A methodology for forecasting knowledge work projects

Abstract

Forecasting project duration for knowledge workers is particularly difficult due to the complexity and variability of their work. This study examines prediction of software development completion times which has traditionally involved either software engineering techniques or purely judgmental forecasts by lead analysts or project managers. In practice, neither approach has achieved much success in forecasting software project duration. This paper proposes a neural network model to modeling software project overruns. Actual software project management data, obtained for a regional grocery chain, were used to develop and test the neural network model as well as a traditional regression model for forecasting project overruns. Comparisons between the forecasts and the actual project overruns revealed that the neural network model outperformed the regression model in terms of forecast accuracy, degree of forecast bias and model fit. The results suggest that neural network modelling can be used to integrate managerial judgments and actual operating data to accurately forecast software project completion times. Scope and purpose This study examines the difficult managerial issue of estimating knowledge work project completion time. Many knowledge worker projects are late and over-budget due to the complexity of knowledge work contexts and the pressures on the project managers to underestimate project duration.This paper compares the project duration estimates of project managers to quantitative prediction techniques that use project history data to predict project overruns. The two techniques used are neural network modeling and regression.