Plain-English definitions of workforce management and contact centre staffing terms.
Where benchmarks appear (service level targets, attrition rates, occupancy thresholds), they represent typical figures from published WFM practice. Your operation may differ: treat them as starting points, not targets.
Callers who hang up before reaching an agent
The percentage of inbound contacts that terminate before being answered by an agent. High abandonment typically signals understaffing or long wait times. A rate above 5% is generally considered poor for inbound voice, though benchmarks vary by industry. Abandonment is not modelled directly by Erlang C; it requires a more complex model (Erlang B handles blocked calls; real abandon curves depend on patience distributions).
Total agent time per contact including wrap-up
AHT = talk time + hold time + after-call work (ACW / wrap time). It is the single most important input to any staffing model: a 10% increase in AHT requires roughly 10% more agents at the same volume. AHT should be measured at the interval level because it often varies by time of day and day of week. Many operations monitor AHT per agent to identify training needs, but using AHT as a performance target can encourage agents to rush calls at the expense of quality.
Mean wait time across all answered contacts
ASA is the arithmetic mean of the time callers wait before an agent answers, averaged across all calls that were answered (not abandoned). It is a lagging indicator: a low ASA average is compatible with brief periods of very long waits. Because of this, ASA is a weaker staffing target than service level, which sets a percentile threshold.
The rate at which employees leave and must be replaced
Annual attrition rate = (leavers in 12 months) ÷ average headcount × 100. Contact centre attrition typically runs 25–40% per year, meaning a quarter to nearly half the team turns over annually. Attrition has two cost components: recruitment cost per hire, and the productivity gap while a new hire ramps to full effectiveness. At steady state, a fraction of agents are always in the ramp period, creating a persistent capacity drain that must be factored into the contracted FTE target.
Work items awaiting processing from prior intervals
For asynchronous channels (email, tickets, cases), backlog is the volume of work received but not yet completed. Unlike voice queues, backlog accumulates and must be modelled with a flow equation rather than Erlang C. Planning for backlog requires knowing the incoming volume, existing backlog, AHT, SLA target, and desired backlog reduction rate.
The medium through which a contact arrives
Common contact channels include inbound voice (phone), live chat, email, messaging apps, and back-office (processing without a customer contact). Each channel has a different staffing model: voice uses Erlang C, live chat uses a concurrency model, and email/back-office uses a backlog flow model. Applying the wrong model to a channel overstates or understates headcount.
Number of simultaneous sessions one agent handles
In live chat, a single agent can handle multiple conversations at once, typically 2 to 5 depending on complexity and response time requirements. This makes Erlang C inapplicable to chat staffing. Instead, concurrency models use Little's Law: offered load = arrival rate × average session duration, and agents required = offered load ÷ max concurrency × (1 + shrinkage).
Labour cost to serve one customer interaction
Cost per contact = total scheduled labour cost ÷ total contacts handled in the same period. It is the most direct link between staffing decisions and business economics. A drop in volume without a proportional reduction in headcount raises cost per contact; overstaffing is visible here in a way that gross headcount numbers hide. For a meaningful number, use actual scheduled labour cost rather than required labour cost.
Total headcount currently employed, regardless of schedule
Contracted FTE is the total employment commitment: how many full-time-equivalent employees are on payroll. It differs from scheduled FTE (who is actually rostered in a given week) and from required FTE (how many the model says you need to meet demand). The gap between required and contracted FTE drives hiring recommendations. Attrition gradually erodes contracted FTE; the hiring needed to maintain capacity includes both the gap and the attrition buffer.
The standard formula for inbound voice queue staffing
Erlang C models an M/M/N queue: Poisson call arrivals, exponentially distributed handle times, and N agents sharing one queue with infinite patience. It gives the probability that an arriving caller must wait (the Erlang C value), from which the service level achievable with N agents can be derived. The formula assumes every call is eventually answered — it cannot model abandonment. For channels where agents handle one contact at a time and calls queue rather than block, Erlang C is the correct model.
Formula for blocked-call systems (no queueing)
Erlang B models a system where calls that arrive when all agents are busy are immediately lost (blocked), rather than queued. It applies to trunk line sizing and overflow routing, not to the staffing of an inbound queue where callers wait. Erlang B is sometimes used to estimate call abandonment but is an imprecise approximation for that purpose.
