Offered Load vs Carried Load

Offered load is the total demand presented to a contact center, measured in Erlangs. Carried load is the demand actually served. Lost load is the difference: offered minus carried. This distinction is one of the most consequential in workforce management because nearly every WFM system measures carried load by default, but staffing calculations require offered load. Confusing the two leads to systematic understaffing.

Overview

When a contact center receives 2,000 calls in an hour but 150 abandon before being answered and 10 receive busy signals, the operation carried 1,840 calls. The ACD reports 1,840 answered calls. If a planner uses 1,840 as the forecast basis for next week's staffing, they are planning for the load that was served — not the load that was offered. The result: fewer agents than needed, which produces more abandonment, which further suppresses the measured carried load, which leads to even less staffing. This is the demand suppression spiral — a feedback loop where understaffing makes itself invisible in the data.

Offered load is the true demand signal. Recovering it from carried load is a critical WFM capability.

Mathematical Foundation

Definitions

Offered load (A) in Erlangs:

$A=\frac{\lambda \times \text{AHT}}{T}$

where λ is the number of arrivals (not answered calls) in a time interval of length T, and AHT is the average handle time per served call. When T is expressed so that λ/T yields a rate per second and AHT is in seconds, the formula simplifies to:

$A={\lambda }_{\text{rate}}\times \text{AHT}$

This is Little's Law applied to the service subsystem.

Carried load (A_c):

$A_c=\frac{{\lambda }_{\text{answered}}\times \text{AHT}}{T}$

Only calls actually answered contribute to carried load. The ACD measures this directly: it is the sum of agent-occupied time in an interval.

Lost load (A_L):

$A_L=A-A_c$

Lost load has two components:

1. Abandoned load — callers who entered the queue, waited, and hung up before being answered.
2. Blocked load — callers who received a busy signal or capacity rejection (see Erlang B).

The Erlang as a Unit

One Erlang equals one server (agent, trunk, port) fully occupied for the entire measurement interval. If a contact center has 100 Erlangs of offered load in a half-hour interval, it means the demand would keep 100 agents continuously busy if no one waited and no one was lost.

Dimensionally, Erlangs = (calls per unit time) × (time per call) = dimensionless. It is a measure of traffic intensity.

Load Type	What It Measures	Source
Offered	Total demand (served + lost)	Must be estimated
Carried	Demand actually served	ACD reports directly
Lost	Demand not served (abandoned + blocked)	Estimated from abandonment data

Estimating Offered Load from Carried Load

The standard method:

$A=\frac{\text{Answered Calls}+\text{Abandoned Calls}+\text{Blocked Calls}}{\text{Interval Length}}\times \text{AHT}$

But this overcounts slightly because abandoned callers would have consumed a full AHT had they been answered — the estimate assumes they would have been served at the average handle time. This is the standard WFM convention and is approximately correct for planning.

A more precise recovery:

$A=\frac{(\text{Answered}\times \text{AHT})+(\text{Abandoned}\times {\text{Estimated AHT}}_{\text{ab}})}{T}$

where AHT_ab is the estimated handle time the abandoned calls would have had. In practice, AHT_ab is assumed equal to AHT for the skill group — there is usually no basis for a different number.

For blocked calls (no queue entry), the estimate requires trunk-level or platform-level data that many ACDs do not surface. In modern SIP environments, blocked calls are rare enough to ignore for staffing purposes but remain relevant for trunk sizing.

WFM Applications

Worked Example: The Understaffing Spiral

Scenario: A center planned for 1,800 calls/hour based on last week's carried load. AHT = 4 minutes.

Planned offered load: 1,800 × 4/60 = 120 Erlangs. Erlang C at 80/20 SL yields 140 agents needed.

What actually happens: True demand is 2,000 calls/hour. Offered load = 2,000 × 4/60 = 133.3 Erlangs. With only 140 agents, the system is severely overloaded for 133.3 Erlangs:

• ASA rises to 45 seconds
• 200 callers abandon (10% abandonment rate)
• ACD reports 1,800 answered calls — exactly what was planned

The planner sees the carried load match the plan and concludes staffing was correct. Next week, the plan again targets 1,800 calls. The spiral continues.

The fix: Forecast offered calls, not answered calls. Add abandoned and blocked to the base: 1,800 + 200 = 2,000. Staff for 2,000.

