Average Speed of Answer (ASA)

From WFM Labs

Average Speed of Answer (ASA) is the mean wait time experienced by callers in a contact-center queue, measured from the moment a contact enters queue to the moment it is answered. It is one of the principal outputs of Erlang-C and Erlang-A, a common SLA in BPO contracts, and the most commonly confused-with-Service-Level metric in the industry. ASA matters because it captures the average caller experience in a single number — but that very compression is what makes it dangerous when used alone.

ASA and Service Level describe the same wait-time distribution from different angles. SL is a quantile (percentage of contacts under threshold T); ASA is the mean. The two move together at any reasonable staffing, but they encode different information, and an operation that uses one without the other is reading half the distribution.

Definition

ASA is the mean of the wait-time distribution, computed across answered contacts:

ASA = Σ wait time of answered contacts / count of answered contacts

Reported in seconds. The denominator includes only contacts that were answered (not abandoned), so ASA ignores the wait of abandoned callers — a structural pitfall when abandonment is high (see Common failure modes below).

Two scopes in operational use:

  • Time in queue. From when the contact entered the agent queue (after IVR) to answer.
  • Time from contact origination. From the customer's perspective, including IVR time. Less commonly used in WFM but increasingly common in CX-led organizations.

The default WFM-industry definition is the time in queue variant.

Formula / mathematics

For an Erlang-C M/M/c queue:

ASA = C(n, a) × AHT / (n − a)

Where a = λ × AHT (offered load in Erlangs), n = agent count, and C(n, a) is the Erlang-C delay-probability.

For Erlang-A, ASA is computed numerically from the queue's steady-state distribution, conditioned on the contact being answered (not abandoned).

ASA decomposes into two components:

ASA = P(Wait > 0) × E[Wait | Wait > 0]

The probability of any wait, times the expected wait given a wait. The first factor is reduced by adding agents; the second is reduced by reducing variance in arrival or AHT.

The relationship to Service Level at threshold T:

  • SL is the proportion of the wait-time distribution below T.
  • ASA is the mean of the same distribution.
  • For an Erlang-C queue, ASA can be computed from SL and T if the distribution is exponential. For real distributions with heavy tails, ASA can be much larger than the median wait — so ASA may look bad while SL looks fine, or vice versa.

Practitioner use

ASA shows up in three contexts:

  1. Capacity planning output. Erlang-C and Erlang-A return ASA alongside SL. Practitioners use both to size staffing.
  2. SLA reporting. Some BPO contracts use ASA as the contractual metric (e.g., "ASA ≤ 30 seconds"). Less common than SL but not rare.
  3. Diagnostic. Comparing reported ASA to reported SL surfaces distribution-shape issues — if ASA is much larger than expected from SL, the wait-time distribution has a heavy tail (some callers wait much longer than others).

Typical industry values, paired with the SL they correspond to (rough):

  • 80/20 SL → ASA ~10–15 seconds for a mid-sized queue at 85% occupancy.
  • 70/30 SL → ASA ~25–40 seconds.
  • Premium queues (90/15) often run ASA < 5 seconds because both the median and tail are tightly controlled.
  • Heavily understaffed peak intervals can have ASA > 120 seconds even if cumulative weekly SL looks adequate.

ASA is more sensitive to tail behavior than SL. A small fraction of very-long-wait calls drives ASA up disproportionately. This is operationally useful — ASA spikes are an early warning of distribution-shape problems before SL responds.

ASA vs Service Level

The two metrics are often used interchangeably in casual conversation. They are not interchangeable.

Service Level ASA
Type Quantile (proportion under threshold T) Mean of wait distribution
Sensitivity to tails Insensitive (a 600s call is "above threshold" same as a 100s call) Highly sensitive (one 600s call shifts the mean materially)
Best at detecting Most callers waiting too long Heavy tail / outlier waits
Common SLA 80/20, 90/30 (X/Y form) ASA ≤ N seconds
Erlang-C output Yes Yes

A practitioner using both reads the joint signal: SL on target with rising ASA → tail spreading (a few outlier waits getting worse). SL falling with stable ASA → bulk shift (most callers waiting longer, but no tail growth). Both metrics are projections of the same distribution; reading both detects shape changes neither one would catch alone.

Common failure modes

  • Treating ASA as Service Level (or vice versa). They report different things. An SLA written in ASA terms cannot be enforced with SL reports.
  • Excluding abandoned-call wait from ASA. The standard ASA definition uses only answered calls. When abandonment is high, the abandoned callers (who waited the longest before giving up) are excluded from the metric, lowering reported ASA exactly when service is worst. Use a paired Abandonment + ASA report, or report both ASA-answered and ASA-with-abandons.
  • Reporting daily/weekly ASA without distribution. A weekly average ASA hides the peak-interval ASA crash. The interval where ASA was 5 minutes blends into a weekly average of 30 seconds.
  • Comparing ASA across queues with different AHTs. A 60-second ASA on a 3-minute-AHT queue is a different operational signal from 60-second ASA on a 12-minute-AHT queue. Normalize when comparing.
  • Using ASA as a target without an SL pair. ASA alone can be hit by short-wait callers while leaving long-tail outliers. SL alone can be hit while average wait deteriorates. The two are complementary; neither alone is sufficient.

Maturity Model Position

  • Level 1 — Initial (Emerging Operations). ASA reported but not understood; treated interchangeably with SL. No distinction between ASA-of-answered and ASA-with-abandons.
  • Level 2 — Foundational (Traditional WFM Excellence). ASA reported daily and weekly alongside SL. Erlang-C planning produces both. Single-number reporting is the norm; abandoned-call wait excluded by convention.
  • Level 3 — Progressive (Breaking the Monolith). ASA reported distributionally — interval-level distribution, peak vs off-peak. ASA-with-abandons reported alongside ASA-answered. Joint SL + ASA reading used to detect distribution-shape changes.
  • Level 4 — Advanced (The Ecosystem Emerges). ASA modeled jointly with SL and abandonment in probabilistic capacity plans. ASA targets vary by composite-value tier — premium queues get tighter ASA targets. ASA reported as a probability band (e.g., ASA ≤ 30s with 90% confidence).
  • Level 5 — Pioneering (Enterprise-Wide Intelligence). ASA targeted dynamically per customer-segment, conditioned on real-time queue state and customer value. The single per-queue ASA target dissolves into a portfolio of per-segment commitments.

References

  • Erlang, A.K. (1917). Solution of some problems in the theory of probabilities of significance in automatic telephone exchanges. Originally in Danish; the foundational delay-formula paper.
  • Cooper, R.B. (1981). Introduction to Queueing Theory (2nd ed.). Standard treatment of mean-wait formulas.
  • Koole, G. (2013). Call Center Optimization. MG Books. ASA in modern contact-center planning.
  • Cleveland, B. (2012). Call Center Management on Fast Forward (3rd ed.). ICMI Press. Operational ASA reporting practice.
  • Lango, T. (2026). Adaptive: The Workforce Transformation Architecture. Joint distributional treatment of SL and ASA.

See Also

Interactive tools

Retrieved from "https://wiki.wfmlabs.org/w/index.php?title=Average_Speed_of_Answer_(ASA)&oldid=746"