Seminar: Peer screening with DEA: evaluating customer segments of a telecom operator
Pieter Jan Kerstens, Postdoc at Section for Environment and Natural Resources, Department of Food and Resource Economics (IFRO), will give the presentation Peer screening with DEA: evaluating customer segments of a telecom operator.
This paper addresses two separate issues:
1) Efficiency analysis can guide management in their (strategic) decisions to improve the firm’s performance. Management decisions are made keeping certain multi-dimensional objectives in mind. Furthermore, not all objectives have equal priority and realizing multiple objectives at once can require more complicated strategies. The direction of projection determines the objectives to focus on. Thus, the choice of these direction vectors is important. Surprisingly, empirical applications usually choose the direction vectors onto the efficient frontier without much ado. Recently, the literature has proposed some ways of determining these direction vectors in some optimal way. Although much can be said in favor of these approaches, an important issue from a management perspective is with regard to the intuitive interpretation of these optimal direction vectors. Lack of intuitive interpretation could preclude widespread usage. We follow a different approach and identify the most interesting objectives to focus on by learning from “key DMUs” (Liu et al. (2009) and Liu and Lu (2010)) in the data. These objectives are identified through a comparative analysis of the key DMUs’ characteristics. The rationale being that key DMUs are the successful DMUs where the others can learn from. In this way hope to strike a balance between intuitive (radial) direction vectors and optimal direction vectors.
2) After calculating efficiency scores, the next stage consists of looking into the dominating peers to determine the feasibility of the proposed benchmark. One way is to use radar plots of the inputs and outputs to compare the different dominating peers. Although radar plots can be illuminating, they have two drawbacks: (i) they quickly become overcrowded when there are many input (output) dimensions and/or when there are many observations; (ii) they do not easily allow for quickly comparing the input-output mix of various DMUs. We propose an input-output mix similarity and scale visualization tool which solves both drawbacks at the cost of the ability to compare individual input (output) dimensions. The tool is therefore intended to be used in conjunction with radar plots.
The empirical application applies both proposed methods to benchmark customer segments of a large European telecom firm using Activity-Based-Costing (ABC) data.
Jens Leth Hougaard