Managing customer expectations in Queue: Estimated Wait Time and position in queue
Why announcing estimated wait time or position in queue is one of the highest-return, lowest-cost levers in the contact centre, and how to do it without it backfiring.
Seven cautions before you proceed.
The case for announcing estimated wait time (EWT) and position in queue (PIQ) is strong. But implementing them poorly is worse than not implementing them at all. The following cautions are not theoretical. They are the most common ways this goes wrong in practice.
- 1Announcements improve the experience of waiting. They do not fix the queue. A contact centre that needs a three-minute EWT announcement is one that is already failing customers. The underlying causes are structural: too few agents on the floor, skills split too finely across too many queues, or contact types that should be resolved through self-service before they reach a person. Improving the waiting experience without fixing the capacity problem can divert leadership attention from the changes that would actually matter. See the companion document, The Physics of Waiting, for the structural levers.
- 2The estimate breaks down when you need it most. EWT models are calibrated on normal, near-steady-state operating conditions. Mass logout events, such as a team leaving for a meeting or a shift changeover handled poorly, can remove a large share of available agents in under a minute. Some platforms can detect agent-state change events and trigger a suppression mode; that capability is worth configuring before you need it.
- 3Trust, once lost, is slow to rebuild. Customers who hear four minutes and wait twelve do not simply discount the announcement. They apply that experience to every future interaction. The asymmetry is significant: a correct EWT produces modest goodwill; an incorrect one degrades long-run trust by a larger amount.
- 4Callbacks have their own failure modes. Virtual queuing is one of the strongest tools available for managing a deep queue. But wrong numbers, missed callbacks, voicemail failures, and calls routed to agents who lack the right skills all generate secondary contacts and additional customer effort.
- 5Walkaways from an EWT announcement include some customers who would have waited happily. The self-selection mechanism described in section 03 is genuinely beneficial on average: low-patience customers balk at the door and free capacity for the rest. But EWT is a blunt instrument. The same announcement that sends an impatient customer home also sends home some customers who had enough patience to wait.
- 6Customers from diverse backgrounds may not benefit equally. EWT announcements assume the caller can hear, understand, and act on spoken information in the announced language. CALD communities, customers with hearing impairments, elderly customers, and customers in noisy environments may not be able to use the announcement at all. Position-in-queue numbers have a modest advantage here since a numeral is more universally understood than a sentence.
- 7None of this works for a sustained capacity shortfall. Everything in this document assumes the queue will recover within hours. When demand persistently exceeds capacity over days or weeks, callback programmes accumulate a backlog that raises average handle time, EWT models lose calibration, position numbers remain permanently in the crisis range, and customer trust in the announcement system collapses. At that point, the correct response is additional headcount, automation of high-volume simple enquiry types, and reducing contact volume at source.
With those cautions stated: the evidence for well-implemented, accurately modelled EWT and PIQ announcements still favours doing this over silence, in normal operating conditions, with appropriate calibration and monitoring. The list above is not a reason to do nothing. It is a checklist of the ways doing it poorly is worse than not doing it at all.
Uncertainty is a hidden tax on your queue.
Customers do not just wait. They suffer through not knowing. Research in behavioural queueing consistently finds that the same objective wait feels longer and triggers abandonment sooner when no information is provided. The pain is not just the time. It is the open question of whether this is ever going to end.
Customer patience is not a fixed number. It is information-sensitive. When customers have no idea how long they will wait, they conservatively abandon early. Tell them four minutes and the customers with five-minute patience stay. Tell them nothing and a meaningful fraction of those same customers leave at minute two, because they cannot know whether they are close. Removing that uncertainty is the mechanism. Everything else follows from it.
Studies of emergency department and call centre queues consistently find that removing wait information increases abandonment by 15 to 30% compared to equivalent queues with basic EWT, even when objective wait is identical (Hui and Tse, 1996; Whitt, 1999). The queue did not get worse. The uncertainty did.
With EWT, callers self-select. Those whose patience genuinely exceeds the stated wait stay; those who cannot wait that long leave at the door rather than partway through. This shifts abandonment to the balking stage, before they have entered the queue, which has very different implications for average wait and service level for those who remain (Armony, Shimkin and Whitt, 2009).
