Average Handle Time

From WFM Labs

Average Handle Time (AHT) is the average end-to-end time an agent spends on a contact, measured from the moment the contact is answered to the moment the agent is available for the next contact. AHT is the supply-side service-time input to every queueing calculation in workforce management — Erlang-C, Erlang-A, simulation, and capacity planning all consume it. It matters because every minute of AHT translates linearly into staffing demand: at fixed volume and service-level target, a 10% AHT increase is roughly a 10% staffing increase.

AHT is also the most commonly mis-measured metric in contact centers. The components are well-defined; the practice of which components count and which intervals to average over is not. The same operation can produce AHT values 30% apart depending on how the calculation is set up.

Definition

AHT is the sum of three components, averaged across contacts:

AHT = Talk Time + Hold Time + After-Call Work (ACW)

Where:

  • Talk Time — the conversation segment, agent connected to customer.
  • Hold Time — time the customer is on hold during the contact, agent still attached to the call.
  • After-Call Work (ACW) or Wrap — post-conversation work (note-writing, code-out, system updates) before the agent returns to available.

Reported AHT is typically:

AHT = (Σ Talk + Σ Hold + Σ ACW) / (count of handled contacts)

The measurement basis depends on three things: what counts as handled (transferred calls, abandoned-after-answer calls, supervisor-monitored calls), whether ACW is bounded (some platforms cap ACW at 5 minutes; uncapped ACW reveals real cost), and whether outliers are excluded (calls over a threshold like 30 minutes or 2 standard deviations).

For non-voice channels:

  • Chat AHT — handle time per conversation. Concurrent chats complicate the math because one agent handles N chats simultaneously; effective AHT is conversation duration / concurrency.
  • Email / case AHT — touch time, often expressed in minutes per case rather than seconds.
  • Back office — task duration; the AHT analog for non-real-time work.

Formula / mathematics

In Erlang-C terms, AHT becomes the service rate input:

μ = 1 / AHT (contacts per agent per second)

The offered load (in Erlangs) is:

ρ = λ × AHT / interval-length

For a 30-minute interval with 100 contacts at 360 seconds AHT:

ρ = (100 × 360) / 1800 = 20 Erlangs

This means 20 fully-utilized agents could just barely cover the workload — actual staffing must exceed ρ to deliver any service level (see Erlang-C and Occupancy).

A 1% increase in AHT, holding volume constant, increases ρ by 1% and increases required staffing by approximately 1% at the relevant occupancy. AHT is a linear lever on cost.

Practitioner use

AHT shows up in nearly every WFM calculation:

  • Forecasting. AHT is forecasted alongside volume by interval and day-of-week. AHT seasonality is real (post-billing-cycle calls run longer than mid-cycle). See Forecasting Methods.
  • Staffing. Erlang-C / Erlang-A consume forecast AHT to compute required agents (see Capacity Planning Methods).
  • Cost modeling. Cost-per-contact = (cost-per-agent-second) × AHT. AHT is the multiplier that turns agent cost into product cost (see Workforce Cost Modeling).
  • Quality and coaching. AHT outliers identify training opportunities — but only when normalized by call type, because complex calls legitimately run longer.
  • Automation cases. The benefit of containment (deflecting a contact to self-service or AI) is roughly AHT × cost-per-second × deflection rate. AHT economics drive automation prioritization.

Typical industry values for inbound voice:

  • Simple transactional queues (account balance, bill pay): 180–300 seconds
  • General customer service: 360–540 seconds
  • Technical support: 540–900 seconds
  • Complex account servicing / retention / claims: 900+ seconds

Chat AHT runs longer in absolute terms (commonly 600–1,500 seconds per conversation) but lower per agent-second when concurrency is 2–3.

