Using Machine Learning to Classify IT Sales Deals by Win-Loss Reasons
Open AccessImagine you are an IT business leader who has just received a monthly performance report that indicates your company has lost 10% market share to one of your competitors. The first question in your mind is, why? Most companies will look to their Competitive Intelligence teams for the answer to the, why, question. The Competitive Intelligence team will tap many sources to help answer this question for their leadership team. One way they will attempt to provide an answer to why their company is winning or losing business to a competitor is through Competitive Win-Loss Analysis. Competitive Win-Loss Analysis is viewed as a powerful tool for business decision-makers. Generating a Win-Loss Analysis requires a Competitive Expert to sanitize, organize, and classify large amounts of a company’s sales data. Given that the pre-work to generate a Win-Loss Analysis is resource intensive, requiring many manual “man-hours,” some companies do not use them. Or they generate the win-loss analysis based on a small subset of sales deals which dilutes the impact of the analysis. For this reason, a Machine Learning Model that classifies an IT company’s competitive sales deals based on win-loss reasons is necessary to reduce the man-hours required to produce meaningful and actionable win-loss analyses. Through research and experimental trials, four supervised algorithms, Random Forest, KNN, MNB, and SVM, were used to build a Win-Loss Classifier. One deep learning algorithm, BERT, was also used to build the Classifier. Each Classifier was tested with two different validation techniques, Train-Test Split and K-Fold Cross Validation. The data that was used to train and validate the performance of the Win-Loss Classifiers was sourced from a Fortune top 10 Global IT company. The Win-Loss Classifiers built with the supervised ML algorithms performed well, with accuracy results in the mid-80% range. The desired accuracy threshold was 90%. Ultimately, a BERT deep learning Win-Loss Classifier that was trained with balanced data and validated using K-Fold Cross validation yielded a classification accuracy of 91%.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.