EWT and position: same goal, different mechanisms.
There are two standard ways to tell a customer where they stand. They have meaningfully different properties for both the customer and your platform's implementation complexity.
Announces a specific time: "Your estimated wait is approximately four minutes." Directly answers the question the customer actually needs answered. More actionable: they can decide to call back, try another channel, or stay. Requires a live model to compute, and that model can fail in the conditions described in section 00.
Announces a rank: "You are caller number five in the queue." Easier to compute accurately since it is always a true count. Requires the caller to mentally convert to time. Research shows it is somewhat less effective at reducing abandonment than EWT but still meaningfully better than silence (Munichor and Rafaeli, 2007).
Provide both: "You are number three in the queue, estimated wait approximately six minutes." Update EWT every 60 to 90 seconds. This removes both uncertainties simultaneously: callers know their rank (fairness signal) and the implied time (decision signal). The combination produces the strongest abandonment reduction in the literature.
Self-selection: the invisible queue manager.
The benefit of EWT is not mainly about keeping callers calm. It is about triggering a sorting effect at the front door. When a caller hears the wait, two things happen: callers whose patience exceeds the announced wait stay in queue; callers whose patience falls below it hang up immediately. Those who balk do so at zero cost to queue capacity and free space for callers who will actually be served.
Armony, Shimkin and Whitt (2009) proved this rigorously: in a many-server queue with abandonment, accurate delay announcements strictly reduce average delay for served customers while preserving throughput. Callers who balk at the front are doing so at zero cost; callers who stay are pre-selected to have enough patience to be served. The queue settles into a lower-wait equilibrium than the no-announcement state, with identical staffing.
Announce a high wait. Many low-patience callers balk. Load drops. Wait drops. Next caller hears a lower wait. Fewer balk, but wait also drops further. The system finds a new, lower equilibrium, with the same agents and the same arriving demand.
See what announcements do to your queue
Adjust the scenario, select a country/region, and compare all announcement strategies side by side. Deep queue warnings appear when position numbers cross the demoralising threshold.
| Metric | No announcement | PIQ only | EWT only | PIQ + EWT |
|---|---|---|---|---|
| Avg wait (served) | -- | -- | -- | -- |
| Service level (30s) | -- | -- | -- | -- |
| Abandonment | -- | -- | -- | -- |
| Balking at entry | -- | -- | -- | -- |
Accuracy matters. Inaccuracy can backfire badly.
EWT only works if customers trust it. A wait announcement is an implicit promise. Break it, tell someone four minutes and make them wait twelve, and you do not just lose that caller. You train future callers to ignore your announcements. An untrustworthy EWT system degrades to something worse than silence, because customers now experience both the wait and the betrayal.
If you announce two minutes and deliver five, callers who would have stayed for three minutes start abandoning around the two-minute mark, because that is when they believe they have been lied to. You have manufactured a new abandonment spike at the announcement threshold, net of what the queue itself would have produced. The U-shaped curve below shows this clearly.
Announcing a slightly longer wait than you will actually deliver removes marginally impatient callers (genuinely positive) and creates a pleasant surprise when callers are answered ahead of schedule. This improves satisfaction scores for answered calls and builds institutional trust over time (Larson, 1987; Maister, 1985).
How EWT bias shapes total abandonment
Slide from underestimating to overestimating. The U-shaped curve shows both the wait (blue) and total abandonment (orange) as the announced EWT drifts from truth.
Position in queue: less precise, easier to trust.
Position in queue has one major structural advantage over EWT: it is always true. You always know exactly how many people are ahead of a caller. There is no model, no estimate, no confidence interval. Just a count. EWT by contrast requires a live forecasting model that can diverge from reality.
When PIQ outperforms EWT
Your handle times are highly variable. Callers are regulars who know your speed. Your modelling capability is limited. You have high volatility and position updates feel fair because they are obviously true. Your EWT model has a history of inaccuracy.
When EWT outperforms PIQ
Callers are infrequent or new. High-value decisions hinge on wait time, such as whether to try another channel. Handle time is consistent enough to model reliably. You have multichannel context and can offer callbacks at the EWT threshold.
