Operations Research

Service Demand Modelling

In call centres and contact centres, customer service is typically measured in terms of a service standard such as no more than 5% of callers will wait in a queue for more than 20 seconds. In order to achieve a given service standard, the call centre must answer the following two questions properly:

1. Forecasting call volumes, and
2. Generating staff requirements.

CSIRO has developed two modules, the Incident Forecaster and Demand Modeller, as part of the Staff Rostering Toolkit.

Incident Forecaster

The Incident Forecaster predicts future incident demands or call volumes based on historical data and other relevant information. The accuracy of call volume forecasting has profound impacts on the business' overall planning and operational strategies.

Various time series forecasting methods have been implemented in the Incident Forecaster. We have also developed a multiple stage forecasting approach that is especially appropriate for the cases where (a) The data type is of hourly or even finer time intervals, (b) Daily starting times vary, and (c) Forecasts are required for several months ahead.

The user graphical interface allows the user to enter relevant data by filling boxes and ticking selections. The forecasts as well as forecasting errors of the statistical model can be viewed graphically in several time scales such as in base time intervals, weekly and monthly, and can be adjusted manually. Some measures of forecasting accuracy are also available to the user.

Demand Modeller

The Demand Modeller converts future incident or call volume forecasts obtained from the Incident Forecaster into staff requirements in different skill classes in each future time period (for example hourly) over a planning horizon in order to provide a satisfactory response to calls that are likely to occur during those periods.

The output of demand modelling will become the input to the scheduling and/or rostering modules. The Demand Modeller can efficiently and accurately convert incident forecasts into staff requirements using various service standards such as the grade of service and average waiting times amongst others.

Several queuing models have been developed and implemented in the Demand Modeller. They are extensions of the famous Erlang-C formula. The new models allows non-stationary call arrival rates, can take service linkage between time periods into account, can handle multi-class customers with multi-skill servers, and can also tackle abandonment and retrials. It is also possible to do sensitivity analysis of impacts of changes of staff requirements on service levels.

As in the Incident Forecaster, the user graphical interface allows the user to enter relevant data by filling various fields. The demand curves of staff requirements along the time can be viewed graphically, and can be adjusted manually. Various service levels corresponding to the demand curves are available to the user.

Other Application Areas

The Incident Forecaster developed for the Staff Rostering Toolkit can either be applicable or be extended for other applications such as hotel room bookings, transportation demand forecasting, airline seat bookings, hospital patients arrival patterns, emergency services like police and ambulances among others. We are aware that specialized forecasting methods are available for hotel room bookings and airline seat bookings that are typical applications of revenue management.

The Demand Modeller, initially developed for the Staff Rostering Toolkit, can be used to generate staff requirements for nurses in health care systems, police and ambulance officers in emergency services, and many other service industries and business where dynamic service volumes exist.

For further information contact Andreas Ernst.

Last updated February 10, 2009 04:15 PM