We establish the context of software maintenance outsourcing of large commercial systems and discuss the problem of estimating the maintenance effort in absence of key parameters such as development effort and maintenance history. To arrive at the drivers that could have a significant effect on the maintenance effort, an experiment was conducted at Syntel, a NASDAQ listed application management and e-business solutions company. We present the list of possible drivers, the profile of the people who participated in the experiment, the details of the 36 experiments and the statistical analysis of the experiment based on Taguchi's methodology. For a system whose size parameter in terms of lines of code is known, we conclude that the five most significant parameters that effect the efforts of maintenance are a)the multi time zone support, b)average number of lines per program, c)percentage of online programs in the total system, d)the nature of service level agreements and e)the complexity of the file system being used.