For most operations: announce position first (always available, always true), then add EWT as a second signal. When they conflict, say position two but EWT says 12 minutes, customers trust position because it is concrete. This is useful information: it signals something unusual, such as two very long calls ahead.
A PIQ or EWT announced once at entry and never updated is almost as useless as silence after 90 seconds. The "you have moved up" signal is actively reassuring and reduces intermediate reneging. Re-announce updated PIQ and EWT at every position change, and on a fixed cadence every 60 to 90 seconds if no position change has occurred. This applies equally to EWT: a static EWT that does not reflect actual queue progress becomes untrustworthy quickly.
Patience and trust vary by market.
The EWT effect is not uniform. Its magnitude depends on two cultural variables: baseline patience (how long callers are willing to wait before abandoning regardless) and institutional trust (how much they believe the announcement). Both vary meaningfully across markets.
Within any single country, CALD communities often carry patience norms from their heritage culture. A caller from a culture with higher baseline patience may wait significantly longer than the average for that market, making EWT announcements less critical for retention but more important for managing service expectations. Conversely, language barriers reduce the effectiveness of IVR-delivered EWT: a caller who does not fully parse "approximately seven minutes" cannot make an informed decision. Organisations serving CALD populations should consider multilingual EWT delivery and test PIQ (a number, universally understood) as the primary signal where language coverage is incomplete. The country selector in the calculator above uses population-average patience; your real caller mix may differ significantly.
Three implementation choices that define your outcome.
The EWT model
Use your platform's native EWT first. Modern CCaaS platforms (Genesys, NICE CXone, Amazon Connect, Avaya, Five9, Salesforce Service Cloud) ship with built-in EWT engines that integrate real-time queue state, agent states, and historical AHT. These outperform hand-rolled estimates because they have access to data your IVR cannot easily replicate, including current call progress, agent skill weights, and routing priority. Configure and calibrate the platform model before building your own.
Where platform EWT is unavailable or unreliable: use a rolling AHT from the last 10 to 15 completed calls (not the daily average, which is too stale), add an estimate of remaining handle time for calls currently in service (approximated as half AHT), and apply a 15 to 25% overestimate buffer. Critically: EWT breaks down on mass logout events such as team meetings, fire drills, shift handovers, and sudden skill reassignments.
Minimum viable: platform EWT with calibration review monthly. If building your own: rolling AHT from last 15 calls plus 20% buffer, suppressed on mass-logout detection.
Update frequency
Update PIQ at every position change. This is non-negotiable and is the primary retention signal: "you are now caller number two" is one of the most effective abandonment reducers available. For EWT, update every 60 to 90 seconds regardless of whether position has changed, because EWT can drift independently of position (if the calls ahead are running long, for example). If EWT is moving upward between updates, acknowledge it: "your estimated wait has increased slightly to approximately eight minutes." A wait that grows without explanation feels dishonest. A wait that grows with a brief acknowledgement remains trustworthy.
Silent stretches longer than 90 seconds meaningfully increase intermediate reneging (Maister, 1985). Callers infer from silence that they have been forgotten or that the system has failed.
Update PIQ at every position change. Update EWT every 60 to 90 seconds. Acknowledge upward drift explicitly.
Callback integration
EWT announcements are most powerful when paired with a callback offer at the moment the EWT is stated. "Your estimated wait is approximately eight minutes. Press one and we will call you back without losing your place." Trigger the callback offer at the same moment as EWT, not as a consolation after callers have suffered longer. Some platforms support position-hold callbacks where the caller's place is literally preserved in the queue as a virtual caller. This is the highest-quality implementation.
Trigger callback offer at the EWT announcement. Do not delay it to a later prompt.
Large position numbers hurt more than they help.
Everything above assumes the queue is manageable. Once the queue runs deeper than roughly position 10 to 12, the playbook changes. Position announcements stop being reassuring and start being demoralising. Beyond that threshold, announced position produces higher abandonment than no announcement at all (Munichor and Rafaeli, 2007). The number is too large to generate a progress narrative. Callers cannot visualise "23 people ahead of me" resolving. They feel trapped.
This section covers two distinct situations that look the same on the surface but require different responses: acute deep queues (a surge that will resolve within hours) and sustained deep queues (chronic over-demand that persists for days, weeks, or longer). The tactics that work for one can make the other worse.
