Forecasting call arrivals at a call centre

22 June 2014

Call and service centres are an essential resource of today’s businesses, from sales and financial services to emergency response. 

More than 70% of today’s customer and business interactions are handled through call centres, making then crucial to the operation of firms. With 60-70% of operating expenses due to human resource costs, one of the main challenges in managing a call centre is determining adequate staffing levels to meet future demand, and to ensure a desired level of service quality and efficiency. This requires the accurate forecasting of call centre inbound calls in order to schedule agents efficiently within a week and during the day (operational planning), or to determine the quantity and timing of hiring and training (tactical planning).

The Lancaster Centre for Forecasting has been involved in research and industry-based projects in call centre forecasting towards balancing service quality, costs and efficiency of call centres. Overestimating call centre demand can lead to planning higher agent workloads than the actual values, resulting in overstaffing and leading to unnecessary costs. In contrast, underestimating the number of incoming calls may result in needlessly long waiting times, high abandonment rates, resulting in poor service quality, and leading to lost revenue and possibly lost customers.

The issues call centre forecasters face are:

  • Multiple seasonality
  • Sales and promotional activities
  • Special events such as sports and emergencies
  • Lost calls and balking

In a recent research study, Dr Devon K Barrow (visiting fellow with a Lancaster PhD) has been studying how to improve the accuracy of call arrivals forecasting using a hybrid approach of the very simple seasonal average method, supplemented by artificial neural networks (ANNs). Such an approach is motivated by observing that:

  1. The seasonal average method performs well at long-term forecasting, but there remain systematic errors.
  2. Research suggests that a combination of forecasts on average improves forecasting accuracy over using a single ‘best’ model

Results in company case studies (in media, finance and utilities) show that in nearly all cases, except for very short seasonal averages, the proposed method leads to substantial improvements over the seasonal average and benchmark methods in forecasting call arrivals. This is observed across various sampling frequencies of call arrival data, from five-minute to hourly observations.

The LCF is always keen to further its collaboration with industry. If you are interested for more information on this work please contact