Abandonment

Abandonment is the share of contacts that leave the queue before being answered. In voice operations it's the caller hanging up; in chat it's the customer closing the window; in messaging or email it's the customer giving up before a response arrives. Abandonment matters because every abandoned contact is at minimum a service failure, often a re-contact later (which inflates the next interval's volume), and frequently a customer-experience event with revenue and CLV consequences. Abandonment is also the assumption that breaks Erlang-C — the model treats the abandoned contact as still in queue, leading to systematic over-staffing.

The mathematical foundation comes from Conny Palm's 1946 patience-distribution model — the distribution of how long a customer is willing to wait — which became the basis of Erlang-A. Modern contact centers measure abandonment as a primary metric alongside Service Level and Average Speed of Answer (ASA).

Definition

Abandonment Rate (ABA or Aban%) is the proportion of offered contacts that abandon the queue:

Abandonment Rate = (Abandoned Contacts) / (Offered Contacts) × 100%

Where Offered = Answered + Abandoned (every contact that entered queue). Some operations subtract short abandons (calls that abandon within a few seconds, often considered miss-dials or accidental disconnects) from the numerator:

Net Abandonment = (Abandoned − Short Abandons) / Offered × 100%

The short-abandon threshold is typically 5–10 seconds and is a measurement convention worth documenting in any SLA. Short-abandons happen because callers misdial, rethink the contact, or briefly experience the queue and decide to use a different channel — they're usually not service failures.

Formula / mathematics

Abandonment is a function of caller patience and queue waiting time. Conny Palm modeled caller patience as an exponential distribution with rate θ — each waiting caller has an instantaneous probability θ × dt of abandoning in the next infinitesimal interval. Mean caller patience is 1/θ.

For an Erlang-A (M/M/c+M) queue:

P(Abandon) = numerically derived from the queue's steady-state distribution

A simplified relationship: at moderate-to-high loads, abandonment rate is approximately:

P(Abandon) ≈ θ × E[Wait | served] + correction terms

Two qualitative regimes:

• Patience >> wait. Abandonment is small. Erlang-C is a good approximation. Most contact centers in steady state.
• Patience ~ wait or shorter. Abandonment dominates. Erlang-C breaks down. Erlang-A or simulation required.

The patience distribution is rarely strictly exponential in real data. Empirical patience distributions often have:

• A small mass of fast-abandoning callers (impatient or accidentally-queued)
• A large body of moderate-patience callers (the bulk of demand)
• A long tail of patient callers (callers with no acceptable alternative)

Modern survival-analysis methods can fit non-exponential patience distributions, which feed simulation models more accurately than the closed-form Erlang-A.

Practitioner use

Abandonment shows up in three contexts:

1. Capacity planning input. Mean caller patience is an Erlang-A input. Empirically estimated from ACD abandon-time data per queue.
2. SLA reporting. BPO contracts increasingly set abandonment caps (e.g., "abandonment ≤ 3%") alongside or instead of Service Level. Abandonment-based SLAs are honest about the customer experience in a way SL alone is not.
3. Operational diagnostic. Abandonment spikes are leading indicators of staffing crashes, queue malfunction, or routing failures.

Estimating mean caller patience for Erlang-A:

• From abandon-time data. The mean of actual abandon times under-estimates patience because patient callers got served (survival bias). Kaplan-Meier-style adjustment is the right approach.
• Quick proxy. Median time-to-abandon from recent intervals is a usable starting point.
• By queue and time-of-day. Patience varies — high-CLV retention queues have longer patience than mass-market support queues; weekend callers wait longer than weekday callers.

Typical industry abandonment rates:

• Top-quartile inbound voice with strong staffing: 1–3%
• Standard inbound voice operations: 3–6%
• Cost-pressured / understaffed peak intervals: 8–15%
• Service-incident / outage intervals: 25%+
• Chat / messaging: 10–25% baseline (different patience regime; customer can multitask while waiting)

Drivers of abandonment, in rough order of impact:

1. Wait time vs caller patience. Most abandonment is patience-driven.
2. Hold music / messaging design. Apologetic, anxiety-inducing messaging shortens patience; informative messaging lengthens it.
3. Channel alternative availability. Easy escape to self-service or chat reduces abandonment without reducing arrival.
4. Time-of-day. Early-morning and lunch-time callers tend to abandon faster.
5. Repeat callers. A caller who abandoned earlier and is re-attempting has shorter remaining patience.