Interval-Level Distortion

The offered-vs-carried gap is worst in the intervals that matter most — peak intervals with the highest abandonment. A planner who averages carried load across the day misses the peak-interval suppression, leading to under-planned peaks and over-planned troughs.

Example:

Interval	True Offered	Abandoned	Carried	% Lost
09:00–09:30	120 calls	5	115	4%
10:00–10:30	250 calls	40	210	16%
14:00–14:30	180 calls	10	170	6%

The 10:00 interval has 16% demand suppression. Planning from carried load understaffs this interval by 40 calls worth of load — approximately 5–6 agents.

Short Abandons

A "short abandon" is a caller who hangs up within a few seconds of entering the queue — often before even hearing a ring or message. Many operations exclude short abandons (< 5 seconds, < 10 seconds) from abandonment counts because they may represent misdials, IVR dropoffs, or callback triggers rather than genuine demand.

Whether to include short abandons in offered load depends on their cause:

• Misdials and accidents — exclude. Not real demand.
• Callers who reached IVR self-service and dropped — exclude from voice queue but include in IVR demand.
• Impatient callers — include. They represent genuine demand that was not met.

The threshold choice (5 sec, 10 sec, 20 sec) materially affects the offered load estimate. Document the convention and apply it consistently.

Common Misconceptions

1. "ACD call volume = demand."

ACD answered volume is carried load. Demand includes abandoned and blocked. In high-abandonment environments, the gap is 10–20%.

2. "Abandonment is small enough to ignore."

Even 3% abandonment in peak intervals can represent 5–7% demand suppression in those intervals. Since these are the intervals that determine staffing adequacy, ignoring a few percent of abandonment propagates disproportionately into understaffing.

3. "We add abandons to our forecast, so we account for offered load."

Verify where abandons are added. If they are added to daily totals and then distributed by historical intraday pattern (based on carried load), the interval-level distortion survives. The correction must happen at the interval level.

4. "Carried load is more 'real' than offered load."

Carried load is an artifact of staffing decisions. Offered load is the customer demand signal. Planning from carried load is circular — you plan based on what you chose to serve, not on what customers wanted.

5. "Offered load in Erlangs must be less than the number of agents."

Offered load can be any positive number. When offered load exceeds agent count, the queue grows and service degrades — but the offered load is still the offered load. See Traffic Intensity and Server Utilization.

Maturity Model Position

• Level 1 — Initial. Forecasts based on ACD-reported answered calls. No adjustment for abandonment. Demand suppression unrecognized.
• Level 2 — Foundational. Abandoned calls added to forecast volumes at the daily level. Short-abandon convention defined. Practitioners understand the distinction between offered and carried.
• Level 3 — Progressive. Offered load reconstructed at the interval level. Abandonment data integrated into the forecast pipeline. Short-abandon threshold empirically validated. Demand suppression measured and reported.
• Level 4 — Advanced. Real-time offered load estimation feeds intraday reforecasting. Blocked-call data from SIP layer included where available. Offered load tracked across channels with consistent definitions. Erlang-A used to model the abandonment/staffing interaction explicitly.
• Level 5 — Pioneering. Latent demand modeled — customers who would have contacted but chose not to (e.g., saw long expected wait in IVR and used self-service instead). Offered load becomes a probabilistic range rather than a point estimate.

See Also

• Little's Law Applied to WFM
• Erlang B
• Erlang C
• Erlang-A
• Traffic Intensity and Server Utilization
• Demand calculation
• Abandonment
• Service Level
• Queueing Theory Fundamentals

References

• Gans, N., Koole, G. & Mandelbaum, A. (2003). "Telephone Call Centers: Tutorial, Review, and Research Prospects." Manufacturing & Service Operations Management 5(2), 79–141.
• Gross, D. & Harris, C.M. (2008). Fundamentals of Queueing Theory, 4th ed. Wiley. Chapter 1.
• Cleveland, B. & Mayben, J. (1997). Call Center Management on Fast Forward. ICMI. Chapter on forecasting.
• Koole, G. (2013). Call Center Optimization. MG Books. Section 3.2: Traffic Measurements.
• ITU-T Recommendation E.501 (1997). "Estimation of traffic offered in the network."