Abstract

Colas is a french subsidiary of Bouygues Construction specialized in the building and maintenance of transportation infrastructure such as roads and railways. Answering calls for public contracts of road construction had been a time consuming activity for the commercial division of the company, with unpredictable yield. We were tasked to analyze a 5 year-long record of road construction contract data to : - identify determinants of bidding success. - prototype a bidding assistant to predict success of a bid based on the characteristics of the project. We conducted a statistical analysis to model the contribution of ~20 different bid features (e.g., main contract price, consortium size, geographic location...) to success of a bid. Based on this exploratory analysis, we trained a logistic regression model to predict "no go" bids with 80% recall and < 1% of False Positive Rate. The deliverables consisted of source code for training the model and display historical data on an interactive map.