Occupancy

Occupancy is the fraction of an agent's logged-in, available time that is actually spent handling contacts. It measures how loaded the workforce is — the dual of Service Level at fixed staffing. Occupancy matters because it is the cost-side companion to service: high occupancy means the operation is squeezing producing time out of paid time, but sustained high occupancy is also the leading indicator of attrition, error rate, and burnout. The occupancy band an operation runs in encodes how it has resolved the cost-vs-experience trade-off.

The conventional industry view treats 85% as the upper bound for sustainable occupancy. The mathematical reality is more nuanced: occupancy is a product of arrivals, AHT, and capacity, and the relationship to Service Level is non-linear. Push occupancy past about 90% and SL collapses; below about 70% and cost discipline collapses. The healthy operating band is narrower than most cost-pressured plans assume.

Definition

Occupancy (often denoted ρ) is the offered load divided by the service capacity:

ρ = (λ × AHT) / (n × interval-length)

Where:

• λ = arrival rate (contacts per second)
• AHT = average handle time (seconds)
• n = number of agents staffed
• interval-length = the reporting window (typically 1800 s for a 30-minute interval)

Equivalently:

ρ = total handle-time / total agent available-time × 100%

In Erlang terms, ρ is the offered load measured in Erlangs divided by n. ρ < 1 is a stable queue; ρ ≥ 1 is an unstable queue (the queue grows without bound).

Two definitions in operational use:

• Theoretical (Erlang) Occupancy — the offered load divided by capacity. Used in planning and Erlang-C math.
• Reported (Operational) Occupancy — handle time divided by available time (logged-in time minus off-phone activities). Used in reporting and adherence.

The two diverge by adherence and shrinkage realization. A plan can call for 85% theoretical occupancy and the operation can report 92% if agents took fewer breaks than planned.

Formula / mathematics

For an M/M/c queue at steady state, occupancy and the Erlang-C delay-probability are coupled:

P(Wait > 0) = C(n, ρ × n) — the Erlang-C probability of any delay

P(Wait > T) = C(n, ρ × n) × e^{−n(1 − ρ) × T / AHT}

The Service Level achieved at occupancy ρ depends nonlinearly on n:

• At ρ = 0.85 with n = 10, SL is poor (long delays for many callers).
• At ρ = 0.85 with n = 50, SL can hit 80/20 with the same offered load.
• This is the pooling principle — large pools tolerate higher occupancy at fixed SL than small pools.

This is why a small specialty queue cannot run at the same occupancy as a large generalist queue without service collapse. It's also why pooling fragmented skill queues (where routing rules permit) lifts occupancy without sacrificing SL.

The cognitive analog of occupancy — applied to AI-supervising humans rather than call-handling agents — is the ρ_max parameter in the Cognitive Portfolio Model (N*). Both express the load-vs-capacity ratio, but the cognitive variant has a much lower upper bound because supervision is more cognitively expensive than handling.

Practitioner use

Occupancy is used three ways:

1. Planning input. At plan time, the staffing model assumes a target occupancy band (e.g., 80–85%) and solves for staffing that delivers SL inside that band.
2. Operational metric. At intraday and weekly review, reported occupancy is checked against plan to detect over- or under-staffing.
3. Health indicator. Sustained occupancy outside the planned band predicts attrition, schedule conformance erosion, and quality drift.

Typical band by environment:

• Cost-pressured BPO operations: 85–90% target. Risk-tolerant; high attrition assumed in the cost model.
• In-house customer service: 75–85%. Balances cost discipline against retention.
• Premium / high-CLV queues: 65–75%. The lower band buys responsiveness and protects the experience.
• Specialty / technical / complex queues: 60–70%. Smaller pools require lower occupancy to deliver SL.
• AI-supervising / cognitive load roles: 40–60% (per Cognitive Portfolio Model (N*)).

Drivers of occupancy at fixed staffing:

• Volume up → ρ up
• AHT up → ρ up
• Schedule fit to interval pattern matters more than total hours; a misshaped schedule can have low average occupancy and intervals at >100% offered load (which means abandonment + SL collapse).

Why 85% matters

The 85% rule of thumb has both a math and a behavior basis. Mathematically, Erlang-C delay grows nonlinearly as ρ → 1. The slope of (delay) with respect to (ρ) is gentle from 0.50 to 0.80 and steepens dramatically from 0.85 to 0.95. Behaviorally, sustained occupancy above 85% is consistently associated with elevated attrition, sick-time, error rate, and adherence breakdown. The number is not a scientific constant — it's a working envelope reconciled across queueing theory and operational evidence.

