Optimizing the Pilot Testing of ERP Systems

In the creation of ERP systems, it is important to plan the pilot testing. In pilot projects (also known as experimental projects), the main functions of the system are tested by a limited set of advanced users. In planning the pilot testing, the key is to determine its extent. With too little testing, fundamental functions of the system may not be checked, and the probability of detecting significant errors only at complete rollout will be high. With excessive testing, the work required in the pilot project will be considerably increased, and the required speed and flexibility will not be attained in testing the basic functions. The effectiveness of pilot testing will be no better than the effectiveness of full-scale introduction. A mathematical formulation of the determination of the optimal extent of pilot testing is developed on the basis of the generation of a portfolio of IT services and their scheduling in the creation of the ERP system for a large steel company. The solution of this problem takes the form of a set of services that must be verified and the relations between them, within the specified constraint on the resources devoted to testing, such that optimal results are obtained. The solution is obtained by network programming, based on a structurally similar network representation of the relevant criteria and constraints. A solution procedure is outlined, along with an example of its use. In this procedure, dichotomous programming is used to solve individual problems. The approximate solutions obtained may be improved by solving a double network-programming problem. The branch and bound method may be used to find the global optimum of the initial problem. In that approach, the boundaries are values of the target functions for the approximate solutions. Generalization of this problem is based on the preferences of IT-service users regarding the quality of verification of different relations between the services. Those preferences may be taken into account by weighting the corresponding relationships. That leaves the basic solution procedure unchanged.