The Complexity Premium and post-deflection AHT

A structural finding documented in the Service Demand Rebound Model: when a contact center deflects simple contacts to self-service or AI, the contacts that remain for human agents are systematically more complex. This is the Complexity Premium — post-deflection AHT runs higher than pre-deflection AHT, and the cost of remaining contacts is higher even though their volume is lower. Practitioners sizing automation cases who use pre-deflection AHT to project post-deflection cost will systematically over-state savings. The right approach: model handled-contact mix shift explicitly, project AHT by call type, and re-derive cost.

Common failure modes

  • Mixing channels. Voice AHT and chat AHT are not the same metric and shouldn't be averaged. Concurrency in chat makes the comparison wrong even after unit conversion.
  • Capped ACW. If the ACR system caps ACW at 5 minutes and the real wrap is 7, the operation is staffing for 5 and paying for 7. The gap shows up as adherence and idle-state pollution.
  • Outlier-stripping in reporting, full-data in planning. Common pitfall: report trimmed AHT to leadership (looks good) and plan with full AHT (need more agents). Or worse, plan with trimmed AHT and miss SL.
  • AHT averaged across heterogeneous mixes. Skill-A calls at 5 minutes and skill-B calls at 15 minutes averaged into "10 minutes" produces wrong staffing at both queues. Forecast and plan AHT by call type.
  • Treating AHT as fixed. AHT shifts with channel mix, automation maturity, agent tenure (new hires run longer; see Speed to proficiency curve), and seasonality. A flat AHT assumption is a Level 2 tell.
  • Using AHT as a coaching metric in isolation. "Reduce AHT" pressure produces premature wraps, repeat contacts, and FCR collapse. AHT is a planning input, not a quality metric.
  • Confusing handle time with talk time. Talk Time alone misses Hold and ACW — the full handle is what staffing needs. Some legacy reports surface only Talk Time.
  • Ignoring post-deflection mix shift. Modeling the post-AI operation with pre-AI AHT systematically inflates expected savings. See Service Demand Rebound Model.

Maturity Model Position

  • Level 1 — Initial (Emerging Operations). AHT reported as a single average; planning uses last month's number; outliers and channel mix not separated.
  • Level 2 — Foundational (Traditional WFM Excellence). AHT forecasted by interval and day-of-week, fed into Erlang-C interval staffing. Channel-specific AHT (voice vs chat) tracked separately. AHT trends visible in WFM reporting; outliers identified but not modeled.
  • Level 3 — Progressive (Breaking the Monolith). AHT modeled by call type and skill, with explicit drivers (tenure, time-of-day, contact reason) and forecasted using regression-style methods (see Regression for Forecasting). AHT distributional spread (not just mean) feeds Probabilistic Forecasting. AHT shift from automation projects projected explicitly via a mix-shift model.
  • Level 4 — Advanced (The Ecosystem Emerges). AHT is segmented by composite-value tier and modeled jointly with channel-mix decisions. The Service Demand Rebound Model is applied to project Complexity Premium for any deflection initiative. AHT becomes a portfolio variable across human, hybrid, and agentic pools (see Three-Pool Architecture).
  • Level 5 — Pioneering (Enterprise-Wide Intelligence). AHT is dynamically set per-customer-segment based on learned value functions; the single AHT number is replaced by a per-interaction service-time distribution managed in real time across human and agentic supply.

References

  • Cooper, R.B. (1981). Introduction to Queueing Theory (2nd ed.) — service-time mathematics.
  • Koole, G. (2013). Call Center Optimization. MG Books. Standard reference for AHT use in capacity planning.
  • Aksin, Z., Armony, M., & Mehrotra, V. (2007). The Modern Call Center: A Multi-Disciplinary Perspective. Production and Operations Management, 16(6), 665–688. AHT segmentation and effective service rate.
  • Lango, T. (2026). Adaptive: The Workforce Transformation Architecture. Complexity Premium and post-deflection AHT modeling.

See Also

Interactive tools

Retrieved from "https://wiki.wfmlabs.org/w/index.php?title=Average_Handle_Time&oldid=741"