well it was carried — a stream of one-time passcodes and a stream of residential calls have no business being compared.
Read against ASR, it becomes the most useful false-answer indicator a buyer has. A route whose ASR climbs while its ACD collapses toward a handful of seconds is very probably answering calls into media that is not the called party: a false answer supervision configuration, a recorded announcement, or a fraud position billing from the moment of the 200 OK. It is one of the most common causes of enterprise bill-shock complaints, and a route showing the pattern should be pulled before the invoice is issued.
PDD: the metric end users actually feel
Post-dial delay is the interval between the INVITE and the first indication of progress returned to the caller, typically a 180 Ringing or a 183 Session Progress. It rarely appears on a rate sheet and generates more end-user complaints than anything else here, because it is the only metric the caller experiences directly. Long enough, and the caller hangs up believing the call failed.
PDD is also strongly regional: a delay noticed immediately in North America is unremarkable into some destinations, where ten seconds or more before ringback is normal. When it degrades suddenly on a previously stable route, the cause is usually signalling — the far end returning progress in an unexpected response, or returning none at all — which makes it a NOC conversation rather than a commercial one.
The transport layer underneath all of it
The metrics above describe call handling. Underneath them sits IP transport, which has published objectives — useful precisely because they are not the carrier's own opinion. ITU-T Recommendation G.114 sets one-way mouth-to-ear transmission time of 0 to 150 ms as acceptable for most user applications, 150 to 400 ms as acceptable provided the operator understands the effect on quality, and anything above 400 ms as unacceptable for general network planning purposes. ITU-T Y.1541 defines the quality-of-service classes IP-carried voice is planned against: Class 0 targets a mean IP packet transfer delay of 100 ms or less, delay variation of 50 ms or less, and a packet loss ratio no worse than one in a thousand, with Class 1 allowing mean delay up to 400 ms at the same variation and loss objectives.
These are planning objectives, not guarantees, and no carrier controls an entire end-to-end path. What they give a buyer is a reference point when a supplier says a route is fine and the audio says otherwise.
MOS, and what it does not settle
Mean opinion score is a quality scale from 1 to 5. Modern MOS figures on carrier dashboards are not opinions at all — they are algorithmic estimates of what a listening panel would have said, most often produced by ITU-T P.863 (POLQA), standardised in 2011 with a third edition in 2018 as the successor to PESQ. The estimate is useful and routinely over-read: it reflects only the samples that were measured, on the codec and bandwidth that were measured. A blended MOS across a mixed route tells a buyer less than a per-destination NER trend does.
Reading them together
Metric | What it counts | What it does not prove |
|---|---|---|
ASR | Answered seizures over total seizures | Route quality — it moves with called-party behaviour and dial-list quality |
NER | Network delivery to the terminal, excluding user busy and no-answer | That the calls which answered were good calls |
ACD | Conversation time over answered calls | Anything on its own; it is a comparison metric, read against ASR |
PDD | INVITE to first progress indication | A fault, until compared with what that destination normally does |
MOS | Estimated listening quality on a 1 to 5 scale | Quality across a route, once it is blended across destinations |
One-way delay | Mouth-to-ear transmission time against G.114 | Which segment of the path introduced the delay |
None of these numbers is dishonest. Each answers a narrow question, and the failure mode in wholesale is asking one metric to answer a question it was never built for.
What to ask a wholesale supplier
Per-destination reporting. A blended country-level figure is not useful. Ask for each metric broken out by destination and, where it matters commercially, by breakout — mobile against fixed, and by prefix range.
The measurement window. A 24-hour ASR and a busy-hour ASR on the same route can differ enough to reverse a routing decision. Agree which is being quoted.
Which codes are counted. Ask how the supplier treats 4xx responses and its own internal rejections in the denominator. This one question resolves a large share of quality disputes.
Route composition. Ask whether a destination is served directly or through resale, and how many hops sit behind the quoted number. Sigma's services run over 1,000+ interconnections so that a degraded route can be changed rather than explained.
The escalation path. A metric is worth what the response to it is worth. Ask what happens at 03:00 when a route degrades, and who picks up. Sigma runs a 24/7 NOC and holds ISO and PCI DSS certification covering how that operation is run.
None of this is unique to voice. A delivery receipt on an A2P route is a status report from an intermediary, not proof a handset displayed a message, and the same questions apply to wholesale SMS.
Frequently asked questions
Is a high ASR a reliable sign of a good route?
No. ASR counts busy and no-answer as failures, so it moves with the called population and the dial list as much as with the route itself. A high ASR paired with a very short average call duration is a warning rather than a good result, because it can indicate calls answered by something other than the called party.
What is the difference between ASR and NER?
NER, defined in ITU-T Recommendation E.425, measures whether the network delivered the call to the terminal, counting user-busy, ring-no-answer and terminal rejection as successes. ASR counts them as failures. NER isolates carrier performance; ASR blends it with end-user behaviour.
What causes high post-dial delay?
Most often signalling: the terminating network returning call progress in an unexpected SIP response, returning none at all, or extra hops in the path. Distance and non-geographic destinations such as satellite numbers also add delay. PDD should be judged against what a destination normally does, not a global average.
What one-way delay is acceptable for voice?
ITU-T Recommendation G.114 treats 0 to 150 ms mouth-to-ear as acceptable for most user applications, 150 to 400 ms as acceptable where the operator understands the impact on quality, and anything above 400 ms as unacceptable for general network planning purposes.
Can a MOS score settle a quality dispute?
Not on its own. A MOS figure is an estimate produced by an algorithm such as ITU-T P.863 (POLQA) from the samples it was given, on a specific codec and bandwidth. It is useful when the sampling method is disclosed and the score is read per destination, and misleading when blended.
Sigma Telecom sells international voice and messaging to operators, carriers, aggregators and enterprises from Istanbul, and has been an ITU sector member since 2018. To discuss a route, or to meet at one of the industry events we attend, use the contact page or write to info@sigmatelecom.com.
Learn more: international voice services · wholesale SMS · ISO and PCI DSS certification · industry events · contact
