Automate RFP Response Generation Process Using FastText Word Embeddings and Soft Cosine Measure

Abstract

Companies frequently receive RFP (Request for Proposal) from interested customers requiring technical expertise, specialized capability regarding to their products or services. Appropriately responding to these RFPs is heavily influential in buyer decision-making. Currently most companies answer RFPs manually, and they (including some major RFP solution providers) mainly use key word(s) matching algorithm to search for similar questions in the knowledge base and choose the one the working analyst thinks most relevant. However, the biggest challenge is the same question can be rephrased using different wordings, so key word matching may lead to irrelevant result(s). The current process is very time-consuming, inefficient, ineffective and sometimes can create inconsistencies. In this paper, we propose using a method which combines state-of-the-art Word Embedding solution in Natural Language Processing (NLP) and AI with the promising new idea of Soft Cosine Measure (SCM) in finding the matching responses. This is by far the first effort to introduce the state-of-the-art technologies in RFP process to improve accuracy, efficiency and effectiveness. From our study of applying this method, we observed high response relevancy rates for non-standardized questionnaires which are out of the knowledge base.