Call center forecasting

Call center forecasting is the practice of estimating how many calls, chats, and emails will land, and how many agents you need. Every workforce plan rests on it, since the forecast sets hiring, shifts, service level, and labor cost across the floor.

Workforce managers own the forecast. They build long-range projections for recruitment, then refresh short-range views each week as schedules change. When the forecast drifts, every number downstream drifts with it.

Good forecasts blend historical data, seasonality, and known business events like product launches, promotions, or billing cycles. They get updated continuously, not filed once and forgotten.

The workforce management team treats the forecast as a living document, not an annual exercise. Every schedule, every hiring requisition, and every service level target traces back to it.

Key takeaways

• Forecasting predicts contact volume and the agent headcount needed to handle it.
• Historical data, seasonality, and known business events are the three core inputs.
• Erlang C, dating to 1917, still converts forecast volume into agent counts.
• Weekly and intraday forecasts matter more than annual ones because volume swings fast.
• Small accuracy gains cut labor cost and lift customer satisfaction at the same time.

How it works

Call center forecasting works in four moves: pick the horizon, model expected volume from history plus known events, translate that volume into agent-hours with Erlang C, then inflate for shrinkage before the schedule goes out.

Historical data usually stretches back 12 to 24 months so seasonal cycles show up. Managers pair that base with same-day drivers: weather, marketing sends, outages, or a new product going live.

Interval buckets of 15 or 30 minutes beat daily totals — contact arrival is uneven inside a shift.

Injixo’s contact-center forecasting fundamentals guide walks new teams through the base mechanics, one interval at a time.

Erlang C, developed by Danish mathematician Agner Krarup Erlang in 1917, is still the workhorse. It converts volume, handle time, and a service level target into a required agent count.

The Erlang C worked example on Call Centre Helper puts it in numbers: 100 calls per half-hour at three-minute handle time needs 20 agents for an 80% service level after 30% shrinkage.

Small changes matter. A 5% forecast error on a 500-seat floor can mean 25 misallocated agents per interval — expensive when idle, painful when short.

Agent-hours then convert into full-time equivalent (FTE) counts, the unit recruiters and finance both understand. That handoff is where a forecasting error turns into a hiring error.

Every forecast also needs an owner for its assumptions. Write down the deflection rate, the shrinkage factor, and the handle time you used, so next quarter’s review can tell a bad model from a bad assumption.

Examples

Forecasting looks different in every sector. Retail banks, online retailers, and telecom operators run the same Erlang math, but their arrival curves and staffing responses diverge sharply, which changes how far ahead each one can plan.

An online retailer sizing its 2025 holiday peak forecasts off the prior two years of November and December data, layered with current marketing spend and a projected order lift. Peak-week staffing often runs two to three times baseline.

Overflow desks at outsourced sites get pre-booked months out, because a 2x volume week cannot be recruited for in October. The call center that waits until November staffs the peak on overtime instead.

Telecom operators face a different curve. When an outage hits, contact volume can spike 400–600% within an hour, so real-time tools re-run the model every few minutes and page in flex agents.

Average response time is the first metric to crack in that scenario, which is why intraday teams watch it minute by minute rather than reading it in a weekly report.

Retail banks work off monthly and quarterly billing cycles. Statement drop dates and card renewal windows drive predictable surges that an annual forecast already captures, so the workforce team just tightens intervals as the date nears.

Airlines sit at the volatile end — one weather event can reroute thousands of passengers at once, so carriers hold a standing reserve pool and treat the same-day forecast as the real plan.

Digital deflection reshapes the voice forecast too. In January 2017, Harvard Business Review’s Kick-Ass Customer Service reported that 81% of customers try self-service before calling a live agent.

So a chat or help-centre improvement lands as a voice volume drop later in the year — which is why forecasters who ignore the deflection curve overstaff the phone queue and understaff the chat desk.

Related terms

Call center forecasting sits inside a wider workforce discipline. The entries below cover the numbers that feed a forecast and the metrics that judge it afterwards, but they stop short of scheduling software and quality assurance.

• Workforce Management: the umbrella process that covers forecasting, scheduling, and adherence tracking.
• Full-Time Equivalent (FTE): the staffing unit forecasters convert contact volume into.
• Key Performance Indicator (KPI): service level, occupancy, and shrinkage are all forecast-driven measures.
• Average Response Time: a lagging indicator that shows whether the forecast matched reality.
• Average Handle Time: a core forecast input where a ten-second shift changes required headcount by roughly 5%.
• Call Center: the operating unit the forecast serves.

FAQ

What data do you need to start forecasting call volume?

At minimum, you need 12 months of interval-level contact history, a calendar of known business events, and an accurate average handle time by contact type. Cleaner inputs beat fancier models nearly every time. The event calendar is usually the piece nobody owns.

How accurate should a call center forecast be?

Most centers target 90–95% accuracy at the daily level and 85–90% at the 30-minute interval level. Anything below 80% forces expensive same-day corrections.

What is Erlang C and why does it matter?

Erlang C is a queuing formula that turns forecast volume, handle time, and a service level target into the number of agents required. Dating to 1917, it remains the default because it accounts for variability in arrival patterns.

How often should forecasts be refreshed?

Review short-term forecasts weekly and adjust intraday when volume drifts more than 10% from plan. Long-term staffing forecasts get a full rebuild each quarter.

Can AI replace traditional forecasting methods?

Machine-learning models often improve accuracy on volatile channels like chat, but Erlang-based staffing math still sits underneath. Most vendors blend both, using AI for volume prediction and Erlang for agent conversion.

What is shrinkage and how does it affect the forecast?

Shrinkage is the share of paid time agents spend off contacts on breaks, training, and absence, and because it typically runs 25% to 35%, forecasters must inflate the raw agent count by the shrinkage factor or the schedule ships short.

Ready to size a forecasting team without hiring in-house? Browse Outsource Accelerator’s outsourcing hubs to shortlist business process outsourcing (BPO) providers with workforce-management depth.