Human Resource Allocation in a Call Centre

Abstract

Last year G. Lecorroller proposed a new approach to the problem of staff allocation in a telephone call centre, based on the work of Prof D. Lowe and Dr I. Nabney, and using a multi-skill workforce. He developed a two stage process, in order to compute the number of agents required to achieve a particular service level. In the first stage, thanks to a stochastic model of the queues, a minimum number of operators was predicted, in the next stage those were allocated in the different pools using a minimisation algorithm. The results were quite encouraging, 25% fewer agents predicted with this new approach compared to the traditional and current one. The main purpose of this thesis is to improve this new model. At first we tested it whilst considering an unlimited pool of people. A tricky problem to solve was the increase of the predicted number of agents with an increasing number of pools given a call volume. The empirical method we proposed to deal with this issue is detailed. The second part of our project explored the effect of different service time distributions. A new distribution based on mixture models has been developed which fits much better the service time distribution from any given call centre. In terms of agents cost less than 2% more were required compared to the new method. Next the effects of abandoned calls have been taken into account by revisiting some approximation formulae. However, the predicted increase of agents, necessary to achieve the same service level as before, remains less than 2% compared to the new model which added to the previous improvement correspond still to 20% fewer agents than with the traditional forecast. Our model is also more precise than the Erlang C model used traditionally. Ultimately we are able to validate the new model with our improvement using both simulation program and new data. All the results are presented in careful details.