Enabling deeper customer-centricity - creating a search engine for project proposal documents at a tech consultancy

Abstract

Information retrieval systems are a crucial business resource. Salespeople at technology consulting companies want to improve their day to day customer-centric processes by researching their organization's collective knowledge. The purpose of this research was to build a search engine application for project proposal documents to help salespeople become more customer-centric.

Proposal Driller, a search engine application for project proposal documents having three app versions was built using modern web technologies after studying the peculiar features of the document and identifying the user's need for information retrieval. Information extraction using Natural Language Processing was done to enrich the documents. The application allowed users to search, filter, sort, view other similar proposals, and browse through the company's proposals in a way that was never available to them before.

An evaluation with fourteen participants showed that the most useful features of the application were filtering the search results by customer name, sorting by document creation date, and viewing the extracted keyphrases and similar documents. Furthermore, the comparisons between the search results of the three app versions showed that searching on a few important pages of the proposal seems to be equally good as searching on the entire proposal document. Also, for a few search queries, it was seen that the app version v3 that searched on the vector embedding of the document made with Google Universal Sentence Encoder was useful when the search terms did not occur in the documents but relevant documents were semantically related to the search terms.

In conclusion, the success of the final application presented in this thesis shows that a search engine application like this is a useful tool for a tech consultancy to enable deeper customer-centricity. However, some limitations of the application and difficulties of relevance evaluation are provided along with recommendations for future work as there are some interesting areas that can be explored next.