100 agents, 5-minute AHT baseline
Slide the queue depth. The chart shows the relative abandonment multiplier by position and classifies the current depth into a phase with a specific response recommendation.
Three-phase response framework.
Position 1 to 10, EWT under 3 minutes (100-agent baseline)
Announce both position and EWT. Update at every position change and every 60 to 90 seconds. Offer callback as an option, not the lead. The standard playbook applies. Your goal is to retain high-patience callers, self-select low-patience ones at the door, and maintain trust through accurate, frequent updates.
Script: "You are caller four, estimated wait approximately two minutes. Press one to keep your place and receive a callback."
Position 11 to 40, EWT 3 to 12 minutes
Suppress the position number. Switch to EWT-only with an immediate, prominent callback offer that leads the message. State that the caller's place is held (fairness signal without the demoralising count). Activate overflow routing: secondary skill groups, blended agents from adjacent queues, or contracted overflow providers. This is the window where proactive load management makes the most difference and where acute strategies are still effective.
Note on WFM alignment: overflow routing, skill reassignment, and ad hoc blending are often constrained by WFM system rules, schedule adherence targets, and reporting configurations. If your WFM system counts overflow handling as out-of-skill activity and penalises it in adherence metrics, agents will resist. Alignment with WFM and reporting teams before a crisis is essential: agree on the rules in advance, or you will spend the crisis arguing about metrics instead of answering calls.
Script: "Your estimated wait is approximately eight minutes. Press one now and we will call you back without losing your place." Drop position from the script entirely.
Position 40+, EWT over 12 minutes, or any sustained multi-day deep queue
Standard queue management has lost. At this depth, the queue is not self-correcting. Critically: several tactics that work in acute situations become counterproductive when sustained.
Callbacks in sustained deep queues cause longer handle times. When callback volume is large and sustained, agents are repeatedly handling callbacks from frustrated customers who have already waited once, had their expectations set and reset, and are calling with compounded anxiety. These calls take longer. They raise AHT, which reduces throughput, which deepens the queue further. Callback is a short-term pressure valve. Sustained, it becomes a handle-time multiplier that worsens the condition it was deployed to treat.
EWT is unreliable when the queue is still growing. Suppress specific EWT at this stage. Use banded language ("under 30 minutes") or drop EWT entirely in favour of a callback-as-default message without a stated time. A wrong 45-minute estimate is worse than no estimate.
Source-level resolution is the highest-leverage action: identify the top contact driver in real time and push proactive outbound (SMS, email, push notification, app message) to the affected population. One well-targeted proactive communication can eliminate hundreds of inbound contacts. This requires knowing your contact reason mix in real time, which is the analytics capability to build before you need it.
Sustained crisis requires staffing solutions, not announcement solutions. See the rules section for sustained-period tactics.
Operational levers and their limits.
Eight rules for the next planning meeting.
- Announce something. Anything beats silence. Position in queue, even without EWT, reduces abandonment. Use your platform's native EWT wherever it exists. Do not let perfect be the enemy of good.
- Overestimate, do not underestimate. Buffer your EWT by 15 to 25%. Delivering early builds trust. Delivering late is a trust breach that erodes future announcement effectiveness. The U-curve is steep on the underestimate side.
- Update frequently, including EWT. Re-announce PIQ at every position change. Update EWT every 60 to 90 seconds. If EWT is drifting upward, acknowledge it. Silent stretches longer than 90 seconds increase reneging.
- Pair EWT with callback, and trigger it immediately. The combination removes callers who cannot wait without losing them as contacts. Offer callback at the moment you announce the wait, not as a consolation after they have suffered through it.
- Suppress position numbers past position 10 to 12. Large queue positions cause more abandonment than silence. Switch to EWT-only above your threshold, lead with callback, and drop the position count from the script entirely.
- Pre-align with WFM and reporting before a crisis hits. Overflow routing, skill blending, and callback pools create reporting anomalies. If adherence metrics penalise the right behaviours, agents will stop doing them. Build crisis-mode reporting rules into your runbook and agree them in advance.
