Contact centre forecasting: a beginner’s guide

Beginner level · ~5 minute read

Introduction

Forecasting is the foundation of effective contact centre workforce planning. Without an accurate view of how much demand is coming through the door, every downstream activity — scheduling, recruitment, budgeting, and service-level performance — becomes a guess. For people new to the discipline, forecasting can feel intimidating, with its mixture of statistics, business judgement, and operational knowledge. The good news is that the core principles are accessible, and applying a few proven habits will lift the quality of your forecast quickly. This article introduces the building blocks and the best practices that help beginners produce reliable, defensible forecasts.

What forecasting actually is

In a contact centre, forecasting is the process of predicting how many contacts (calls, chats, emails, or other interactions) will arrive in a future period, and how long each will take to handle. A complete forecast typically covers three things: contact volume, average handle time (AHT), and shrinkage — the percentage of paid time agents spend off the phones for breaks, training, meetings, or absence. Together these inputs feed an Erlang or simulation calculation that tells planners how many agents are required in each interval to meet a service-level target.

Why it matters

A small forecasting error can have a disproportionate impact on customer experience and cost. Under-forecast by ten percent and queues balloon, abandonment rises, and agents burn out. Over-forecast and you carry expensive idle time, eroding the operating budget. Because workforce costs typically represent 60 to 70 percent of contact centre operating expense, the forecast is one of the highest-leverage numbers in the business. Investing time to do it well pays back many times over.

Use clean, granular historical data

Good forecasts start with good data. Pull at least 12 to 24 months of interval-level history (15- or 30-minute buckets are standard) so the model can see seasonality, day-of-week patterns, and intraday curves. Cleanse the data before modelling: remove outliers caused by outages, system failures, marketing spikes, or one-off events, and document why each adjustment was made. Treat each contact channel separately — voice, chat, email, and back-office work behave very differently and should never be blended into a single forecast. The same discipline applies to AHT: track it by channel, by skill, and ideally by call type, because aggregated averages hide the variation that drives staffing accuracy.

Choose the right method for the horizon

Different forecasting horizons need different techniques. For long-range planning (six to eighteen months ahead), top-down methods that combine historical trend, business growth assumptions, and known marketing or product events tend to work best. For the operational horizon (one to twelve weeks ahead), time-series methods such as Holt-Winters exponential smoothing or ARIMA do an excellent job of capturing trend and seasonality. For the very short term (today and tomorrow), real-time adjustments based on actual arrivals and same-day intelligence outperform any model. Start simple — a well-tuned exponential smoothing model often beats a complex machine-learning approach, especially when data is limited.

Bring the business into the room

Forecasting is not a purely statistical exercise. Marketing campaigns, product launches, billing cycles, weather, and public holidays all shift demand in ways the history alone cannot anticipate. Build a regular forecasting forum with marketing, operations, IT, and finance to capture upcoming events and translate them into adjustments. Document every override on top of the statistical baseline so you can review whether the assumption proved correct. Over time this creates a library of event impacts that improves future accuracy.

Measure accuracy and learn from it

You cannot improve what you do not measure. The two standard metrics are Mean Absolute Percentage Error (MAPE) for volume and Weighted Absolute Percentage Error (WAPE) when intervals vary widely in size. Track accuracy at multiple levels — daily, weekly, interval — and against multiple horizons (one day out, one week out, one month out). A mature contact centre aims for 5 to 10 percent MAPE at the daily level for stable queues. Hold a brief weekly review to look at where the forecast missed and why, then feed those lessons back into the model and the assumption library.

Common pitfalls to avoid

Beginners typically fall into a small set of traps. They forecast volume without forecasting AHT, and then wonder why staffing is wrong even when arrivals are right. They fail to separate channels, blending email backlog work into voice arrival patterns. They ignore shrinkage, producing requirements that look fine on paper but collapse the moment real-world absence hits. And they treat the forecast as a finished product rather than a living estimate that should be refreshed as new information arrives.

A simple best-practice checklist

Begin every forecast with cleansed, granular history. Forecast volume, AHT, and shrinkage separately for each channel. Match the method to the horizon. Layer business intelligence on top of the statistical baseline, and document every assumption. Measure MAPE weekly and act on the gaps. Above all, treat forecasting as a team sport: the best forecasts come from combining mathematical rigour with operational judgement.

Ready to put the numbers into practice? Try the Erlang C staffing calculator or the simple volume forecaster.