A normalised unit of headcount based on standard hours
One FTE represents one employee working a full standard week, typically 37.5 or 40 hours of paid time. Part-time workers contribute a fractional FTE (a 20h/wk worker is 0.5 FTE). FTE is the planning currency for budgets, capacity models, and headcount approvals because it normalises across contract types. Note that FTE measures paid time; productive (seated) FTE is lower due to shrinkage.
A prediction of future contact volume
A WFM forecast predicts interval-level contact volume for a future planning horizon, typically 2–13 weeks. Common approaches include time-series models (Holt-Winters, ARIMA), regression on drivers (seasonality, campaigns), and hybrid methods. Forecasts should be measured against actual outcomes using metrics like WAPE and MAPE, and the model retuned if accuracy falls below operational thresholds (typically 10–15% WAPE for voice).
Systematic directional error in the forecast
Bias = mean(forecast - actual) / mean(actual) × 100. A positive bias means the model consistently over-forecasts, leading to overstaffing. A negative bias means consistent under-forecasting, leading to understaffing. Bias is distinct from accuracy: a model can have low WAPE but high bias if errors in opposite directions cancel out in aggregate. Bias by day of week (e.g. consistently over-forecasting Mondays) reveals patterns the model has not captured.
Matching employment contracts to demand over a long horizon
Headcount planning sits above scheduling. It determines how many people to employ, not when they work. A headcount plan derives from the weekly FTE requirement over the planning horizon, adds an attrition buffer (how many people will leave and need replacing), and deducts contracted FTE. The output is a gross hiring target and a timeline. Headcount planning is typically done quarterly and aligned to budget cycles.
Real-time adjustment of staffing on the day of operation
Intraday management is the practice of monitoring actual volume and adherence against the plan on the day, and making tactical adjustments: moving breaks, offering voluntary time off, requesting overtime, or routing contacts differently. It requires real-time data feeds from the contact platform and is the layer between strategic forecasting and moment-to-moment performance.
The relationship between throughput, queue length, and wait time
Little's Law states: L = λW, where L is the average number of items in a system, λ is the arrival rate, and W is the average time an item spends in the system. In a chat queue, offered load (the equivalent of Erlang C traffic intensity for voice) = sessions arriving per interval × average session duration / interval length. This is the foundation of chat staffing models.
Average forecast error as a percentage of actual
MAPE = mean of |actual - forecast| / actual × 100. It treats every interval equally regardless of volume, which means a noisy low-volume overnight interval has the same weight as a busy midday peak. For this reason, MAPE is a secondary metric in WFM; WAPE is preferred for its volume-weighting. MAPE remains useful for detecting if there are specific intervals the model consistently misses.
An upper cap on how busy agents are allowed to be
Occupancy is the fraction of an agent's available time spent handling contacts. Setting a maximum occupancy (e.g. 85%) as a staffing guardrail adds a margin above the mathematically minimum headcount. Without it, Erlang C will recommend the minimum possible agents, which at high utilisation produces long, unstable queue dynamics: a small volume increase causes a disproportionate wait time spike. Most operations set max occupancy between 80% and 90%.
Fraction of agent time spent on active contacts
Occupancy = (time spent handling contacts) ÷ (total available time). At 80% occupancy, an agent spends 48 minutes per hour on contacts and has 12 minutes free. High occupancy (>90%) increases agent stress and error rates and reduces the ability to absorb sudden volume spikes. Low occupancy (<65%) is inefficient. The occupancy floor is determined by Erlang C at the minimum-staffing solution; adding more agents reduces occupancy.
Employer costs above the basic wage
On-cost (also written on-costs) is the additional employer expenditure beyond the base salary, covering payroll taxes, pension contributions, employer national insurance, health insurance, and statutory benefits. A typical on-cost uplift in the UK is 20 to 30% of base salary. Multiplying hourly wage by (1 + on-cost rate) gives the loaded labour cost used in cost-per-contact calculations.
Assigning agents to specific shifts and days
Rostering (UK) or scheduling (US) is the process of determining when each agent works. It starts from the interval-level staffing requirement and produces a set of shift patterns that cover demand as efficiently as possible. Constraints include contracted hours, working time regulations, minimum rest periods, and agent preferences. WFM tools optimise this across thousands of possible combinations; for smaller teams it is often done manually from shift templates.
