Health identification and outcome prediction for outsourcing services based on textual comments

Abstract

Outsourcing service opportunities go through a complex and laborious process beginning with RFP processing and continuing through solutioning, proposal development, costing, pricing, contract definition and closing. A key issue for service providers is to understand the health status of a given service opportunity engagement based on available information at any point in time during the engagement. This capability enables them to bring excelling engagements to a faster close, and identify and help troubled engagements. As a related problem, service providers are interested in understanding the win chances of engagements in order to prioritize their pipelines for planning and resource management purposes. In this paper, we propose a new sentiment-based approach for identifying and monitoring the health of opportunities by analyzing weekly comments made by sales teams for each engagement. We then present a novel method for feature extraction and selection from the engagements' comments. This approach combines text-based and sentiment-based features and identifies highly predictive features that are then used to predict the outcome of engagements. We report on experiments performed on an industrial dataset, which shows our prediction approach achieves a high win prediction accuracy. We describe a method for providing a supplementary confidence level and linguistic features that facilitates the interpretation of the predicted outcome.