Abandonment and caller patience — the metric Erlang C ignores
Introduction
Service level and average speed of answer get all the attention, but there is a quieter metric that tells you more about how a queue actually feels to a customer: abandonment. It’s the percentage of contacts that give up before they’re answered. Classic Erlang C — the model most planners staff with — pretends it never happens, assuming every caller waits patiently forever. Real callers don’t. This article is about what drives abandonment, how to think about caller patience, and how to forecast and target it honestly.
What abandonment actually measures
Abandonment is the share of offered contacts that disconnect before reaching an agent. It matters for two opposite reasons at once. It’s a service failure — a customer who wanted help and didn’t get it, who may call back (inflating your volume), complain, or churn. But it’s also a relief valve: every caller who abandons is one fewer in the queue, which is why a badly understaffed queue doesn’t grow without limit. Erlang C, by assuming infinite patience, captures neither effect — it will happily tell you a queue is stable when in reality it only “works” because a third of callers are hanging up.
Caller patience: the hidden input
Patience is how long a caller will wait before giving up, and it varies enormously by context. For voice, mean patience is often in the 60–90 second range, but it depends on the reason for contact (someone chasing a refund waits far longer than someone with a quick query), the alternatives available (if your web chat is good, phone patience falls), and whether the caller has been told their position or wait time. Patience is also not constant during a call: there’s often a cluster of very early abandons (wrong number, changed their mind in the first few seconds) and then a slower bleed as the wait drags on.
The awkward truth is that patience is hard to measure, because you only ever observe the patience of the people who abandoned. The callers who waited and got through might have waited far longer — you simply don’t know. That censoring is why measured patience tends to understate true patience, and why abandonment figures should be treated as an indicator rather than a precise dial.
What drives it
Three things move abandonment more than anything else. The first is wait time — unsurprisingly, the longer the queue, the more people leave, and the relationship is non-linear, as the diagram shows. The second is expectation management: an announced position in queue or estimated wait can either reduce abandonment (callers settle in) or increase it (callers hear “20 minutes” and leave) depending on how it’s framed. The third is the value of the contact to the customer — high-stakes reasons hold people far longer than low-stakes ones, which is why a blended abandonment number across all call types can be misleading.
Forecasting and targeting it
To plan for abandonment rather than just report it, you need a model that includes patience. That’s exactly what Erlang A adds — a patience parameter on top of Erlang C, so it predicts how many callers will abandon at a given staffing level. Erlang X goes further again, adding busy-tone blocking and the redials that abandoned callers generate. Both will typically tell you that you can hit your service target with slightly fewer agents than Erlang C demands, because abandonment quietly relieves the queue — but lean on that too hard and you’re effectively staffing on the back of customers giving up, which is no plan at all.
On targets: a common standard is to keep abandonment below 5%, but the right number depends on your customers and your reasons for contact. The more useful discipline is to track abandonment alongside service level and by call type, and to watch the short-abandon band separately — calls lost in the first five seconds usually aren’t a staffing problem and shouldn’t be allowed to flatter or damn your numbers.
The practical takeaway
Treat abandonment as a customer-experience metric first and a staffing relief valve second — never the other way round. Measure it by call type, separate the early abandons, and when you size an impatient queue, use a model that actually accounts for patience rather than pretending it away. If you want to see how staffing, patience and abandonment move together, the Erlang A and Erlang X calculators make the trade-offs visible in a few clicks, and the From Erlang to Excel paper builds the maths from scratch.
Put it into practice: model abandonment with the Erlang A calculator, or add redials and blocking with Erlang X.