The percentage of contacts answered within a target time
Service level is expressed as X% within Y seconds, for example 80% of calls answered within 20 seconds (80/20 or 80-20). It is computed as (contacts answered within Y seconds) ÷ (total contacts offered) × 100. Service level is the primary staffing target for real-time queues because it controls customer wait experience. Unlike ASA, it sets a percentile threshold rather than an average, making it more representative of the worst-case experience.
A contractual commitment on service quality
An SLA is the formal agreement with an internal department, outsourcer, or customer on the minimum acceptable service level. Breaking an SLA typically carries a financial or contractual penalty. WFM teams plan to meet or exceed the SLA. The internal staffing target is often set higher than the SLA to create a buffer against forecast error and unexpected volume spikes.
Fraction of paid time not available for handling contacts
Shrinkage converts a seated-agent requirement into a scheduled headcount: scheduled = seated ÷ (1 - shrinkage). It encompasses planned shrinkage (breaks, meetings, training, public holidays, annual leave) and unplanned shrinkage (sick leave, systems outages, unscheduled absence). A shrinkage rate of 30% means that to have 100 agents available at any time, you must schedule 143. Shrinkage rates in contact centres typically range from 25% to 40%.
A defined work period with a start time, end time, and break
A shift template defines a standard work pattern: start time, end time, break length, headcount, and the days of the week it runs. WFM teams build a library of shift patterns that collectively cover the operating hours at the right staffing levels. Common patterns include early/day/late rotations for extended-hours operations. The overlap between shifts creates the staffing profile visible in the 24-hour coverage chart.
The mathematical method used to convert volume into headcount
A staffing model translates a forecasted contact volume into a headcount requirement. The correct model depends on the channel: Erlang C for inbound voice (one agent handles one call at a time, calls queue); concurrency model for live chat (one agent handles multiple sessions); backlog flow model for email and tickets (deferred work with SLA deadlines). Applying Erlang C to chat, or chat models to voice, overstates or understates headcount significantly.
The degree to which agents follow their planned schedule
Schedule adherence = (time in adherence) ÷ (scheduled time) × 100. An agent is 'in adherence' when they are doing what the schedule says they should be doing at that moment: on a call, on break, in training. Low adherence (agents not on the phones when scheduled) is a common cause of unexpected service level drops. Adherence is typically monitored in real-time via ACD/WFM integration and targeted at 90 to 95%.
Volume of work completed per unit of time
Throughput is the rate at which work is processed: contacts handled per hour, or cases closed per day. It is a function of headcount and AHT: throughput = agent hours ÷ AHT. For backlog channels (email, tickets), throughput relative to incoming volume determines whether the backlog grows, shrinks, or stays flat. If incoming > throughput, the backlog grows; staffing must be added until the two balance.
The volume of work per unit time in Erlangs
Traffic intensity (A, in Erlangs) = (call arrival rate × AHT) ÷ interval length. With 100 calls per 30 minutes at 3 minutes AHT: A = (100 × 180) ÷ 1800 = 10 Erlangs. If A ≥ N (number of agents), the queue is mathematically unstable and wait times grow without bound. Erlang C calculates the service level given A and N.
Reduction in contact volume from automation or self-service
Deflection is the percentage of contacts that are resolved by IVR, chatbot, knowledge base, or other self-service tools before reaching an agent. A deflection rate of 20% means only 80% of the raw contact volume reaches staffed queues. Deflection typically grows over time as automation matures, so a forecast modelled with a fixed deflection rate will overstate future staffing need if automation is improving.
Volume-weighted forecast accuracy metric
WAPE = Σ|actual - forecast| ÷ Σactual × 100. By weighting each interval by its actual volume, WAPE prevents low-volume intervals from inflating the error score. A busy midday interval with a 5% error contributes more to WAPE than a quiet 2am interval with a 50% error. WAPE is the preferred headline forecast accuracy metric in WFM. Below 10% is strong; 10–20% is acceptable for most operations; above 20% indicates the model needs improvement.
The discipline of planning and managing operational headcount
WFM covers the full cycle from long-range headcount planning through to intraday management: demand forecasting, requirement calculation, scheduling, real-time adherence monitoring, and performance reporting. In a contact centre context it is sometimes called Contact Centre WFM or CC WFM. WFM tools automate the mathematical steps (Erlang C, scheduling optimisation) that would otherwise require manual spreadsheet work.
Turnella uses the correct staffing model for each channel type: Erlang C for voice, concurrency for chat, and backlog flow for email. You get the honest numbers to present to your team.