- Plan for the unexpected: build acute surge runbooks. For short periods of deep queues (hours to one day), your response is: suppress PIQ, lead with EWT and callback-as-default, activate overflow, identify and address the contact driver. These tactics work for acute surges. Define the thresholds that trigger each action and test them in a non-crisis period.Acute surge tactics: overflow routing, callback-as-default, digital deflection, proactive outbound to known affected customers, EWT suppression if model is unreliable.
- Sustained deep queues require staffing solutions, not announcement solutions. Announcements cannot resolve a chronic capacity shortfall. For periods lasting more than one to two days: callbacks accumulate and raise AHT; EWT loses calibration; customer trust degrades. The response set shifts to: allocating overtime immediately, engaging agency or contractor staffing (lead time is days to weeks, plan ahead), accelerating automation for high-volume simple enquiry types, reviewing IVR containment rates for self-serviceable contacts, and for longer-horizon demand growth, workforce planning to hire in advance of the demand curve. Each announcement tactic should have a defined expiry condition: "we switch to this when queue is less than X, and we escalate to that when the condition persists beyond Y hours."Sustained period tactics: overtime, contractor augmentation, automation of simple enquiry types, IVR containment improvement, proactive source-level resolution, long-run hiring ahead of forecast demand growth.
The mathematical extension: Erlang-A with strategic balking.
Everything above is grounded in an extension of the standard Erlang-A (M/M/c+M) model. Here is the formal bridge for your WFM team, analytics engineers, and anyone building the EWT model.
| Concept | Formulation | Symbol |
|---|---|---|
| Arrival rate (no announcement) | Poisson with rate | λ |
| Customer patience | Exponential distribution, mean | 1/θ |
| Balking probability given announced EWT | P(patience < EWT_ann) = 1 − exp(−θ × EWT_ann) | β |
| Effective arrival rate after balking | λ × exp(−θ × EWT_ann) | λeff |
| Honest EWT fixed point | W* such that Wq(λ × exp(−θW*)) = W* | W* |
| Expectation violation (underestimation) | Additional abandonment for callers with patience in (ann, W_true) | δ |
| Total abandonment | balk + violation + (1 − balk − violation) × renege | Ptotal |
Scientific references
- Armony, M., Shimkin, N. and Whitt, W. (2009). "The Impact of Delay Announcements in Many-Server Queues with Abandonment." Operations Research, 57(1), 66-81. Core theoretical result: accurate EWT strictly reduces average delay for served customers in M/M/c+M queues.
- Munichor, N. and Rafaeli, A. (2007). "Numbers or Apologies? Customer Reactions to Telephone Waiting Time Fillers." Journal of Applied Psychology, 92(2), 511-518. Empirical evidence that position numbers above 10-12 increase abandonment above silence baseline.
- Whitt, W. (1999). "Improving Service by Informing Customers About Anticipated Delays." Management Science, 45(2), 192-207. Early theoretical treatment of delay announcements and customer behaviour.
- Hui, M.K. and Tse, D.K. (1996). "What to Tell Consumers in Waits of Different Lengths: An Integrative Model of Service Evaluation." Journal of Marketing, 60(2), 81-90. Establishes that any information is better than silence, and that the benefit varies with actual wait length.
- Maister, D.H. (1985). "The Psychology of Waiting Lines." In The Service Encounter, Czepiel, Solomon and Surprenant (eds). The foundational taxonomy: uncertain waits feel longer, unexplained waits feel unfair, and progress signals reduce perceived wait. Underpins the update-frequency recommendation.
- Larson, R.C. (1987). "Perspectives on Queues: Social Justice and the Psychology of Queueing." Operations Research, 35(6), 895-905. On queue fairness, the value of visible progress, and why FIFO is not merely efficient but experienced as just.
- Zhou, R. and Soman, D. (2003). "Looking Back: Exploring the Psychology of Queuing and the Effect of the Number of People Behind." Journal of Consumer Research, 29(4), 517-530. Demonstrates that relative position (people behind you) affects persistence in queue, relevant to PIQ framing.
Work with us
Want to apply these principles to your call centre?
Book a 15-minute working session with Graeme to explore Estimated Waiting Time and Positon In Queue for your contact centre platform.
Book a working session