Common failure modes

• Confusing occupancy with utilization. Occupancy = handle / available. Utilization = handle / paid. They differ by Shrinkage. Both are useful; conflating them produces wrong staffing.
• Reporting actual occupancy and using it to estimate variance. When the operation is understaffed, occupancy rises to compensate, masking the real staffing gap. Variance analysis must use planned occupancy, not actual. (See the WFM Variance Analysis tool.)
• Single-number occupancy hides interval distribution. A shift can average 80% occupancy while spending 30% of intervals over 95%. The over-95% intervals are where SL fails, attrition risk concentrates, and customer abandonment spikes.
• Ignoring pool size. Small queues can't run at the same occupancy as big ones at fixed SL. A 5-agent specialty queue at 85% occupancy will deliver much worse SL than a 50-agent generalist queue at 85%.
• Cost-pressured operations targeting >90%. This is a Level 1/Level 2 pattern that hides cost in attrition and quality. The savings are real on the labor line and are usually larger on the attrition + quality lines as a leakage.
• Treating occupancy as the lever rather than the consequence. Occupancy is what falls out of (volume + AHT + staffing). The levers are forecast, schedule, and the staffing decision; occupancy is the readout.

Maturity Model Position

• Level 1 — Initial (Emerging Operations). Occupancy reported but not targeted. Staffing decisions made on volume and SL alone; occupancy looked at after the fact. Variance analyses use actual occupancy, hiding gaps.
• Level 2 — Foundational (Traditional WFM Excellence). Single-number occupancy target (often 85%) drives capacity planning. Daily and weekly occupancy reported. The 85% rule of thumb anchors staffing across the operation, regardless of pool size.
• Level 3 — Progressive (Breaking the Monolith). Occupancy bands set by pool size, channel, and skill. Variance decomposed using planned occupancy (correct method). Occupancy distribution (not just mean) tracked. The relationship to attrition and quality measured rather than asserted.
• Level 4 — Advanced (The Ecosystem Emerges). Occupancy treated as a portfolio variable. Different pools (Pool AA, Collab, Spec — see Three-Pool Architecture) carry different occupancy targets calibrated to their value contribution. Cognitive occupancy (ρ_max from Cognitive Portfolio Model (N*)) used for AI-supervising roles. Multi-objective plans surface the cost / CX / EX / risk trade-off explicitly.
• Level 5 — Pioneering (Enterprise-Wide Intelligence). Occupancy adapts in real time per agent and per pool, calibrated to learned cognitive-load and value functions. Occupancy targets become continuous, not discrete; the 85% threshold becomes a special case of a learned schedule.

References

• Erlang, A.K. (1909, 1917). The original telephone-traffic papers — basis of the offered-load concept.
• Cooper, R.B. (1981). Introduction to Queueing Theory (2nd ed.). Foundational ρ math.
• Koole, G. (2013). Call Center Optimization. MG Books. Pooling principle and the ρ-vs-SL relationship.
• Aksin, Z., Armony, M., & Mehrotra, V. (2007). The Modern Call Center: A Multi-Disciplinary Perspective. Production and Operations Management, 16(6), 665–688. Empirical occupancy-attrition relationships.
• Lango, T. (2026). Adaptive: The Workforce Transformation Architecture. Pool-specific occupancy targets and the cognitive ρ_max framing.

See Also

• Service Level — the dual quantity
• Erlang-C — the staffing formula
• Erlang-A — abandonment-aware variant
• Average Speed of Answer (ASA) — wait-side metric coupled to occupancy
• Average Handle Time — service-time input
• Shrinkage — distinguishes occupancy from utilization
• Cognitive Portfolio Model (N*) — cognitive ρ_max for AI-supervising roles
• Three-Pool Architecture — pool-specific occupancy targets
• Capacity Planning Methods — where occupancy targets are set
• Workforce Cost Modeling — occupancy in cost-per-FTE economics
• Power of One — single-agent SL/occupancy sensitivity
• Variance Harvesting — Level 3 use of occupancy variance
• Adherence and Conformance — adherence drives reported occupancy

Interactive tools

• Power One — powerone.wfmlabs.com. Visualizes the SL / occupancy / cost trade-off across a full staffing range, including burnout and boredom risk bands.
• WFM Variance Analysis — occupancy-variance-analysis.wfmlabs.com. Decomposes variance using planned (not actual) occupancy — the correct method for separating volume, AHT, and staffing variance.
• Erlang Suite — erlangcalculator.wfmlabs.com. Computes occupancy as a function of Erlang inputs.