Effort estimation in large-scale software development: An industrial case study

Context: Software projects frequently incur schedule and budget overruns. Planning and estimation are particularly challenging in large and globally distributed agile projects. While software engineering researchers have been investigating effort estimation for many years to help practitioners to improve their estimation processes, there is little empirical research about effort estimation in large-scale distributed projects involving agile teams. Objective: The objective of this paper is three-fold: i) To identify how effort estimation is carried out in large-scale distributed agile projects; ii) to analyze the accuracy of the effort estimation processes in large-scale distributed agile projects; and iii) to identify and investigate the factors that impact the accuracy of effort estimates in large-scale distributed agile projects. Method: We performed an exploratory longitudinal case study. The data collection was operationalized through archival research and semi-structured interviews. Results: The main findings of the studied case are: 1) A two-stage estimation process, with re-estimation at the analysis stage, improves the accuracy of the effort estimates; 2) underestimation is the dominant trend; 3) less mature teams incur larger effort overruns; 4) requirements with larger size/scope incur larger effort overruns; 5) requirements developed in multi-site settings incur larger effort overruns as compared to requirements developed in a co-located setting; 6) requirements priorities impact the accuracy of the effort estimates. Conclusion: A two-stage effort estimation process can improve effort estimation accuracy and seems to address some of the challenges in large-scale agile software development. To improve effort estimates one needs to consider team maturity, distribution as well as requirements size and priorities.