WFM guideForecasting

Contact centre forecast error impact

Q: What is the difference between over-forecasting and under-forecasting in a contact centre?

Over-forecasting (predicting more contacts than arrive) produces labour waste: agents are staffed and available but contacts do not arrive to fill their time. Occupancy is lower than planned; agents are idle; the labour cost for the interval is incurred without the corresponding revenue-generating or customer-serving activity. Over-forecasting is expensive but recoverable — the operation meets its SL target easily because it is overstaffed, and the cost is visible in the monthly labour report. Under-forecasting (predicting fewer contacts than arrive) produces service level failure: the staffed agents are insufficient to handle the actual volume at the target SL. Because of the non-linear relationship between staffing and SL in queuing systems, a 10% staffing shortfall does not produce a 10% SL reduction. At moderate occupancy levels, a 10% staffing shortfall can produce a 30–50% SL reduction in the affected interval because the queue builds faster than it clears. Under-forecasting is more damaging than the error magnitude suggests, and the damage is immediate and customer-visible. From a business perspective: over-forecasting is a cost problem; under-forecasting is a customer experience problem. Both matter, but they require different responses.

Forecast error is not symmetric. A 10% over-forecast produces quantifiable labour waste. A 10% under-forecast can produce 30–50% SL failure in the affected interval — because of the non-linear relationship between staffed agents and service level in queuing systems.

Why under-forecasting is worse than its error percentage suggests

In a voice queue running at 75% occupancy (a typical operational level), removing 1 agent from a 10-agent team is a 10% staffing reduction. But the SL impact is not 10% — because the queue that was clearing steadily now has insufficient capacity to keep up with arrivals. Queue length increases rapidly. Wait times extend. Contacts that were going to be answered within the SL window are now pushed beyond it.

At 85% occupancy, the same 1-agent reduction from 10 to 9 agents may cause SL to fall from 80% to 55% — a 25-point drop from a 10% staffing reduction. At 90%+ occupancy (a high-traffic peak interval), removing 1 agent can collapse SL entirely because the queue builds faster than it clears.

This is the non-linear queuing relationship. It is why WFM functions must be held to accuracy targets, not just directional forecasting. "We were approximately right" is not sufficient when the queuing math converts approximate errors into severe customer experience failures.

Forecast error impact by error band

0–5% forecast error (WAPE)

Over-forecast consequence (volume lower than predicted)

Minimal. Occupancy 2–4pp below plan. Labour cost marginally above plan per contact. Within normal operational variation — no action required.

Under-forecast consequence (volume higher than predicted)

Minimal. Erlang C efficiency buffers absorb small shortfalls at typical occupancy levels. SL may decline 2–5pp in peak intervals but recovers naturally as volume normalises. Intraday team can usually manage with flex tools.

Management response

No specific action required. Monitor trend. Within-tolerance variation at this level is expected from any forecast model and does not indicate a systematic bias.

5–10% forecast error (WAPE)

Over-forecast consequence (volume lower than predicted)

Measurable. Occupancy 4–8pp below plan. Cost per contact increases noticeably. Operations will notice agents are under-utilised. Finance will query the labour efficiency variance.

Under-forecast consequence (volume higher than predicted)

Significant. SL may miss target by 10–20pp in peak intervals. Intraday team must actively manage — adjusting breaks, releasing overtime, activating overflow. Not recoverable with standard flex tools if the error persists across multiple intervals.

Management response

Review the forecast model for systematic bias. Check whether the error is consistent (indicating a model assumption is wrong) or random (indicating high volume variability). At this level, investigate cause before re-calibrating the model.

10–20% forecast error (WAPE)

Over-forecast consequence (volume lower than predicted)

Significant. Occupancy materially below plan. Labour cost variance is large enough to appear in the P&L. Risk of agents disengaging due to persistent low occupancy. May signal over-caution in forecast assumptions or a volume source that has reduced.

Under-forecast consequence (volume higher than predicted)

Severe. SL deteriorates sharply in under-staffed intervals — due to the non-linear queuing relationship, a 15% staffing shortfall may produce 40–60% of contacts failing to meet the SL threshold. Customer satisfaction impact is significant. Escalation to Operations Director level is appropriate if this error band is sustained across multiple weeks.

Management response

Forecast model review is mandatory. Convene a formal forecasting review meeting with WFM, Operations, and commercial/product teams. Identify whether the error is driven by volume (the forecast model) or AHT (the handle time assumption). Produce a corrective action plan with a target re-calibration date.

Greater than 20% forecast error (WAPE)

Over-forecast consequence (volume lower than predicted)

Severe. Labour budget is materially wasted. Finance intervention is likely. Either the forecast model has a fundamental flaw, or an unplanned volume reduction has occurred (e.g. significant self-service deflection, product discontinuation, demand seasonality not captured in the model).

Under-forecast consequence (volume higher than predicted)

Catastrophic. SL targets will not be met. Customer experience will deteriorate severely. Queue wait times will be multiple times the SL threshold in peak intervals. Reputational risk for the operation. Emergency resource options (agency, mandatory overtime, inter-site routing) must be deployed immediately while the root cause is identified.

Management response

Escalate immediately to senior leadership. The forecast has broken down — either the model is wrong, an external event has occurred that changed contact volume beyond the model's tolerance, or forecast data is unreliable. Treat as a forecasting incident with a root-cause analysis, not a routine monthly review item.

Forecast error questions

What is the difference between over-forecasting and under-forecasting in a contact centre?

Over-forecasting (more contacts predicted than arrive) produces labour waste: agents staffed and available with no contacts to handle. Occupancy drops below plan; cost per contact rises; the operation meets SL easily but expensively. Under-forecasting (fewer contacts predicted than arrive) produces SL failure: insufficient agents to handle actual volume. Because of the non-linear queuing relationship, a 10% staffing shortfall typically produces 30–50% SL failure in peak intervals — not 10% SL failure. Under-forecasting is a customer experience problem; over-forecasting is a cost problem. Both are failures, but under-forecasting has the more severe and immediate consequence.