Common failure modes

• Treating abandonment as zero in Erlang-C planning. Erlang-C assumes infinite patience. Plans built on this assumption over-state required staffing in any operation with material abandonment.
• Ignoring re-contact volume from abandons. An abandoned caller often calls back. The next interval's volume is partly the previous interval's abandons. The math closes the loop in simulation; Erlang-C doesn't.
• Conflating patience-driven abandonment with technical abandonment. Network drops, ACD failures, and routing errors are not patience-driven and shouldn't feed Erlang-A patience estimates.
• Reporting abandonment without short-abandon adjustment. A 6% raw abandon rate including 2% miss-dials looks worse than the 4% of real abandons; document the convention.
• Single-number abandonment hides distribution. Like every queue metric, abandonment is interval-distributed. A weekly 5% can hide a 30% peak-interval crash.
• Treating abandonment purely as a service metric. Abandonment has a CX/CLV cost beyond the immediate service failure. A high-CLV customer who abandoned has a higher cost than a routine-inquiry customer who abandoned. The Value Routing Model frame makes this explicit.
• Assuming patience is fixed. Patience varies by call type, time, and customer state. A static patience number is a Level 2 simplification.

Maturity Model Position

• Level 1 — Initial (Emerging Operations). Abandonment reported but not modeled; treated as residual rather than as a planning variable. Erlang-C planning ignores it.
• Level 2 — Foundational (Traditional WFM Excellence). Abandonment reported daily and weekly. Erlang-C is the staffing default; Erlang-A available but rarely used. Abandonment caps appear in some BPO contracts.
• Level 3 — Progressive (Breaking the Monolith). Erlang-A is the staffing default for material-abandonment queues. Patience estimated empirically from ACD data per queue. Abandonment reported distributionally; short-abandon adjustment standardized. Re-contact volume estimated and fed back into next-cycle forecasts.
• Level 4 — Advanced (The Ecosystem Emerges). Patience modeled per queue, time-of-day, and call type. Abandonment cost weighted by composite value — a high-CLV abandon has higher cost than a routine abandon, and routing decisions reflect this. Abandonment reported as risk distribution per WFM Labs Risk Score™. Patience-distribution modeling moves beyond exponential to empirical.
• Level 5 — Pioneering (Enterprise-Wide Intelligence). Patience predicted per customer in real time. Abandonment cost computed against customer-specific value functions. Routing and queue-management decisions adapt continuously to predicted patience and re-contact probability.

References

• Palm, C. (1946). Methods of judging the annoyance caused by congestion. Tele 4, 189–208. Original patience-distribution paper.
• Garnett, O., Mandelbaum, A., & Reiman, M. (2002). Designing a call center with impatient customers. Manufacturing & Service Operations Management, 4(3), 208–227. Modern Erlang-A analysis.
• Mandelbaum, A., & Zeltyn, S. (2007). Service engineering in action: the Palm/Erlang-A queue. Operations Research, 55(2), 165–183.
• Brown, L., Gans, N., Mandelbaum, A., et al. (2005). Statistical analysis of a telephone call center: A queueing-science perspective. Journal of the American Statistical Association, 100(469), 36–50. Empirical patience distributions in real call-center data.
• Aksin, Z., Armony, M., & Mehrotra, V. (2007). The Modern Call Center: A Multi-Disciplinary Perspective. Production and Operations Management, 16(6), 665–688.
• Koole, G. (2013). Call Center Optimization. MG Books. Practical abandonment management.
• Lango, T. (2026). Adaptive: The Workforce Transformation Architecture. Value-weighted abandonment in distributional planning.

See Also

• Erlang-A — the model that explicitly handles abandonment
• Erlang-C — the model that ignores it
• Service Level — paired SLA metric
• Average Speed of Answer (ASA) — paired wait metric
• Average Handle Time — service-time input
• Occupancy — capacity-utilization companion
• Shrinkage — gross-up factor in capacity planning
• Capacity Planning Methods — where abandonment enters planning
• Discrete-Event vs. Monte Carlo Simulation Models — when simulation models patience better than closed-form Erlang-A
• Probabilistic Forecasting — distributional inputs to abandonment risk
• Value Routing Model — value-weighted abandonment cost
• WFM Labs Risk Score™ — abandonment risk in plan-rating

Interactive tools

• Erlang Suite — erlangcalculator.wfmlabs.com. Includes Erlang A, the right tool for abandonment-aware staffing.
• Power of One — powerofone.wfmlabs.com. Visualizes the single-agent effect on ASA, which propagates directly to abandonment.